Value Propositions of Current CPUs Put to the Test

J. Dzhugashvili writes "Processors are typically compared by their performance alone. However, the folks at The Tech Report have put together an article that attempts to quantify the value propositions of AMD's and Intel's latest processors. The article takes 16 processors through an extensive battery of tests that range from gaming and video encoding to Folding@Home and energy efficiency, and examines the value they offer in each. The results may surprise you."

Comments

Wait for the next price drop

[cerqon]

In a few weeks the E6600 will be at $170, quite a bargain...

Re:Wait for the next price drop

[pushing-robot]

And the Q6600 will be $266, not $530. It'll be a much different playing field.

At least they lay out all the figures so you can recalculate price/performance when the big drop hits.

Re:Wait for the next price drop

[ceeam]

Saying "wait" in these comparisons is foul-play.

Re:Wait for the next price drop

[nbowman]

Yah, as of July 22 all of this changes. but then Overclocking would change the results now too. Interesting test, but the results are definitely to be taken with a grain of salt. man I can't wait for the 22nd, E6850 here I come.

Re:Wait for the next price drop

[Anonymous Brave Guy]

Coincidentally, I started serious research into building my next PC last weekend. I initially looked into the E6600, Asus P5B, DDR2 combination that has been popular in recent months, but was rapidly warned off by others because the next round of hardware is starting to hit.

As an aside for anyone else who's in the same boat and hasn't found the info yet: Intel 3-series chipset motherboards are already available, 6x50 Core 2 chips are due any time now, and DDR3 RAM is starting to filter through the retail channels. All of these like a 1333MHz front-side bus, which potentially brings quite large performance benefits. There are advantages in terms of power consumption and overclocking potential as well. Also, they bring major price cuts in older kit: one retailer here in the UK was listing a much lower price for advance orders of a top-of-the-range E6850 chip than they are listing for buying an E6700 today, which presumably means the price of the latter will plummet when the 6x50 chips arrive.

All of this makes it an odd sort of time to run a data-based comparison like TFA. There are things that matter about processors beyond raw performance and price, but those are the dominant factors. Similarly, you can always say something better is just a few months away in this business, but right now, the annual upgrades are literally starting to arrive in retail channels as I type this and in less than a month many of the data points used in the article will be obsolete. That said, some of their notes are likely to remain valid, so it's worth a read for anyone not familiar with the recent processor line-ups. Some of the results are counter-intuitive if all you do is look at the theoretical numbers and go for what looks like the best price performance: the relative lack of advantage of a Q6600 over an E6600 for many applications because the extra cores aren't used effectively in practice is one example.

Re:Wait for the next price drop

[RMingin]

DDR3 is worthless so far. It's hotter, burns more juice, same speed, and expensive. In time it'll become desirable, but that time is still quite a ways off. If you're really horny for 1333FSB (which isn't the second coming of performance Christ, BTW), you can do what I did two months ago: Core 2 Duo e6600. I got it for about 220$. eVGA nForce 680i. All the checkboxes, official 1333FSB, and as low as 150$ with a good mail-in rebate. 2GB or more of DDR2-667. I got DDR2-800 because I hadn't done all the math first, and I like headroom. Once set to 1333FSB, your e6600@2.4 magically becomes an e6850@3GHz. Say hello to 1333. While you're here, you'll want to make sure your ram is set to sync with the FSB, which puts it at 666MHz. What's that you say? You simply must have DDR3-1333? Hell no. The FSB is QDR, the ram is DDR. FSB=RAM*2 for a proper sync. Your FSB won't be able to meaningfully use more than that, you'll just be nudging the FSB saturation up from 99% to 99.5%, for a performance gain of near 0. The price cuts next month will rework this a bit, mainly I'll swap out the e6600 for a q6600 and attempt to get the same settings. 4 cores at 3GHz each ought to be possible, and I'm looking forward to it.

Re:Wait for the next price drop

[default luser]

Just to add a little extra info, and clarify a bit:

The reason you want a 2:1 ratio in FSB vs ram speed is because the FSB for Intel systems is 64 bits wide. Dual-channel ram is 2x64, or 128 bits wide. 128 * 667 = 64 * 1333.

The only reason why DDR 800 or faster memory is important is if you plan to use an integrated graphics solution, because that data doesn't have to travel over the FSB.

Re:Wait for the next price drop

[cheater512]

If you wait long enough, you'll be able to get a 256 core beast.
Saying 'Oh but if you wait' is completely pointless.

Its comparing the cpus in a certain point of time which is the only accurate way to do it.

Re:Wait for the next price drop

[Anonymous Brave Guy]

DDR3 is worthless so far. It's hotter, burns more juice, same speed, and expensive. In time it'll become desirable, but that time is still quite a ways off.

Sure it is, and I fully expect I'll be buying DDR2 when I put my new machine together. But my machines typically last around 4 years, and RAM is one of the few upgrades easy, cheap and useful enough to be worth doing without throwing out the whole core of the machine. Why would I buy a board from last month that was restricted forever to DDR2, when I can buy a board now that takes both DDR2 and DDR3? Gigabyte, for one, already have such an offering on the market.

It goes to show

[Bullfish]

That, now more than ever, the best processor for a person's needs depends on what they are going to do with their machine. The large number of choices in CPUs means that if your needs are simple, you can put together a fast machine with relatively few dollars. Ditto for video cards really. If you aren't married to the fastest cards, there are a lot of cards for around $100 give or take that will give great performance in most things, and even run a few games decently. The hype that CPU makers love to throw out there and the cost of high end parts belies that you can put together a machine cheaper now for most needs than ever before.

Re:It goes to show

[JasterBobaMereel]

Since most people do not need a cutting edge PC to do what they require the latest and greatest chips/graphics cards are of no interest unless you are actually pushing them to the limit already. And the only users who do this are,

    Playing newer games
    Video Editing
    Rendering 3D Graphics ... and other intensive applications

Of these most people will only play games (and consoles are better at this!)

Re:It goes to show

[jshriverWVU]

Of these most people will only play games (and consoles are better at this!)

Matter of opinion. I enjoy my Wii, but for most gaming you just can't beat a mouse and WASD on the keyboard :) just more esthetically pleasing.

Re:It goes to show

[SwordsmanLuke]

I agree... for first person shooters. It's hard to match the accuracy and agility you can get with the mouse/keyboard combo with a couple of thumb sticks (or one thumb stick and a Wiimote). But, I also find that many games (esp. fighting games) work much better on consoles. I think it really just depends on what the game genre was intended for.

FPSes (in their current incarnation. Battlezone[1] doesn't count.) were born on the PC[2] and have evolved to use take advantage of the mouse/keyboard control scheme. OTOH, fighting games (to stick with my example) started out in the arcades[3] and were designed with joysticks in mind. I am a long time gamer and own most of the major consoles from the Atari 2600 to the present and I also have a collection of PC games going back to the original King's Quest[4] series (And yes, I still have an old PC with a 5.25" drive to play them on). I don't think there's any real conflict vis-a-vis PC vs Console games. They're just different platforms with their own strengths and weaknesses.

1) http://en.wikipedia.org/wiki/Battlezone
2) http://en.wikipedia.org/wiki/Wolfenstein_3D
3) http://en.wikipedia.org/wiki/Warrior_(arcade_game)
4) http://en.wikipedia.org/wiki/Kings_Quest

Re:It goes to show

[JasterBobaMereel]

Well get a keyboard and mouse for the Wii then ...?

http://www.wiikeyboard.com/

Re:It goes to show

[bdulac]

True, most people don't even scratch the surface of what their PC can do. That's why they don't want to spend much when comes to a new system. Most people figure that the cheapest PC will suit their needs.....until they realize how slow the thing is! There is something to be said for getting the latest technology but mostly only real geeks are going to care whether a CPU actually is capable of the claims made by the manufacturer. :')

Re:It goes to show

[gfxguy]

Exactly... for the 3D guys where I work it's ultimately worth the extra thousands in billable time. If they merely save an hour a week, then that's 52 hours/year, over a week's worth of work, billed at about $150/hour. The fast processor is worth it.

On the other hand, I'm old school... I always thought we wanted computers to do things faster so that we'd have more time off. Oh well, that didn't work out so well.

Re:It goes to show

[WuphonsReach]

True, most people don't even scratch the surface of what their PC can do. That's why they don't want to spend much when comes to a new system. Most people figure that the cheapest PC will suit their needs.....until they realize how slow the thing is! There is something to be said for getting the latest technology but mostly only real geeks are going to care whether a CPU actually is capable of the claims made by the manufacturer. :')

The trick (if you're giving good general advice) is to steer the person towards something that is just slightly more expensive, but will result in better performance. Such as ordering a machine with 1GB of RAM instead of 512MB (for WinXP) then scaling back slightly on the CPU speed (go with the 2.2GHz instead of the 2.4GHz).

Personally, I won't recommend anything less then dual-core w/ 2GB of RAM. Dual-core (or dual-CPU) makes a huge difference in responsiveness, and 2GB of RAM is fairly inexpensive at the moment. (That's been our baseline system for the last year at work. Now that CPU prices are dropping, we're starting to buy the faster dual-core CPUs.)

Re:It goes to show

[Wolfrider]

--Yep. Dual-core Pentium-D @ 2660MHz with 2GB RAM can do 64-bit, and actually makes a pretty good Vmware build machine. (Only disadvantage is it doesn't do 64-bit guests, but I haven't really needed that yet.)

Accuracy somewhat questionable

[PoliTech]

FTA: "For instance, our Core 2 Duo E6600 and E6700 processors are actually a Core 2 Extreme X6800 processor clocked down to the appropriate speeds. Their performance should be identical to that of the real thing."

One must question the accuracy of the results due to the above verbiage.

Re:Accuracy somewhat questionable

[Splab]

Not only that, but in my experience the fraps hook can have huge impact on your framerate. On top of that the motherboards are running different chipsets, simple stuff like drivers for onboard sound cards can have a huge impact on the processing power if they are written poorly. They really aren't testing anything when they don't have a uniform rig to do so.

Re:Accuracy somewhat questionable

[CajunArson]

Maybe you need to look up the term 'speed binning' sometime. The X6800 is simply an e6600 or e6700 only at a higher clock speed. There is no differnce in cache or in architecture so I see nothing wrong with using it in this manner. There is one way in which the results might change, and that would be to differences in memory bandwidth due to messing with the FSB. However, the X6800 has an unlocked memory multiplier (Intel unlocks this on the 'X' series CPU's to make overclocking easier). By changing the multiplier and leaving the FSB alone, the reviewers are able to turn the chip into an e6600 or e6700 for any practical purpose. I mean, there will probably be some minute differences between different batches of the same e6600/e6700 production run that would have miniscule differences just as big as the difference between a properly clocked X6800 and the 'real thing'.

Re:Accuracy somewhat questionable

[MoxFulder]

Yeah, I think the underclocking approach they've used is completely valid. If two processors are built on the same core revision and the same CMOS process, then clock differences are just a matter of changing the multiplier.

I thought this was an excellent review. Probably the most useful CPU benchmark review I've ever read, actually. Very thorough, very wide range of benchmarks, very useful commentary on different factors that might be affecting the results on the different benchmarks, and not loaded up with ads (*cough* TomsHardware *cough*).

One thing I found very interesting was the considerable advantage AMD had in the 64-bit benchmarks, especially Cinebench and POV-RAY, and the Linux-based Folding@Home. It seems that the AMD's really shine in 64-bit performance. For those of us, who run 64-bit Linux, where *all* of our open-source software is natively 64-bit, AMD64 gives a pretty great performance boost. Yet another point in favor of open-source :-)

Re:Accuracy somewhat questionable

[PoliTech]

I'll defer to your expertise regarding the architecture, however the author of the article used a qualifier, "Their performance should be identical to that of the real thing." and I also used a qualifier "Accuracy somewhat questionable".

I did think the scatter plot was cool but IMHO the testing can only provide a rough idea of cost to performance of the various tested CPUs.

It was a very interesting review none the less.

Re:Accuracy somewhat questionable

[default luser]

Not only that, but in my experience the fraps hook can have huge impact on your framerate.

I doubt the overhead of FRAPS is all that significant these days. Most games aren't heavily multithreaded to the point that they can stress two cores, so there's plenty of horespower leftover in all these multicore CPUs to power FRAPS.

And even if the FRAPS overhead is significant, you do realize that this will actually hurt the lower-end CPUs more in a comparison like this, right?

If you're worried about variances caused by FRAPS, the FRAPS-captured tests (all two of them!) were the average of five runs, which pretty much removes this issue.

On top of that the motherboards are running different chipsets, simple stuff like drivers for onboard sound cards can have a huge impact on the processing power if they are written poorly.

The motherboards chosen are the best you can get on their respective platforms, and are priced at about the same. You really think using the best motherboards available to both camps is going to negatively affect the results of this review?

The platforms are exactly the same within Intel models and within AMD models. Even the FX-series has the same sound chip and driver as the rest of the Athlon 64s.

Now, they could have gone to the trouble to get matching sound chips, but it ultimately doesn't matter. The benchmarks they published are very close to those I've seen on other websites, making your arguement moot.

There are also only two tests that could be affected by the sound chip used: the two game tests at the start. Ever other test uses ZERO sound.

They really aren't testing anything when they don't have a uniform rig to do so.

Well, I'm still waiting for you to conjure me a motherboard that takes both AMD and Intel processors, all without stupid hacks like entire motherboards on an expansion card. Until you have that, there's obviously NO WAY we can compare these two chip lines in any way.

Re:Accuracy somewhat questionable

[Anonymous Coward]

Except fraps hooks to the graphics board and reads from it which is a very expensive thing to do.

And no, since fraps hurts the bus it will hurt poor performing MBs.

About the MBs, yes I do think so, why else would I have said so? And you are pointing that MBs among series are same, yes they are, but my point still holds, that only makes a uniform test for that series of chips.

Oh and it does ultimately matter what chips you are using, everything counts!

And thank you for once again not understanding the point of my post and point out that you can't get MBs that support both architectures.

Impressive you can get so much wrong at once.

The most useful information

[WuphonsReach]

Ignoring the game-based benchmarks - which are somewhat interesting, I think the most useful information is the Windows Media Encoding & LAME encoding (page 6). And some of the other charts on other CPU-focused benchmarks.

It shows that for the CPUs priced under $250-$300, there's not a lot of difference in performance for a particular dollar value. Both AMD and Intel seem to be on parity in that market segment in terms of performance per dollar. (One exception would seem to be benchmarks like POVRay/Cinebench where there's a distinct gap between the two product lines, which flips around on the Myrimatch/STARS page.)

They divide by the price of the processor only

[Anonymous Coward]

They should divide by the price of the complete machine. To illustrate: if those processors happened to cost 113-1199 cents instead of 113-1199 dollars, we would all buy the best performing, to hell with performance/cost.

Summary, and Flawed Analysis

[MojoRilla]

Bargain basement CPUs do better at $ / work than faster, more expensive ones, because they are so cheap. AMD does well at the low end.

But this doesn't consider the total price of a computer which would help mid priced chips. A $113 CPU is 54% more expensive than a $73 one, so it would have to perform 54% better. But when you throw them into identical $200 systems (case, hard drive, fan, power supply, memory, etc), the $113 CPU (with a total system cost of $313) is only 14% more expensive than the $73 CPU (with a total system cost of $273).

So, while the extremely low end chips do well with this analysis, they make much less sense when you consider total system costs.

Re:Summary, and Flawed Analysis

[MMC Monster]

The problem is now: Is a person is going to use that extra few percent? It used to be (in the days of sub 500MHz CPUs) that everyone could use more power. Now that's not always true. There are people that do fine with a low power, low GHz CPU. And that 30 buck savings may be put to something else (ie: extra hard drive space), and the slower CPU may also waste less electricity as well.

Re:Summary, and Flawed Analysis

[ciroknight]

You mean waste more electricity right? Cheaper the processor, the more goodies they knock out of the chip to keep the price down. That $70 Celeron is built without SpeedStep. The $110 Core 2 has the full sized Smart cache and SpeedStep. The Celeron might be 80% as fast as the Core 2, but the Core 2 will probably use 60% of the energy, meaning the net win (if you can afford to spend $40 more bucks) goes to the Core 2.

In all honesty, it makes the most sense to buy the most "featureful" chip at the bottom of the clock bracket and overclock it. Not only will you have all of the features those chips ship with, you'll likely have more performance than the CPUs at the top of the bracket. You might cut the lifetime of the chip down, but computers today are such commodities that hardly anyone cares if the chip burns out after 3 years instead of 10; they won't be using that chip by then anyways.

Re:Summary, and Flawed Analysis

[gEvil (beta)]

The problem is now: Is a person is going to use that extra few percent?

How true. I just picked up an Athlon64 X2 3800+ to replace my single core 3500+. Naturally, there isn't much of a performance difference between the two. However, having the dual cores does seem to balance the machine out a bit. For instance, I can play a video fullscreen on one monitor and browse the web on the other without having the video stutter if I happened to hit a page with lots going on. Did I need the new processor? Not at all. But I figured it'll at least hold me over a bit longer, especially as software that can take advantage of multi-cores becomes more prevalent.

Re:Summary, and Flawed Analysis

[Ruie]

But this doesn't consider the total price of a computer which would help mid priced chips. A $113 CPU is 54% more expensive than a $73 one, so it would have to perform 54% better. But when you throw them into identical $200 systems (case, hard drive, fan, power supply, memory, etc), the $113 CPU (with a total system cost of $313) is only 14% more expensive than the $73 CPU (with a total system cost of $273).

I want to second that - I am always astounded at how closely the price increments of the cpu's match the cost of a well-configured system. You start configuring a workstation or a server and lo and behold upping the clock rate by 10% costs almost exactly (or slightly below) 10% of the total system price.

The only case when lower priced processors matter is when configuring for capability rather than raw performance. I.e. larger cache, 64bit vs 32bit, thermal power dissipation, etc.

Re:Summary, and Flawed Analysis

[wangerx]

I would add to this notion of diminishing impact. What about the value of time? If a quad core will run 4x but costs 7x the cost impact will diminish within two weeks of use! The 7x cost is a one time cost; say for example $500 more. The 4x impact on performance and time is forever. For an exaggerated point, if I can get four times the work done, every hour saves me three. If my time were a paltry $10 per hour, I would pay for the CPU cost difference in less than a week. Anything after that it gravy! You might say that the article it is relates to the impact on gaming and not real work (I see encoding as real work) but quality has a cost associated with it too. Entertainment can be largely impacted by quality. It used to be said, that you are not having fun until you spend $20 per hour (circa. 2000). Heck, one might even say that bragging rights have a certain value.

Re:Summary, and Flawed Analysis

[Kjella]

But this doesn't consider the total price of a computer which would help mid priced chips. A $113 CPU is 54% more expensive than a $73 one, so it would have to perform 54% better.

Which is of course a meaningless metric to most people, maybe it'd be relevant for a data center but it'd probably be overshadowed by power costs. There are three factors IMO:
1. Can and can't dos. Can you play your game at your desired resolution/fps? Can you play back 1080p HD-DVDs/Blu-Rays? Those have values in themselves.
2. The annoyance factor - tired of your machine locking up or annoying response time? Get a dual/quad core and there's always a core free to kill whichever process went beserk.
3. Time savings. 54% * hours you'll be waiting on the CPU over the lifetime * your time value in dollars. The first depends massivly on what you're doing, if you're a developer constantly compiling, doing video editing, graphics rendering, simulations or anything intense compared to a casual web surfer. The other factor depends a lot on the wage level, general and personal. If you're living in the US or Europe, it might make sense but for a 3rd world citizen it would not, even if you're doing exactly the same. Or if you're a highly prized consultant compared to a burger flipper.

For none of these, speed/dollar is even a remotely meaningful statistic.

Re:Summary, and Flawed Analysis

[LWATCDR]

I was going to say that the Intel motherboards where a little more expensive than AM2 motherboards but I am shocked that this is no longer so.
However ever I will say that for a lot of people that extra CPU performance really doesn't matter. If you don't use it then it it valueless. For most people I would bet they would be better off spending that extra money on more ram then a faster CPU.

Re:Summary, and Flawed Analysis

[darkwhite]

$113 CPU (with a total system cost of $313) is only 14% more expensive than the $73 CPU (with a total system cost of $273). The only case when lower priced processors matter is when configuring for capability rather than raw performance. I.e. larger cache, 64bit vs 32bit, thermal power dissipation, etc. This notion is a logical fallacy. The CPU speed is by no means the only performance-defining part of the system. Upping the amount of memory, the GPU or the speed of the hard drive may well cost 10% of the system as well, but somehow I don't hear a lot of people say "Oh, but a $200 GPU is only 15% more expensive than a $50 one, because it only costs 15% more in a $1000 system".

Lower priced processors matter a lot to those of us who want cheap, fast, cool systems that kick bigger systems' ass in price/performance, and are happy to let others fund the chip companies' R&D budgets. I, for instance, hold a big grudge against AMD for abandoning Socket 939, and I'm worried about its future because Core 2 kicks Barcelona's ass for any PC from $500 up, but I still bought an Athlon 3900+, which can actually be had for $60 in a combo, with the intent to overclock. It makes perfect sense because I'm assembling a tiny sub-$300 cube PC in which anything with a bigger price/performance/power ratio just doesn't appeal to me.

Re:Summary, and Flawed Analysis

[Mick Ohrberg]

Did I need the new processor? Not at all.

How true. I recently upgraded from an Athlon64 X2 4200+ to a 5600+. Did I need that upgrade? Heck no. However, in my case, the motherboard (socket 939) broke down, and rather than chasing around for a comparable 939 mother board (they ARE getting harder to find), I decided to go with an AM2 board, and "while I was at it" upgrade CPU. Oh and I had to get new memory too (DDR2 instead of the old DDR). But I didn't *need* any of it. When it comes to upgrading, it's sometimes something that just makes plain sense - moving off of obsolete sockets for instance. Of course, budgets play in. If you're really on the cheap, an old Duron 900 in a Socket A motherboard may do you just fine, but newer OS bloat doesn't help either.

Re:Summary, and Flawed Analysis

[Anonymous Coward]

Decent S775 LGA motherboard: 150$
Core 2 Duo E4300: 140$
Total: 290$

Athlon 64 X2 3800+ EE: 95$
Asus M2NPV-VM*: 100$ tops.
Total: 195$

Guess which one I'd be using?

But I resent the statement that the X2 3600+ is any good. 3800+ is just 10$ more, it's faster MHz-wise, more cache, and runs with the same amount of electricity.

(*Reported to work with Ubuntu 6.06 & Fedora 6)

PS: Does anyone know how much your average 250W-300W system uses in a year? (And yes, that's all a setup with 4 sticks of DDR2, 1 SATA HDD, 1 DVDRW, 1 floppy, 3 USB devices, and a case fan or two uses.)

(Prices in Canadian, found from random stores in Ottawa)

Re:Summary, and Flawed Analysis

[WuphonsReach]

was going to say that the Intel motherboards where a little more expensive than AM2 motherboards but I am shocked that this is no longer so.

Off-hand, I'd say that in reality AMD and Intel boards of equal (and moderate) capability are almost always within a few dollars of each other. Go take a look at prices on pricescan.com or MWave's motherboard bundles. Last year at this time, Intel board were running about $10 more expensive, but it seesaws back and forth.

It's only the esoteric motherboards (all multi-CPU boards are expensive - and high performance SLI boards are pricey) where this may not hold true. But in those boards, when you're paying $300+, a few dollars either way isn't much.

Re:Summary, and Flawed Analysis

[evilviper]

Bargain basement CPUs do better at $ / work than faster, more expensive ones, because they are so cheap. AMD does well at the low end. But this doesn't consider the total price of a computer which would help mid priced chips.

Actually, I see exactly the opposite...

When AMD* turns out slow, low-end chips, they do so after the high-end original has passed on, and in the process they update the tech, and so tend to give the "value" chips lower power consumption for free. That is of course baring the possibility of buying your CPU right during the transitional period, while the high-end chips use a newer, smaller die, and the low-end chips don't, yet.

(* I can't use Intel as an example because right now their low-end chips are old Pentium4s, which suck power like crazy)

Re:Summary, and Flawed Analysis

[gfxguy]

I would say the vast majority of typical users have overbought their systems, period, including disk space and memory.

Some schmuck comes home from work (where the quality of the computer might matter) and maybe reads email, surfs for the latest movie schedule, and watches a couple of youtube videos his friend sent him... even play a DVD at full screen.

The lowest end computer is going to do all of these things and still waste electricity and cpu cycles.

He doesn't want HD on the computer, he's got an entertainment center that does all that. He's got Tivo. A $5000 system is not going to sync your iPod any faster than a $500 system.

But the real reason a lot of these things are pointless is because for the people who KNOW what they need, and need speed, cost of one component is not as big of a consideration. If people are making money with their computers, $1000 versus $2000 is not a big deal in the long run if they are going to be more productive with the $2000 one.

I'm at about the lowest end on that scale they have there, with AMD X2 3800+. My video editing, even with 1GB memory (it's never fully used, according task manager), is stuck on I/O, not processing. Perhaps more complex editing would need more CPU to keep up with I/O.

I play games. 100+ fps is pointless on my 60hz LCD monitor.

So it's my contention that people are over buying, and have overbought, computer systems for several years now. Frankly, in my opinion, from my observations, U.S. consumers have overbought just about everything. I'm not complaining... just making an observation. If that's all they want, then it's great, because I'm a patient man, and eventually it translates into lower prices for me when it might actually be useful to have.

Case in point: the people who spent the most on Blu-Ray and HD-DVD players had the least selection of movies. By the time I buy one of these systems, I'll have thousands to choose from.

Re:Summary, and Flawed Analysis

[Gospodin]

The proper way to do this analysis is to consider the "efficient frontier" of CPUs. CPU X is on the efficient frontier if all equally priced or less expensive CPUs are slower. If you buy a CPU on the efficient frontier, then you couldn't have paid less and gotten more speed (although of course you could have chosen to accept less speed). This factors out the problem of diminishing returns which plagues "frames per dollar" comparisons.

The efficient frontier is analogous to the same concept in Modern Portfolio Theory, where you plot risk vs. return for a set of investments. The idea is that there is no best risk/return position, but for a given risk you want to maximize your return. Choosing investment on the efficient frontier guarantees this.

Re:Summary, and Flawed Analysis

[Bishop]

Your average 250w-300w system does not use anywhere near that much power most of the time. I would guess that at idle either of those systems would use about 60-70w (assuming onboard video). At full cpu load I would expect the system to consume 70-80w. Possibly as high as 90w. A video card will make a big difference. In a system I tested the nvidia 7900 GTX consumed about 40w at idle and much more under load. When looking at yearly power consumption the power usage of the DVD-rw and floppy drive are effectively zero as those devices are so seldom used. (Unless you are running a DVD piracy shop :-)

For most users the systems will be idle most of the time. (Even when playing mp3s my Core 2 cpu is running at its slowest cpu speed.) Either of the above systems will likely consume about 550 kW-hours of electricity a year. If you run Folding@Home the systems will likely consume about 650 and maybe as high as 780 kW-hours. If you estimate electricity at 8cents per kWhr. You are looking at operating costs in the range of $45 a year. Running folding@home would cost $55 to $65 a year. Using the suspend feature can save you a few bucks every year.

If you are running Linux use the "ondemand" or "conservative" speed governor.

Re:Summary, and Flawed Analysis

[PitaBred]

100+ FPS may be pointless, but having a minimum of 60FPS is ideal. If your FPS ever drops below 60, then you start getting closer to having stuttering, etc.

Re:Summary, and Flawed Analysis

[florescent_beige]

If this had been factored in to the scatter plots shown in TFA, it would have caused a compression of the results along the horizontal (price) axis. Which means yes, the lower priced processors would move away from the maximum price/performance tangent and the higher priced ones would move towards it.

Re:Summary, and Flawed Analysis

[KingMotley]

I agree completely. The cost of the CPU is only one factor. Although I tend to look at the higher end of the spectrum, so I'm deciding what CPU makes the most sense to throw in the $1400 system I've built around it. Putting in a $73 CPU really wouldn't make a whole lot of sense.

Unfortunately, they would then have to run the complete set of tests to be accurate with atleast 3-4 different base systems to give a clear picture. El-Cheapo - Using the noname brands of whatever is cheapest, Budget - Use name brand/common parts, Gamer - Same as budget but toss in a high end video card, High End - Same as gamer but toss in a raid array, more memory, maybe SLI.

Re:Summary, and Flawed Analysis

[RzUpAnmsCwrds]

You mean waste more electricity right? Cheaper the processor, the more goodies they knock out of the chip to keep the price down. That $70 Celeron is built without SpeedStep. The $110 Core 2 has the full sized Smart cache and SpeedStep. The Celeron might be 80% as fast as the Core 2, but the Core 2 will probably use 60% of the energy, meaning the net win (if you can afford to spend $40 more bucks) goes to the Core 2.

The problem is that you're looking at Intel, whose low-end CPUs suck. AMD's $59 Athlon 64 X2 3600+ has the latest process (65nm), power saving features (Cool n Quiet), full AMD-V support, and two cores.

Oh, and the X2 3600+ is massively overclockable, too. Mine hit 2.85GHz (300x9.5) with no trouble - and it probably would have gone higher if my mobo supported higher LDT frequencies.

My system is 100% stable (as far as I know, based on a 36-hour two-process Prime95 run that pegged both cores at 100%, and based on a 12 hour Memtest86+ run). My motherboard is a $49 GeForce 6100 chipset board (right now, my board plus an X2 3600+ sells for $94 on Newegg). My heatsink is a $10 Arctic Cooling Alpine 7. My memory is cheapshit Kingston DDR2-667 (2x1GB).

Including my HDD, DVD burner, GeForce 8600GT, Motherboard, CPU, DDR, and case, I've put maybe $500 into my machine.

Re:Summary, and Flawed Analysis

[LWATCDR]

I often see low end PCs with only 512 megs of ram. Which while usable isn't as nice as a gig.
If I had a choice between an AM X2 3800+ with 1 gigabyte of ram and X2 5000 with only 512 I would go for the extra ram everytime.

I agree with you for the most part. I am compiling code on an X2 3800 every day. I use the Eclipse CDT and debug with it every day as well. It really is fast enough. I can compile the Kernel in a reasonable amount of time. It really is pretty freaking fast. For the average person the X2 3800 is more than good enough. The next machine I build will probably be an X2 with one of the new 45 watt X2s in it.
I would say that for the average person they should get more ram before they worry about the CPU. Probably should get more video card speed also before worrying about getting more CPU.

Re:Summary, and Flawed Analysis

[viper66]

I think your numbers are a little low. My Celeron 566 with an old low-powered AGP card, 2 sticks of memory, 2 network cards and 1 hard drive takes about 80W idle. My watercooled A64 X2 with a 7900GT and 6 hard drives takes about 200W when idle.

Re:Summary, and Flawed Analysis

[NatteringNabob]

[So, while the extremely low end chips do well with this analysis, they make much less sense when you consider total system costs.]

Maybe, maybe not. On Newegg right now (and other similar sites) an AMD 3800+ 65W goes for about $67. A 4600+ 65W goes for about $113.00, a difference of about $45. That $45 will buy an extra GB of memory and most people will benefit more from a 100% jump in memory from 1GB to 2GB than they will from the 25% jump in clock rate, though it probably won't show up in these benchmarks at all. Like the article points out, 'best value' depends a lot on what you plan on doing with the computer.

Re:Summary, and Flawed Analysis

[fishbowl]

"If I had a choice between an AM X2 3800+ with 1 gigabyte of ram and X2 5000 with only 512 I would go for the extra ram everytime."

In one of our applications, RAM (size and speed) makes a much bigger difference than CPU power.
I've seen people put together what they thought were high end systems, only to skimp on RAM, because they've got it in their heads that memory isn't of major importance.

We pay attention to memory timings, and we test them.

Re:Summary, and Flawed Analysis

[Fweeky]

"Core 2 kicks Barcelona's ass"

Do you know something the rest of us don't?

Re:Summary, and Flawed Analysis

[LWATCDR]

I have noticed that when you buy a system they often stick the cheapest ram possible in them.
What people don't get is that it is a SYSTEM! People will put the cheapest RAM and as little as possible into a system , and then throw a big but slow drive on it and wonder why the new CPU isn't that much faster.

Re:Summary, and Flawed Analysis

[WuphonsReach]

My system is 100% stable (as far as I know, based on a 36-hour two-process Prime95 run that pegged both cores at 100%, and based on a 12 hour Memtest86+ run).

That's a pretty good stability test - but may want to also be exercise the disks at the same time. I've seen a situation in the past where the disk controller would corrupt data when the CPU was under load. That one was a real bear - get the OS installed, start using it, then start getting all sorts of odd errors.

(And Prime95 is a lot better for finding transient or timing errors then MemTest86+.)

Re:Summary, and Flawed Analysis

[darkwhite]

I really don't.

The article you cited is AMD marketspeak from April 23.

Barcelona's launch prospects are looking grim. If you recall this time last year, Intel had been showing off Conroe for almost half a year before launch. I have no doubt AMD can make it work in the long run (witness the evolution of K8) but they seem to be having serious problems and running out of time.

Re:Summary, and Flawed Analysis

[511pf]

The problem is that you're looking at Intel, whose low-end CPUs suck. Really? Because I just paid $140 for a Core 2 Duo 4400 that runs at 2.8 Ghz on stock cooling and voltage. It's as fast as any sub-$300 dual core CPU on the market. I wouldn't call that "sucking."

Re:Summary, and Flawed Analysis

[that this is not und]

Can you get the four times as much work done, though? What do you define as 'work'?

Are you rendering animation or designing big complex FPGAs?

There aren't that many 'seats' except in high end engineering work where the performance 'bump' of the current processor lines matters. If your job involves Excel and Powerpoint, the power to render ever more complex garbage psuedo-data with the flick of a mouse is actually detrimental.

Nobody cares if you're writing your novel on an Osborne I or on a machine where you're forced to watch the widgets fling little animated pieces of paper across dialogue boxes in the latest wizzy-whoo turd from Redmond. You're more likely to become distracted and LESS creative if the all-singing, all-dancing bullshit is flashing in your face.

So what kinda work are you talking about?

Re:Summary, and Flawed Analysis

[cibyr]

You didn't buy a low-end CPU. Intel's low-end CPUs are labeled "Celeron", and they do indeed suck.

Re:Summary, and Flawed Analysis

[gfxguy]

And that was my problem with video editing... sure, if I was compositing 8 images together, CPU and memory might be a bigger issue - but I'm just chopping out the boring parts of my home videos and adding transitions between what I kept; so CPU and Memory are NEVER 100% full, I'm always I/O bound... and so my biggest benefit was making a RAID 0 scratch drive to write to.

I'd posit that with faster CPU and more memory, someone might be able to composite a complicated scene as fast as I compile my simple home videos; but that they wouldn't be able to get faster simply by doing less - they'd level off at about the same speed I get, unless they had some super high speed drives.

Re:Summary, and Flawed Analysis

[Bishop]

The numbers I posted are reasonably accurate. Measurements were made with true RMS Fluke amp and volt meters. Taking power factor into account the numbers should be within 15%. Two of the systems tested were tested with the amp and volt meters, and a watt meter. The numbers were within 8%. Are you sure that your meter is accurate?

Is your water cooled AMD X2 overclocked? Are you sure "Cool'n'Quiet" is enabled? This will make a difference at idle. I am not surprised that your Celeron 566 burns 80w. A P3 550 I tested used about 85w at idle. At the time I was surprised as it was assumed that these older P3 systems ran cool.

One of the systems I tested was close to the configuration the grandparent was looking at. It was a socket 939 mainboard with AMD64 X2 3800+, 2x512MiB ram, Nvidia chipset with onboard video and a Seagate hard drive. At idle it consumed 67w.

Re:Summary, and Flawed Analysis

[RzUpAnmsCwrds]

Really? Because I just paid $140 for a Core 2 Duo 4400

That's not low-end. My X2 3600+ sells for $59 on Newegg. I have no doubt that your CPU is better than mine. It's also more than twice the price.

And, FYI, it's not that Intel's low-end CPUs have crappy performance. They suck because they don't have EIST (power saving) or VT/AMD-V. They also run on a crappy 800MHz FSB.

$60 on the Intel side buys you a 2.66GHz Pentium D. The cheapest Core 2 CPU is the E2140 for around $80, which runs at 1.6GHz.

Re:Summary, and Flawed Analysis

[viper66]

I am using a Kill-A-Watt power meter. The X2 is overclocked and Cool'n'Quiet is not enabled because it causes problems on an overclocked CPU.

Re:Summary, and Flawed Analysis

[fishbowl]

>CPU and Memory are NEVER 100% full

You're thinking in terms of memory capacity; our bottleneck in this case is memory bandwidth and capacity is far less significant.

Question: Cost of the energy to run the CPU?

[Futurepower(R)]

Note that the Power consumption and efficiency section in the linked article shows CPU power plotted against cost of the CPU, rather than CPU power plotted against the cost of the electricity.

For computers that are on much of the day, the cost of the electricity over the perhaps 4-year life of the system is significant, and more important than relatively small differences in the cost of the CPU.

Although the article has some flaws, it is very useful.

Re:Question: Cost of the energy to run the CPU?

[Anonymous Coward]

The cost of electricity over a 4-year life depends on where you live and how much you use your computer. In a place like Quebec where it is cold and electricity is cheap, the cost is low and the heat produced reduces the heating bill. In California the price is high and you need extra power to evacuate the heat.

Re:Question: Cost of the energy to run the CPU?

[Anonymous Coward]

They give a bunch of charts, and rank the processors; AMD wins, with single Intel item in the top 3. Then their text says:

"Despite having nearly the highest render energy of the lot, the X2 3600+'s bargain basement price keeps it atop the power efficiency per dollar standings. The E4300 claims second place, offering higher power efficiency and a lower price than the 4400+. Our bronze medal winner is the E6400, whose power efficiency is quite substantially above the E6300's, despite the small pricing gap between the two chips.

That said, we couldn't get away without mentioning the Core 2 Quad Q6600, which tops the power efficiency scale despite its fairly reasonable price tag. If you do 3D rendering work for Al Gore, this is the chip to get. Still, it's worth pointing out that the Q6600 doesn't have the lowest idle power consumption (see our full results here), so the X2 3600+ may yet be the friendliest to your power bill if you don't run compute-intensive tasks very often."

Which makes no sense against their own data, and in particulary the Q6600 turned up middle of the pack all around on anything-per-price. Did the same guy make the charts and write the text ?

This is why, when I saw this article in the firehouse, I voted it down. It's so non-sensical that it makes me question the accuracy of all the numbers in the review. If I were buying a chip, I would not depend on anything in this article.

Re:Question: Cost of the energy to run the CPU?

[ljw1004]

It doesn't show CPU *power*. It shows total *energy* required to complete given a rendering task. Total energy is roughly similar to power/performance. So for computers that are on all day, you just look for a CPU that's low on the y-axis for this graph.

Re:Question: Cost of the energy to run the CPU?

[Nanoda]

This was my top concern when evaluating for a new PC. My current PC (an old dual 2Ghz Xeon workstation) draws more power than a standard 300Va UPS can provide, and heats up my small living room almost unbearably.
I calculated that one watt year (using a single watt constantly for an year) costs about $0.90 at current prices (~10 cents/KWh). I figure my PC costs me at least $150 a year (it's on 24/7 for remote access, etc).

My planned replacement components include a power supply with PFC, one of AMDs new 45W CPUs available in a few weeks, and Nvidia's 8600GT which is reported to use only 42 watts at peak. None of these components are specialty items, in fact they're all fairly inexpensive.

Re:Question: Cost of the energy to run the CPU?

[xs650]

"My planned replacement components include a power supply with PFC [wikipedia.org],"

That's a nice thing to do, but home users are rarely billed based on power factor. Industrial and business users frequently are. If your PFC power supply is also more efficient, then that's a reason for home users to buy one.

Re:Question: Cost of the energy to run the CPU?

[turing_m]

Maybe he runs a UPS as well? Lower VA will allow him to get away with either a lower spec unit or run longer.

E4300 best because of huge o/c headroom

[dhakbar]

Considering the E4300 can often be run at ~3 GHz per core on stock air cooling, it ends up having more bang per buck for those inclined to overclock.

Re:E4300 best because of huge o/c headroom

[Etrias]

You could say the same for the AMD 3600+.

It's the same for everything

[Colin Smith]

Buy last year's hardware at a fraction of the cost and let someone else take the depreciation hit/development cost. You benefit from lower depreciation and usually, better reliability. There's always a dogleg increase in cost for the latest and greatest.

Works for cars too.

Re:It's the same for everything

[dgatwood]

Flip side, you're buying older, slower hardware. When the next release of your choice of OS comes out and it needs more CPU or GPU power to behave well, or when you suddenly find that video playback is stuttering because the new codecs eat CPU like candy, you have to upgrade, while you might not have needed to do so if you had bought the top machine.

There are costs associated with upgrading, and I don't just mean the raw financial cost. There's a cost of time to move all your data over, a cost to the environment from disposing of that old computer, a cost to your sanity from the number of times you lose your cool and start swearing at the computer when it stutters, etc.

If all you do with your computer is send email and do basic word processing and maybe lightweight web browsing, you're probably right. You could get away with a low end computer (assuming that's all you will ever do with the computer). On the other hand, you could probably get by with buying a used computer from five years ago just as well. For anybody who does anything even remotely significant with the machines (even something as simple as watching YouTube videos), that CPU power is still important and will become much more so in the future.

Re:It's the same for everything

[couchslug]

If you are poor enough that the money matters (I sure am!) then do the geek thing and perform some computer work for the money.
I upgrade computers for other people, and part of the fee is the hardware I replace. I now appreciate why everyone (else) NEEDS bleeding edge hardware.

"Works for cars too."
Boy howdy it does! Fix them yourself and you move from saving some money to saving a sh1tload and paying for top-quality tools while you are at it.

"There's always a dogleg increase in cost for the latest and greatest."
My "doglegs" are cheap too. $60 per neuter = $15 per dogleg.
We only get rescue animals, often from a$$holes who wanted a "fashionable" animal and lost interest.
I have all the computers/vehicles/critters I want, and didn't break my wallet to do it.

"The results may surprise you"

[tkw954]

The results may surprise you.

How come does it seem like all the summaries lately have to have a cliffhanger? I'm all for reading the articles, but give us the results. This is a news site, not a murder-mystery.

Re:"The results may surprise you"

[jd]

Call that a cliffhanger? In my day, we had to make our own cliffhangers. With our bare hands! And we were grateful for it.

Re:"The results may surprise you"

[starwed]

That's been bugging me too. Enough with the trite write-ups!

Re:"The results may surprise you"

[noidentity]

Tune in next time to see if future articles end with a cliffhanger...

Re:"The results may surprise you"

[evilviper]

It's not actually a cliffhanger, but pure hyperbole.

I, for instance, wasn't surprised in the least by the results, nor can I see any reason anyone would be.

Re:"The results may surprise you"

[Dracolytch]

Because ending in a cliffhanger will result in higher traffic, and thus more revenues for slashdot.

~D

Re:"The results may surprise you"

[Anonymous Coward]

I hate to say it, but I _was_ surprised.

Re:Wait for the next price drop...

[DogDude]

You can say that every single day and still be right. I'm a big fan of buying PC's from my local thrift stores for $20 each. PC's are, by themselves, probably the worst investment that I can think of. Not even American cars depreciate as fast as PC's. I always tell people that unless you're playing games on a PC (which is an insanely expensive hobby), or doing something important, just get the cheapest thing you can find.

Zero offset

[Harlequeen]

It all depends where you draw the line for zero. If you only include the price of the processor of your system, then price will win. However if you factor the price of the whole system, the charts would look very different. Cheap spark plugs maybe ok in some VFM terms, but you wouldn't fit them in a Ferrari, or more realistically any new car.

Re:Zero offset

[that this is not und]

Cheap spark plugs maybe ok in some VFM terms, but you wouldn't fit them in a Ferrari, or more realistically any new car.

That depends on what makes the spark plugs 'cheap.' More analysis is needed. Are the spark plugs cheap because they're made with lower-cost inferior materials, or is it because the logo on the box doesn't match any seen on vehicles shown on TeeVee in NASCAR events?

Which gives rise to articles like the one that inspired this discussion. Some of us will actually go out of our way to choose the 'brand' that subsidizes the advertising 'industry' the least as long as there is equivalent value in the product.

Would you be better off with several computers?

[Anonymous Coward]

For twice the money you don't get twice the performance. The aggregate difference was 275% between the most powerful and the weakest. The difference in price was about 1300%. Ignoring energy, you would be much better off with two cheap computers as long as you could divide the tasks. For instance, you could have one computer to compile on and another for listening to music/ wordprocessing/ browsing the web/ answering email. Not only that but you could turn one computer off when you didn't need it and save energy thereby.

For tasks that can be shared, it would seem that a cluster of cheap computers is more cost effective than one hotrod box.

Re:Would you be better off with several computers?

[sethstorm]

Ignoring energy, you would be much better off with two cheap computers as long as you could divide the tasks.
...while ignoring quality, and space taken up by both machines.


For instance, you could have one computer to compile on and another for listening to music/ wordprocessing/ browsing the web/ answering email. Not only that but you could turn one computer off when you didn't need it and save energy thereby.


For tasks that can be shared, it would seem that a cluster of cheap computers is more cost effective than one hotrod box.

Until you end up having to deal with returns and questionable quality of the components. Might as well go with the hotrod and end up with less replacement going on until it's really needed.

Just wait!

[dtolman]

Once the page is slashdotted it really will be a cliffhanger...

Is AMD faster for 64-bit?

[emil]

I heard somewhere that certain Core 2 Duo high-performance functions were disabled for 64-bit code.

Given that Vista will be the last 32-bit Microsoft OS, is AMD a better choice, also given the Intel errata?

Re:Is AMD faster for 64-bit?

[ceeam]

If you don't need OSS graphics drivers AMD is a damn good choice. And it has been for the last 20 years except small K5 vs Pentium period.

As for Vista being the last 32-bit MS OS, it could well be the last MS OS period. Anyway - it really stretches 32-bit address space as it is trying to fit its fat ass within i386 pants.

Re:Is AMD faster for 64-bit?

[Anonymous Coward]

If you need the highest possible graphics performance you can get on fully open source drivers, AMD is your only choice. Or rather, ATI Radeon X850. Older Radeons have working fully open source drivers, unlike nvidia cards.

Re:Is AMD faster for 64-bit?

[ceeam]

I have Radeon 8500 in this machine - is that old enough? Where do I get a "fully working" (as in "beryl works great and does not hang") open-source driver? The one in xorg is not it IME.

Re:Is AMD faster for 64-bit?

[that this is not und]

AMD is a damn good choice. And it has been for the last 20 years except small K5 vs Pentium period.

Uh, I guess so. But you were plugging an Intel 80287 in alongside that AMD '286 chip, and if not, you're just another Wordperfect jockey.

Re:Is AMD faster for 64-bit?

[Anonymous Coward]

Note that I said "fully open source", not "fully working", since bugs and gotchas still occur especially with hardware revisions that are not used by the developers; at least the bugs are fixable in OSS. However, I am nevertheless surprised that you have problems with Radeon 8500, which is certainly old enough that it should be both stable and fully working (see here for OSS radeon driver status). My own Radeon 9250 works perfectly, and it's a couple of revisions later hardware than the 8500. However, the default x.org configuration used AGP 1x, and when I tried to maximize performance I couldn't get it to run stably at AGP 8x, although AGP 4x works. There's no telling if the reason is my el-cheapo VIA motherboard (VIA is well known for hardware bugs) or some compatibility problem in the kernel or the radeon drivers. But nevertheless it worked out of the box on Debian unstable and has kept working for nearly two years now. Perhaps you could try updating to the latest x.org or even the nightly builds.

Power management still has a ways to go

[bhmit1]

I'd like to see computer manufacturers get to the point where all non-essential components can power down when a machine isn't using them. This would be huge for server rooms, where most machines are there waiting for users to connect. For my mythtv server, it's running non-stop, but hardly using any cpu until it's recording or playing back. Same goes for a mail/file server. The 100-200 watt idle numbers are wasteful, lets get this down to 10-20 watts. Hibernating or suspending doesn't work when you need to be standing by to service a user.

Re:Power management still has a ways to go

[Anonymous Coward]

HDs, CPUs (speedstep etc), monitors and video cards do this already - it's not perfect, of course, you need the CPU running at some level so it knows when to come back... Even PSUs power down to some extent.

p.s. If a server room is hosting unused machines then chucking them out or virtualising them (running many virtual servers on a single box) is the solution.

Re:Power management still has a ways to go

[evilviper]

The 100-200 watt idle numbers are wasteful, lets get this down to 10-20 watts.

Stay away from the high-end, and computer are nearly there already, thanks to AMD pushing CnQ/PowerNow on everything (while Intel STILL omits SpeedStep on their low-end chips).

The only real problem/exception seems to be GPUs, so for now, you have to go for a lower-end, (preferably fanless) video card to be safe.

Re:Power management still has a ways to go

[moeinvt]

"The 100-200 watt idle numbers are wasteful, lets get this down to 10-20 watts."

Sorry dude, there are a lot of brilliant minds in the semiconductor industry working on this, but it's a perpetual challenge. The problem you described with the static power dissipation is the natural result of shrinking device geometries. As you scale down the physical sizes of the devices, they dissipate more power just by means of being turned on (i.e. whether they are switching or not).

It would take a major innovation in device fabrication or power management to get tens of millions of transistors manufactured in a 65nm (or smaller) process down to 10-20 Watts of static power consumption. You could probably make a few million $ and possibly win a Nobel prize if you can figure that one out.

Re:Power management still has a ways to go

[pz]

I've been thinking for a long time that CPU manufacturers could do well to promote low power chips through the following tact: take an older CPU design (say the venerable PIII) and reimplement in the newer technology. A 1GHz PIII is a reasonable CPU for every-day things like surfing the web, reading email, watching videos (with the help of an MP4 chip), and so forth. At original spec, they dissipate 35 W or so. Current-generation CPUs dissipate 2-3 times as much power, have 3-4 times as much cache, run at 2-3 times the clock rate, have memory systems 4-5 times faster, and on the whole run somewhere between 5 to 10 times faster. How about taking the massive improvements in device design, fabrication, architecture, power management, and so forth that went into these impressive achievements and re-implementing the lowly PIII 1GHz, but at 5-10W maximum power?

Re:Power management still has a ways to go

[Quikah]

This is the ultimate goal of virtualization in the data center. Once everything is virtualized you can cram as many machines as possible on a few servers, once they start getting loaded you move the virtualized systems to a new host, then back again when the load goes away. Maybe even power off some of the unused systems at night and bring them up in the morning. All the while keeping 24/7 access to your systems.

Re:Power management still has a ways to go

[Plekto]

How about taking the massive improvements in device design, fabrication, architecture, power management, and so forth that went into these impressive achievements and re-implementing the lowly PIII 1GHz, but at 5-10W maximum power?
****

You can do this with some motherboards by underclocking the processor. That Celeron D 360, for instance, is essentially a 3.6Ghz Northwood with much lower heat. Underclock it and presto - it runs at 20W power(we're talking stock cooler with the fan *off*!) as fast as an old 2.4Ghz P4.

These can make for great appliances.

P.S. - Could Slashdot implement a quote function? Please?

Re:Power management still has a ways to go

[bhmit1]

It would take a major innovation in device fabrication or power management to get tens of millions of transistors manufactured in a 65nm (or smaller) process down to 10-20 Watts of static power consumption. You could probably make a few million $ and possibly win a Nobel prize if you can figure that one out.

They already have stuff now that can turn on and off certain parts of the processor to conserve energy. So, thinking a little out of the box, put an additional core on a chip that runs similar to what you see in embedded systems (think <10 watts, maybe around 50mhz or so). Transfer your registers from one core to the other and use some fancy power distribution technology to turn off whichever core you aren't using. Suddenly you have a dual speed cpu that goes from a few watts to handle the pings, interrupts, and other minimal cpu task, but when you need it, can be ramped up to a 2ghz chip. Where's my prize? :-)

The cpu would probably be relatively easy in the grand scheme. Getting all the components to work efficiently is more of the challenge. GPU's could be configured to power completely down with the monitors. Harddrives are already coming out with hybrid flash storage to reduce spin up's for small writes. The power supplies could use a few tweaks to efficiently handle the significant drop in power consumption. Then, all we need is usb 3.0 to come along and have the ability to tell devices to turn on and off so they can stop drawing power. Personally, I think the problem is less technology and more motivation, and silicon valley is just starting to get motivated. If they don't, China will, they have far too many new power plants not to see the need for conservation.

Re:Power management still has a ways to go

[Anonymous Coward]

> re-implementing the lowly PIII 1GHz, but at 5-10W maximum power?

Hasn't Intel already done exactly that ?
Core Solo and Core (2) Duo ULV are already within that range,
while A100 and A110 TDP is 3 W.

Re:Power management still has a ways to go

[darkwhite]

Bullshit.

Any chip, if properly engineered, can turn off parts of itself completely, eliminating all gate leakage in those parts. It can also clock down to the lowest rate necessary to process interrupts, which is millions of times slower than its operating clock speed (though anything below 1/4 the top speed is usually not implemented due to diminishing returns). So there's nothing preventing the CPU from flushing the caches (or not, depending on depth of sleep) and powering down everything except the interrupt scheduler and handler and a few other small modules. Future designs from Intel have the ability to power down lots of their components, and AMD has stated that they can at least turn off individual cores in Barcelona.

Progress in power efficient chip design is slow because often the biggest power wasters are elsewhere, so making the chip super-efficient wouldn't make a huge difference (aka passing the buck) - hard drives are not slowed down, unused memory modules are not powered down, PSUs are inefficient, LCD backlights are inefficient (yay LED backlights), etc. But it is being made, since datacenter operators and laptop consumers are very interested in it.

pure economic sense = not surprising

[irw]

Im sorry, but this article shows nothing that isnt obvious from the economics of the CPU market.

The time a CPU spends at the top end of the market is small compared to the overall time it is available. This time is also at the lead-end of its production when availability is small. Rarity (and demand) create a premium over and above the actual worth of the CPU.

By the time other CPUs displace it, such as CPU will no longer be in as much demand and the manufacturer may have to drop prices below or close to cost to sell end stocks. Thus they take advantage of the novelty of top-end CPUs to recover their development costs in advance.

Basically, new CPUs are overpriced for what they can do, and tail-end CPUs are underpriced, which is exactly what the article shows.

The results may surprise you...

[Anonymous Coward]

The results may surprise you

The sheer number of animating ads surprised me the most. How can anyone even read the article with that many moving ads right next to the article?

Re:The results may surprise you...

[Jaqenn]

How can anyone even read the article with that many moving ads right next to the article? Firefox. Adblock. Bliss.

If you need links for those, here you go:
Firefox: http://www.getfirefox.com/
Adblock: https://addons.mozilla.org/en-US/firefox/addon/186 5/
Bliss: http://www.youtube.com/watch?v=ppnaHjcypTY&mode=re lated&search=/

Re:The results may surprise you...

[Anonymous Coward]

Oh man, you are genius. I've been looking for "More" for a while. It used to be on a site for shorts, but it has long since passed. I eventually gave up.

Quite an unexpected and pleasant surprise.

FYI, the song used is Elegia by New Order

If the results suprise you...

[Anonymous Coward]

...then you've never bought hardware, or worked on an application where CPU performance was a bottleneck.

There's always a price curve. The latest-and-greatest always sells for a greater premium than it warrants in strict performance terms. This has been the case since The Megahertz Wars and before.

For some applications, "performance per dollar" isn't the paramount concern--it's raw performance. The quad-core chips, on a raw level, outperform the dual-core and single-core for most applications. Some people want that. Those that do, those at the bleeding edge, are willing to pay a premium for it. And the companies selling chips will want to extract that premium from the "gotta have it" users before lowering the pricing on a chip to line up with the low end of the spectrum. This isn't news. Nor is it surprising.

If you're looking to build one box that can do the most possible, shoot for the high end, and damn the cost. If you can scale well and price is the main concern, hit the low end of the scale and buy in bulk.

As others have pointed out, the fact that the middle of the spectrum in point of fact sells well shows us not that consumers are necessarilly irrational, but that the tests are flawed. I could probably get 50 PIII's for the price of one Q6800, and the PIII's have more horsepower. Great, but by the time I've built them into machines, racked them, powered them, cooled them, etc., in my datacenter, did I save money? On the other hand, does my web server really need 4 cores, when it hits 10% CPU load on a good day on a single core Athlon-64? The middle of the spectrum is usually the rational tradeoff of "pure CPU performance" vs. "cost to build out machines."

Which doesn't mean that the study is bad--it's a great example of being what you pretend to be. It shows the performance tradeoffs against cost. But if you use this to determine what CPU to put in that new server you're ordering, without looking at other factors, you have rocks in your head.

don't forget the e2160

[mo]

Too bad they left out Intel's newest low-end core2 duo proc, the e2160.
It's basically an e4300 with less l2 cache and 40 bucks cheaper.
The e2160 starts to put a lot of pressure on AMD's low cost dual-core offerings.

Re:don't forget the e2160

[sjwaste]

2160 is $91, x2 3600 is $64 (both Newegg prices). That Core2Duo costs 40% more than the x2, so while it's a step in the right direction, its not there yet. Plus, factor in that you can get a good AM2 board for ~70, whereas you have to spend north of 100 to get comparable features for Core2 (ie, Nforce 550 vs 965P chipsets).

Since they both take DDR2 memory, you can quickly add up the cost of moving from one platform to the next (assuming one already has a PCI-E card, just for the sake of argument). Figure $70 for 2 gigs of DDR2, and we're at 205ish for the 3600 system vs 280ish for the 2160 system (figuring a gigabyte 965p board). Definitely not a trivial difference to overcome.

That said, as consumers, we're really benefitting from this. I hope that it can continue, we're getting great CPU's out at a fast pace, and the bottom offerings like the 3600 and 2160 are still VERY good processors. Back in the day, the bottom of the barrel truly sucked.

Re:don't forget the e2160

[supremebob]

They also left off the new Core 2 Duo 6420, which is basically the 6400 with double the cache at the same price point. I was a bit annoyed by that omission, since I just got one for myself.

Ascending sort based on price...

[fitten]

So... the conclusion is to buy the cheapest CPU. The charts all look pretty much like they're just sorted by price, ascending. Guess what... that's the way it is and always has been... each speed grade increase adds a premium (both for artificial/marketing reasons and because the yield is typically lower, at first, but after the process matures, it's mostly artificial).

X6800 owners...

[ricepudd]

Just a word of warning, don't look at the graphs. You'll not like them.

Insanely Expensive?

[thestuckmud]

playing games on a PC (which is an insanely expensive hobby)

I'm going to go out on a limb and guess that you have never tried yacht racing, which really is an expensive hobby. I play games on a computer which cost under US$2000 three years ago. New games are stressing the system, so I'll have to upgrade to a faster CPU and graphics card (another $600) this year. Including games, that works out to less than I spend on my bicycle.

Games cost $60, less if you are willing to wait, often for a hundred hours of play time. No travel expenses. No special clothes required. Hell, gamers don't even need to buy deodorant. Ever priced a round of golf at a good course? How about membership at the club so you can play there?

I know... if you want to compete with the best players or impress your friends you may choose to buy lots of bleeding edge hardware. My point is you don't have to do so if you just want to play games.

Re:Insanely Expensive?

[grumbel]

### which cost under US$2000 three years ago. New games are stressing the system, so I'll have to upgrade to a faster CPU and graphics card (another $600) this year.

A game console costs between $200 - $600 and lasts you for five years or more. There are certainly more expensive hobbies then PC gaming, but then there are also much cheaper solution to play a game.

Re:Insanely Expensive?

[object88]

Hell, gamers don't even need to buy deodorant.

I call foul on this! Something smells rank about your statement...

Re:Insanely Expensive?

[rthille]

On the other hand, most of the expenses for yacht racing go solely to the boat owner...
Sure you want some clothing, but other than that there isn't much crew has to spend money on...

Re:Insanely Expensive?

[Anonymous Coward]

Except that a game console will ONLY play games, whereas a PC does numerous other things. I would guess that nearly every house with a game console also has one or more PCs. As such, the question is not "spend $1200 on a PC or $400 on a console" but "spend an extra $400 on my PC to make it able to run modern games or $400 on a console". Much more parity. And if you only play games that don't need a lot of computing power, such as World of Warcraft, there is no extra cost, making the PC CLEARLY a less expensive gaming platform.

Re:Insanely Expensive?

[petermgreen]

A game console costs between $200 - $600 and lasts you for five years or more. There are certainly more expensive hobbies then PC gaming, but then there are also much cheaper solution to play a game.
depends on the genre you want. Beat-em-ups and driving games are well suited to console controllers and have a wide selection availible on consoles. shoot-em-up and RTS games are not well suited to console controllers and as a result many simply never come out on consoles or are crappy compared to the PC versions. Also some of us like mods which are pretty much non existant in the console space.

P.S. the last peice of computer hardware i bought for gaming reasons was a graphics card costing less than £25 (approx $12.5). What game is it that requires $600 worth of processor and graphics card?

Re:Insanely Expensive?

[petermgreen]

£25 (approx $12.5)
sorry that should have been about $50. I got the conversion factor backwards.

Re:Insanely Expensive?

[Anonymous Coward]

"...expenses. No clothes required. Hell, gamers..."

Fixed.

Re:Insanely Expensive?

[Castar]

Sure. And if golf is too expensive, just buy a soccer ball, right? They're both sports, so they're the same thing...

Needs overclocking

[llZENll]

The best performance per dollar has almost always been through overclocking, which they don't even cover. The e6600 can be overclocked from 2.6Ghz to near 4Ghz on air almost doubling its performance per dollar at every level, probably putting it at the number one position in every test.

Re:Needs overclocking

[that this is not und]

That depends on what you mean by the term 'performance.' If 'reliability' is one of your performance factors, then 'overclocking' is not a panacea, rather it's another negative factor, and one that requires expensive evaluation, since it's not included on the vendor's spec sheet for obvious reasons. (the vendor already 'overclocks' to the degree possible while still meeting the specsheet requirements)

So it's okay to 'overclock' as long as your applications are trivial and it the occasional 'unexplained' crash is okay (you can blame it on Microsoft!)

Very happy with Dell AMD x2 3600+

[adrew]

You can't argue with the value of the entry-level processors. I bought the system below a couple of weeks ago for $688, including shipping and tax. Dell had a coupon for $350 off any system $999 or more, so I played with the options until it was exactly $999, then applied the coupon.

Dell Dimension E521
AMD x2 3600+, 2 GB RAM, 160 GB HDD, nVidia 7300LE, 16x DVD burner, card reader, keyboard, mouse, Vista Home Premium, Dell 22" widescreen LCD

For a home computer, performance is great. Vista is nice and snappy and it runs everything I've thrown at it without any problems. I work at a university, so I bought a copy of Office Enterprise for $30 and it runs beautifully. The speech- and handwriting recognition works great and doesn't bog down the machine at all.

Re:Very happy with Dell AMD x2 3600+

[neonmonk]

I saw Bigfoot once! It made a sound that I don't care to hear twice in my life.

Re:Very happy with Dell AMD x2 3600+

[adrew]

I'm a Mac user, so I wasn't too thrilled about buying a PC either. But due to an arm injury I needed a machine with voice recognition that can take dictation. Vista has it and OS X doesn't, so I got the Dell.

It was time for a new computer anyway. Initially I was going to get a Mac Mini and dual-boot Vista, but it would've cost more than twice as much, even without a monitor...

From a Mac user's perspective, Vista is a damn sight better than XP or Win9x. It's still a little annoying, but you don't have to constantly tinker with it like you did with previous versions of Windows.

Re:Very happy with Dell AMD x2 3600+

[stewbacca]

Yeah, I'm surprised more of us Mac users don't pipe up and at least give credit to MS for a more Mac-like OS. I wish we would use Vista at work, but none of our hardware is fast enough. As much ridicule that Vista is getting, the early reports (at least my perception) is that it IS more stable, and things do "just work" better than the past 10 years of frustration....as long as your hardware is ultra-modern. I think a lot of the Vista venom is geared towards the fact that it looks nice and is "different" than XP. It is probably the same phenomena that makes XP lovers vitriolic against Mac OS X. With this said, though, I STILL can't find anyone who uses Vista, and all the stores password their computers...arghhh...Anyone want to send me theirs so I can install it on my Intel Mac?

As for the Mac-mini price comment, I'm not sure how it is 2x as much as the $700 Dell machine. For $1400, you'd have a maxed-out mini with a pretty big screen to boot. Personally, I'd rather have a full-form PC than a Mac mini, even as hardcore Mac that I am.

Re:Very happy with Dell AMD x2 3600+

[adrew]

Well, a comparably-equipped Mac Mini, with 2 GB of RAM, 160 GB hard drive, DVD burner, keyboard/mouse and USB modem is $1,376 on Apple's site. And that doesn't include a dedicated graphics card or monitor. Figure $300 for a nice 3rd party monitor and that's $1,700, nearly $1,000 more....

Need underclocking info

[fishbowl]

I don't want to overclock for performance, I want to *underclock* for cooling (for sound reduction.)

I never seem to find any information about what CPUs and motherboards allow you to purposely drop the performance, or how that would affect cooling. Even in silent PC circles, I find mainly people trying to maximize performance and then compensate for the cooling issues and associated noise. Suggestions?

Re:Need underclocking info

[viper66]

Search for 'Quiet and Cool' for AMD and 'Speedstep' for Intel. All Athlon 64s support Q&C. I don't know the status of Speedstep in Intel CPUs. The motherboard has to support Q&C also so you'll want to confirm that if you are purchasing one but I think most do.

Re:Need underclocking info

[Anonymous Coward]

Core 2 Duo E4300's support Speedstep, and in the case of the ECS (Via Chipset) P4M800 series mobos, they have an in-bio option to run at minimum cpu speed on bootup (which you may not like, since that's 200mhz!). However they're the cheapest new-gen intel chip out, only missing the VT extensions (which most people won't have a use for anyways!) and they perform decently well (at least as good as my p4 3.0 ghz had, and with generally lower core thermal temps, although peak is about the same.)

Just my 2c

Re:Need underclocking info

[Anonymous Coward]

Check the forums at www.silentpcreview.com you'll most likely get an answer there.

Comment removed

[account_deleted]

Comment removed based on user account deletion

same goes for

[TheAxeMaster]

Hibernating or suspending doesn't work when you need to be standing by to service a user. Same goes for hookers.

J. Dzhugashvili?

[XNormal]

J. Dzhugashvili? Is the full name Josif ("call me Stalin") Vissarionovich Dzhugashvili by any chance?

Re:J. Dzhugashvili?

[gr8dude]

Yes, you're correct; Stalin was just a nickname (its meaning is close to "made of steel",).

I own that 3600+ X2 and... no.

[Tatarize]

The damn thing is a trooper. Seriously for the 70 bucks I can clock the thing at 2.7 ghz even though it's base is 1.9. That damned Brisbane core is a trooper. Seriously, it overclocks 42%. Really it's enough for what I do, that I don't bother (also as a downside the chip has poor temp readings, everything says something different so I can't tell a meltdown condition from room temp) -- Honestly, I'd buy another of the $73 if it cost $113 and the $113 was $73.

Looking at it from a business lens

[ducomputergeek]

Had a total of 263 computers, all Macs, and was involved in Video Production. That including 3D rendering and a full post setup with Shake. We had a full time 150 node rendering cluster set up on Qmaster & Xgrid and then the other 103 units would render tasks when available. Now all but 3 units in the building was running Macs, this included the Office and accounting people too. We kept 3 Windows boxes in case we had something come in on MAX that needed to be converted for use in Lightwave. Plus there was a couple tools on Windows that just didn't exist on Mac.

In that business, time was money. The quicker the project rendered, the sooner you could get to the next project in the queue. I was reading the article and scratching my head on some of it. Especially the "points" on render time. How about a real world measurement as in "Seconds it took to render" for the ration. On one of the tools, I believe it was the video encoding, the top line Intel rendered almost 10 minutes faster.

You may say, "Yeah, but you can buy 15 the processors of the low end AMD for the cost of the QuadCore Intel chip". Yes, but when you start adding up all the other costs, I bet the total system costs are like $500 vs. $2000 all things being equal. So your talking a price advantage of 4 to 1. Still, it takes 4 times more space to house those units. (Space does cost money) Not to mention that 4 computers generally are going to produce more heat and draw more power. 1 700watt supply vs 4 500watt supplies. Start adding up those electricity rates over 36 months and start to figure out what the TCO is going to be.

That's just how I look at the data.

They forgot a few...

[Plekto]

As expected, the less expensive processors work best for the dollar.

But they forgot the real budget models like the newest Celerons. It would be interesting to see how an overclocked budget AMD or Celeron(air cooled, say no more than $30 spent on the cooler) would fare versus the big boys.

They may be single core. And have less cache. And run at a slower FSB. But for $50-$70, I bet they are only slightly slower when overclocked.

Not a useful comparison

[podom]

This article is deserving of ridicule. Only if the total system cost of a computer were the cost of the CPU would the testers' performance rations (MHz/$, FPS/$, etc.) mean anything, but that's not the case.

Take for example the X2 3600+ and 6000+ at $73 and $241, respectively. They test these chips in identical systems whose only cost difference would be the price of the processor. The 6000+ costs 3.3x as much as the 3600+, but if you used the system cost, you might be looking at $900 vs. $1068, for example, a difference of ~19%.

The testers aggregate performance index shows the 6000+ at 150% the performance of the 3600+. With my two example systems, I get 50% more performance for only 19% more dollars. That makes the 6000+ seems like a pretty good value to me.

The article points out that a system based on the FX-72 or FX-74 would suffer in value from the requirement of using an expensive motherboard. This tends to make my point, that the total system cost is a much more reasonable number to use in calculating value. If you're building an otherwise outrageously expensive system, skimping on the processor saves you very little. In a bare-bones budget computer, it makes sense to use a cheaper processor because it reduces the system cost by a much larger percentage than in a high-end box.

-podom

(whew)

[Ayanami Rei]

Thank you for that spot-on response; save me the trouble.

Can you also inform Slashdot about the benefits of overclocked Santa Ana cores from AMD, and the different between software RAID, fake hardware RAID, and real hardware RAID?

Doesn't scale.

[Ayanami Rei]

Timeframe to the stuttering of your codec will be in the same generation window of the lower cost version of the same model CPU. You just push back the need to upgrade by about a year. So instead of a 3 year upgrade cycle you're on a 4 year cycle. Yet you're not paying 33% more, but 50-80% more for the priviledge.

Does the time-cost of more frequent upgrades have a value that exceeds 15% of the purchase price of a new system? I should hope not. Perhaps you should use a different approach to migration -- for example, keeping your hard drive between builds and upgrading that as space demands independantly.

Re: Yacht racing

[r_jensen11]

But, you see, the thing with Yacht racing is that those people actually have lives...

Re:Wait for the next price drop...

[petermgreen]

playing games on a PC (which is an insanely expensive hobby)
It is possible to get a quite reasonable machine that will play most PC games released for a few years acceptablly albiet not at the highest quality settings while paying less than £1000 Games rarely cost more than £50 each.

If you buy 10 games a year (which i'd say is more than enough if you check out reviews or demos first) and spend £1K every two years on a new PC thats a total of 1K a year. In reality you can get away with spending a lot less especially if you don't buy games the instant they come out.

£1000 a year is NOT an insanely expensive hobby. A single clubbing trip could eailly cost you £50, do that on a weekly basis and you are up to £2600 a year!

Where the hell is VIA?

[r_jensen11]

I know they're a smaller player, but they're still out there, especially for price-sensitive and energy conscious consumers. How do VIA chips compare to this lot?

Motherboards!

[Anonymous Coward]

you know, no one ever mentions motherboards in these conversations. telling me the Intel CPUs are inexpensive now means nothing to me when the motherboards cost TWICE as much as what will run comparable AMD CPUs.

Energy Efficiency and Idle

[ooloogi]

For a huge number of computers, the energy efficiency is governed by the idle power draw, and not the loaded power. In that case AMD is way ahead of Intel at the moment on account of the motherboards for AMD chips using much less power at idle. Ecologically speaking though, the energy consumed while a PC is on is dwarfed by the energy consumed in manufacturing and transporting the parts.

Re:Wait for the next price drop...

[darkwhite]

you're playing games on a PC (which is an insanely expensive hobby) That was funny! Sorry, I don't have mod points...