Slashdot Mirror
Which CPU Is Tops in Price/Performance?
mikemuch writes "You can spend 150 bucks or over a thousand on a processor, but how do you know which gives you the most power for your money? It's a little like MPG for CPUs. ExtremeTech's Loyd Case does extensive benchmarking on twenty-three current desktop processor flavors from AMD and Intel. While of course most folks won't make dollar-efficiency the sole basis for their chip decisions, it's interesting to see which CPUs get you, for example, the most frames per second in Far Cry for a dollar." From the article: "Take PC games, for example. The cheapest CPU available may have the best frame rate per dollar ratio. But you still need an adequate frame rate for an optimum gaming experience, and the cheapest CPU may not deliver that. On the other hand, office applications are generally not as sensitive to raw performance, and the lower cost processor may be better. It's all in what you do."
I would say a Pent Pro 200mhz processor, given that most are given away for free now.
Almost all power consumed by processor get converted to heat any idea about Power consumed vs work done?
By narrowing the field to intel and amd, dont we cut the pie awefully thin?
What about IBM, Sun, Motorola, Transmeta, and hell even VIA?
What I'd really like to see is how the "normal" chips stack up in price/performance effeciency vs the "non standard" lineup....
-GenTimJS
What is the metric equivalent of fpsFC/dollar?
Monstar L
i have an msi board with an amd 64 3200 processor...im wondering if a dual core drop-in is available for this board ? does amd offer dual core drop ins for their 64 lines ?
There's no clear-cut price/performance leader.
there, and without cutting it up into pointless pages and appendixes (?!) to generate more ad dollars.
Not exactly surprising, but I wonder how much of that is tied to the OS (f'rinstance, dual core kicks ass on OS X for processor-intensive tasks). Similarly, I wonder how much of it is simply benchmarking the wrong kinds of things. Comparing "office productivity" is mostly useless, as they say in the article, yet it still gets benched. Similarly, graphics, while still relying on the CPU, uses the GPU more and more.
I've found in my own little "tests" that heavy-duty rendering and long-term CPU processes are really where the benchmark tests are at. Fire up something like VirtualDub and compare the time it takes to transcode video files, for instance, or use ffmpegX on Mac OS X. That's where the real CPU tests come into play. Not office and games.
(I'd also be curious to see what happens if you start switching around operating systems. Test to see if an AMD chip and NVIDIA board is better running a Linux flavor compared to Intel, for instance).
Once you've decided on which company to go with (and most of us already have and stick with our choices), you look for the most powerful CPU just before the price break. Come on - this is nothing new. This is how people have been picking CPUs for at least a couple decades. And if powe consumption matters to you - and you're only buying a couple of these things and not hundreds - then maybe you should rethink the whole computer thing focus on affording your top ramen or whatever.
You buy the fastest CPU you can afford at the time. Stay away from the one or two top-of-the line chips unless you have mad money. And know that in another year you could buy twice the performance for the same price.
...to keep up with invention, you're never going to own the fastest processor anyhow.
Do higher frame rates really let you shoot more bad guys or see the enemy coming quicker? If so, how much of an edge are we talking? Is it really worth spending more moolah for that gaming advantage?
Nothing But Coupons - Your no-frills site for online coupons and discou
A used computer is far cheaper than a new computer. As with cars, people seem to want to pay through the nose for something new and shiny. The price then drops the minute they get the box home. For most computing activities these days, a top-of-line machine that is a year or so old does a good job relative to a brand new low-end or mid-range system. And for basic work (web/email/word), a 5 year old machine can be both very affordable and provide acceptable performance.
Two wrongs don't make a right, but three lefts do.
Looking at the charts basically says one thing - go for the cheapest processor if you want the most 'bang' for your buck.
Makes sense seeing how performance increases incrementally as you go up the different proc models. I bet that if you did a study for framerates/cost for graphics cards you would see a similar trend.
I guess bang/buck is not really as important as I would have though.
No surprise there. Now if they could add celerons and semprons to the benchmark, we might see which is really the better value, otherwise they've wasted a lot of their own time and money.
You really need to look at your long term costs ... the power to run your computer in the long run is likely to account for a significant fraction of the overall price, so you should factor that in.
"Who is the Journal of Quantum Physics going to believe?" --Stephen Hawking
It seems it would be more cost effective simply to buy a cheaper processor and upgrade your system more often than it would be to spend on the more expensive processors.
I found it interesting how well the AMD 3000+ did in the benchmarks. On almost every benchmark it had the highest score for price/performance in 3d and gaming related tests. It seems like buying this cpu and putting money in a better video card are the smart choice. I'm basing this on the fact that most gamers go through quicker upgrade cycles anyway.
MidnightBSD: The BSD for Everyone
hmm. forgot the line break in my sig. there should be a preview.
"Who is the Journal of Quantum Physics going to believe?" --Stephen Hawking
I'm surprised that they don't give results for lower-end CPUs (Sempron, Celeron, Pentium M etc.) Those things are rediculously cheap and would probably wipe the floor with expensive Intel&AMD offerings, except maybe in game tests. I guess extremetech is simply oriented towards gamers and thus not interested in workstation CPUs.
I've been extremely satisfied with the Athlon XP processor line. They perform well and are cheap. I actually wish they wouldn't phase them out. Grab them will you can.
ads...
words...
benchmarks...
ads...
conclusion: there is no conclusion.
this article was the longest bit of nothing ive ever read.
dude.
I haven't been deep in numbers for processor performance over the last couple of years. I've found the processing speed to be so fast lately that the software I use or care about runs FAST on most modern processors.
That said, when someone asks me for advice, unless they have some specific high-end gaming requirement, the only advice I offer is don't buy a Celeron!
Other than the poor performing Celeron I suspect most processing bottlenecks today are more from insufficient memory, bad or slow bus architectures, network latencies, and disk I/O bandwidth.
Frustrating to me is the non-sequitur naming of technology, I don't know if it's done intentionally to confuse the buying public. A friend of mine saw the ads for some manufacturer's laptop with Centrino technology (which really isn't about processor anyway), and went to her favorite electronics store and got talked into a laptop with the Celeron (mistakenly remembering the "C" word incorrectly).
I made her take it back and exchange for Centrino.
I'm fairly confident that I'll never have to uprgade again. I'm running on 2ghz, and it's more than adequate for my linux window manager - it runs perfectly adequately. In fact probably 500mhz would have done a suitable job. If you're not a gamer or a windows user, then you shouldn't need that much to run an eye candy laden os. I'm fairly sure the only thing pushing the cpu market is the gaming industry, and the necessity for Microsoft to push a new generation of their products in coming years. CPU speeds have increased usefully over the last 15 years, but I can't come up with any home-computer user applications (apart from gaming) that would need more.
"You know you don't act like a scientist, you're more like a game show host." Dana Barret
This could have been an awesome metric. However there are no older or low-end chips in the whole thing. (!)
I think the results might be really interesting because the low-end chips are really cheap compared to the top-dollar stuff.
For example, I have an AMD Mobile 2600+ that cost like $90 and when clocked at 2.4+ Ghz is damn near close to the performance of my Opteron 250 (which was $800 at the time) at many tasks. I would say the low-end Athlon would have more FPS/$ than the Opteron.
That's what I was hoping to see anyway. Now I'm sad after looking at the list of processors they used.
The ratio of people to cake is too big
They say price/performance higher-is-better... Higher would mean more price for less performance... I don't understand how they're coming up with that metric, maybe they're actually saying performance/price, but they don't know how ratios work, or am I just missing something important?
-Jesse
Nothing says "unprofessional job" like wrinkles in your duct tape.
Given away by whom?
I recently got an old server from my office for free. It was dual capable, so I figured I'd stick another PII 450 in there for fun. What could it cost, like $10?
Called up Dell to make sure that it could handle the 450, and I was offered to buy it from them. Get ready for the price:
$457.
That doesn't include installation or anything. I literally laughed out loud and the guy on the other end said "Yeah, you should probably get it somewhere else."
I went to pricewatch and I got it for $12 with shipping. For that price, I could have bought a whole fleet of PII 450s...maybe that's not a bad idea...
the power to run your computer in the long run is likely to account for a significant fraction of the overall price, so you should factor that in.
That's not the price, that's the cost. The price is whatever you pay to make it yours. The cost is whatever you pay to use it (including making it yours). Subtle difference, I know, but I'm a pedant.
.. is the Athlon 64 3800+
The review calculates price/performance based on the price of the CPUs instead of total system cost. A useless measure, since a CPU on its own cannot do anything useful. It also hides the added system costs for CPUs that consume a lot of power: larger PSU, more cooling and noise reduction measures. And then there are the additional platform costs for CPUs that only work with particular chipsets or expensive motherboards. Never mind the increment to your electricity bill.
What this smells like is yet another bullshit metric invented by the Intel marketing department. One wonders how much these review sites get paid for prostituting themselves.
Both of these processors need $500 of ancillary equipment in order to function. Therefore, a system with processor A gives 100 units for $600, or 0.167 units/$, whereas processor B gives 150 units for $650, or 0.231 units/$. This analysis shows that processor B is better value when speccing out a new system
But what about the case where you're just upgrading your cpu? Well, in that case it's moot to compare the AMD with the Intel processors, as you would need a new motherboard too. But simply dividing the performance by the cost of the cpu is meaningless here, too, because staying with your existing processor ($0) would give you a performance/price ratio of infinity.
Conclusion: you have to calculate your total outlay in order to figure out which cpu is the better value.
Really, if you're going to give a PC to your grandma just for typing papers, a 386 with 4 megs of RAM and a 120 MB HD is more than enough. Stick Windows for Workgroups 3.11 and Word Perfect or Word (if you must) and you have a winner. Just in time for the holidays....
I haven't bought a new CPU for years. In the past I would look at the fastest CPU on the market, and then buy one that runs at about half the speed. It helps to also buy a motherboard that supports faster CPUs.
For example, I bought a new system back in 2000. I think the top of the range Intel chips then were P3s @ 700-800 MHz. I bought two P2-450s for my computer. A few years later I bought two P3-850s, which was the max the motherboard would take. For those four CPUs, I paid less than the price of a single P3-550 back when I was originally shopping around.
Buying top-of-the-range CPUs is just a waste of money. Gamers are the biggest fools of the bunch with their obsession to have the latest and greatest.
time-testing-hardware/common-sense -> Infinity
The Quad G5 I just stole in a daring heist. Ninjas nearly caught be, but they're not sweet enough for me.
... and then they built the supercollider.
Wouldn't it be better to have a Top Performance/Price? Just wondering...
The answer IS 42.
I work in a high performance computing center, and we just recently acquired a new cluster. One of the major items that we looked at was the amount of heat that is generated by the systems, as cooling systems for large amounts of equipment can be quite costly. We went with AMD dual core systems because we were able to load up with twice the number of systems and cores (thus a 4X overall improvement in number of processors) for a heat load that was actually less than the old system we had that was running Intel Xeon processors.
Shifting to a DC powergrid helped a lot too.
I'll bet the best price/performance chip in years was the AMD 2500+. I bought mine a few years ago for about $60.00 and clocked it to a 3200+. I saved hundreds of dollars then for the same performance, and today I am still doing great with it. Best chip ever.
If I'm willing to run Windows 3.1 or Mandrake 8.1, a 200 MHz system will be fine. On the other hand, if I'm using the system for the office and I can save an hour a week by buying a system, my payback will be measured in weeks not months or years. Like tfa says, it really depends on what you're doing.
I've spent a lot of time thinking about what I can use my pile of obsolete equipment for. Usually the answer is that I'm much better off forgetting it and going with something more modern. It would be interesting to see a similar article that takes all the costs into account; capital, wages, electricity, maintenance. Basically, once you have to pay wages, the capital cost of a computer is peanuts.
i have an AMD athlonXP 1600+ (1.4Gig MHz) CPU and i just ordered a new motherboard to slide under it that has better support for different BUS speeds (266/333/400).
:^)
RAM @ PC1600, PC2100, PC2700, PC3200
IDE Controller @ ATA33/ATA66/ATA100/ATA133
i know it is not new, but for what i do i dont need bleeding edge. but if i want to slap a new CPU in then no problem, faster RAM no problem, and salvaging used components is gravy too
Politics is Treachery, Religion is Brainwashing
There's no clear-cut price/performance leader, unless you simply believe that the lowest-cost CPU will offer the most bang for the buck. We were impressed with how the Athlon 64 3800+ placed, generally offering a slightly better position on the overall curve than most other CPUs. On the other hand, the Athlon 64 3700+, one notch below the 3700+ in price, looks to be the odd duck, and we'd recommend you avoid this one if possible.
On the Intel front, the Pentium 4 660 does seem to have a slightly better position on most curves than the other Intel CPUs. At under $400 for 3.60GHz, it's certainly better positioned than the pricier 670. Note that the weakest link in Intel's lineup is the very pricey 3.73GHz Pentium 4 Extreme Edition.
Aside from the HPC circles, I've never really seen a price/performance analysis like this before. I was even more surprised at the results. The P4 3.06 GHz processor came out in November of 2002, the Athlon 3000+ came out in February of 2003 and the P4 3.0 GHz with the 800MHz FSB came out in April of 2003. And these were the winners in terms of price performance with not much more raw performance in the newer and much more expensive chips.
So, what does this say? To me, it says that SMP and mulitcore is here to stay. We've known this for quite some time for servers, higher end workstations, and again in HPC environments. Its cool to see this trickling down to the more basic end user machines. Although the multicore processors are still new, they are really showing promise. At this time, it looks like the PowerPC dual cores are the best in terms of performance and power handling (they are the only ones I know of that can dynamically power down or reduce power to one core at a time), then the Opteron dual core for price performance, and the Intel x86 dualcore/hyperthreading last. Although I don't have the common slashdot attitude towards Intel, I do believe that they have really lost tons of ground over the past couple of years, and their current offerings in the multicore and hyperthreading hack are not much more than blanks to be filled in the current marketers' scripts.
SPARCs are supposed to be multicore soon, along with Itaniums, and I guess everything else. AMD has the current lead in core/processor to memory path, but that still needs improvement.
What I want to see, especially for the "business" or "general purpose" type of machine (read NOT games) or even in the server market is to have more cores, maybe 4 to 8 or so, and have them dynamically come to life when needed, and sleep when not needed. Unless your machine is loaded with adware, spyware, and flash advertisements, when you look at desktop CPU utilization, its very bursty. Usually its at 10% or less of use, and at times it can go much beyond that, and the processor and memory systems should be designed to accommodate this kind of use.
Subtle difference, I know, but I'm a pedant.
no kidding.
That was the worst conclusion ever concieved.
-Rick
"Most people in the U.S. wouldn't know they live in a tyrannical state if it walked up and grabbed their junk." - MyFirs
Of course, as the article states, gauging a chips' performance isn't a simple matter since everyone has different performance metrics based on the software they use. How many John Q Public will use 3DS Max and encode MPEG4 and give a rats ass about miniscule performance differences anyway? In my expereince, non-power users who do alot of CPU-intensive stuff just leave their computer overnight when they're converting the birthday DV into an MPEG or whatnot - for them, "slow" usually equates to "Swapping between apps takes an age cos I don't have nearly enough RAM in my bargain basement beige box" rather than a lack of high floating point scores.
And is overclocking allowed? And what memory/chipsets/I/O are we able to use? Do power bills count towards overall system cost? All of these points are glossed over or omitted entirely from the article.
That said, since the article is talking about general purpose computing, I'm gonna take a wild guess and say one of the AMD 64bit Semprons. Great performance across the board at a very low price (some very cheap and decent mobo's out there too). For more server/workstation-like workloads, I'd go for something like an overclocked Athlon X2 3800 (I'm informed almost all will overclock from 2.0GHz to 2.4GHz without even raising the core voltage - not tried it myself though as I'm not into overclocking - which nets you the power of an X2 4600 for ~£220 less based on todays UK prices).
All in all, the article was just another excuse to say "Woooweee! We've got lots of processors, let's use all our benchmarks du jour as an excuse for an article". Sigh.
Moderation Total: -1 Troll, +3 Goat
If one's work is important, one should not accept the possibility of undetected random bit-flipping.
Many 'desktop class' CPUs these days will not support "egistered" memory. Heck, even ECC is "an option".
In Soviet Washington the swamp drains you.
Step 1: Rescue 386 with Windows from Dumpster Step 2: Give to relative Step 3: Upgrade phone system to handle influx of calls Step 4: Commit suicide after seeing the horror of it all
... and then they built the supercollider.
A fantastic overclocker, an even more fantastic price! Don't let the M(obile) suffix fool you, you can put it in a desktop PC without issue, and they run at a lower voltage than their desktop counterparts.
The Athlon 64 X2 3800 may be a close second. It is an absolutely great processor(s) for the price.
No worries, it's an evolving language, not a dead one. :-)
"Who is the Journal of Quantum Physics going to believe?" --Stephen Hawking
Sure you can solve little man's syndrome by buying an "efficient" powerhouse Processor, but what good is it when you wont see any difference 99% of the time and you can save $400.
CS: It is all sink or swim...oh and did I mention there are sharks in that water?
I agree. I've got a 6600GT in my gaming rig, which on benchmarks comes in somewhere around one-half the performance of a 7800-class card, but it runs Quake4 just fine and costs around a third of the price of a 7800.
A 3000+ might be cheap, but factor in other costs and it might not be the fastest (indeed a quick check suggests the 3500+ or 3800+ can win out).
It would be nice to see these calculations redone to take the real world into account.
... means Miles Per Gallon... the analog to kilometers per liters in a certain more sensible system of units and measures.
Despite what EULAs say, most software is sold, not licensed.
I see 2 things that could seriously improve this article.
First the graph of the price performance is not real informative. They had all the data they just needed to add it to the graph. It really needs to be a scatter plot of performance and price/performance. That way you may be able to target a region of the graph based on your application.
The other thing is that when I buy a cpu I may actually spend 20% more on the cpu for a 10% cpu boost. The reason is that the faster cpu can actually utilize the gpu 10% more, and the memory, and the motherboard, and every other component in the machine. Bottlenecks will of course reduce overall % gain but paying an extra $30 on a $150 cpu is actually cheaper in the long run if you can get a 5% gain on $800 in parts. The way they present the numbers it looks like you need a %10 gain in performance to justify a %10 gain in price.
It's spelled ATHLON, not ATHALON.
It's a little like MPG for CPUs.
Close, but not quite. The real analogy is Miles per Gallon per Pound. SUVs get relatively poor MPGs, but they're performing much more work.
Just noticed that the very last page of TFA where they are benchmarking Splinter Cell 3, all of the benchmark results for the High test at capped at exactly 47 FPS (in the raw data). I wonder if that was the motherboard, ram latency, video card, bug, etc... I really wonder if the motherboards / chipsets have a lot of play in how well the CPU's perform overall. I have had quite a few computers where the only part to get changed is the motherboard, but the computer seems to run multitudes of times faster. I have seen many many crap chipsets drag down otherwise good CPU's. Also from what I have been able to tell the P-III 733 - 1.13(?), or AMD Athlon at similar clock, seems to be the fastest processor the average desktop user really needs to have. They can play movies, browse the web, type documents, chat with friends, burn cd's, rip MP3's, with clock cycles to spare. It doesn't hurt any that these processors and motherboards are being discarded in many corporate environments at this point in time, and can be picked up for very cheap if not free. Also linux runs absolutely wonderful on P-III generation chips. It supports everything on the board, generally all of the bugs have been worked out, and it just makes for a very stable platform (and cheap) that you could give to your grandma, kids, cousins, etc.
While it's interesting to know price/perf for the CPU per se, it would be a much more interesting metric the TCO/Perf. At least, adding the CPU price to the motherboard and other components and then dividing that amount by the raw power of the configuration would be a much better metric than the CPU alone. Additionally, factoring other costs such as energy and cooling might give a more complete metric.
Why would I need a different OS for testing on any of those chips? .. they all run Linux 2.6.10 just fine ..
-GenTimJS
You will do much better using a 800 -1.2 ish machine and Win2k and word/office 97. ANd the price these days is comparable to a 386.
Food not Bombs is a nice platitude but it breaks down when you notice that the Bombees are usually well fed
Read Firing Squad's article about F.E.A.R. game with the recent CPUs (not just the newest and expensive ones). It said video cards are the big factors in CPU performance for this game. I was a bit surprised by that. I really need to upgrade my video card soon (probably a NVIDIA 6800 for its decent price and can't use it again if I upgrade my motherboard next year).
My game is sometimes choppy at 1152x864 resolution with most graphic options to the maximum. I have an Athlon 64 3200+ with 1 GB of RAM and ATI Radeon 9800 Pro AIW (128 MB). I get 20 FPS average (10 minimum) according to its benchmark test.
Ant(Dude) @ Quality Foraged Links (AQFL.net) & The Ant Farm (antfarm.ma.cx / antfarm.home.dhs.org).
> I had thought it a strange summary actually
It's customary for review sites not to take sides. How else will they continue to get free product to test?
Is it at all suspect that they were using ram with quite a bit better CAS values in the AMD setup?
AMD:
2 x 512MB Corsair XMS 3200XL (CAS 2-2-2-5)
Intel:
2 x 512MB Corsair XMS2 Pro (CAS 3-3-3-8)
Personally I only have AMD systems at home so I'm not sure what the limitations are in having lower latency ram in the intel system but I still think that would account for some performance differences.
I think a more useful chart would set price points and then find the best proc within those points.
What is the best proc for $100? $200? $500? $1000?
If he'd spent 15 minutes thinking about this before he started, he would have realised that lower-priced CPUs would rate higher. Reminds me of how my grandpa used to tell me that the WW2 PT boat had more firepower than a battleship based on firepower to displacement ratio.
I'd rather you do it wrong, than for me to have to do it at all.
Oops, I meant to say 1024x768 resolution. I forgot I turned the resolution down to gain a few more precious FPS. :) I like eyecandy. So pretty. ;)
Ant(Dude) @ Quality Foraged Links (AQFL.net) & The Ant Farm (antfarm.ma.cx / antfarm.home.dhs.org).
This article concluded that the AMD64 3700 was an "odd duck" and maybe try to avoid this one. It usually scored lower than the 3800. The biggest difference I saw between these two was that the 3700+ had twice the L2 cache of the 3800 (1MB vs 512k). It was lower in absolute performance for most of the Games benchmarks. But, I noticed in the appendix that for Farcry, the 3700 scored better fps than the 3800!
Can someone explain why doubling the cache would hurt many performance benchmarks?
It is opposite from my own experience: I have two AMD 2600+'s at home. One is a Barton (which has double the L2 cache), and it is clearly a better gaming machine. (I have tested them against eachother extensively)
What is the deal here?
I agree. Their CPU list is all well and good if you plan on building a new system, but worthless if you are upgrading. I will probably change from my XP1700 tbred to the AMD Mobile 2600+. That way, I can keep using my current memory and video card, and as a bonus, will use even less power as well.
"He's lost in a 'floyd hole"
Also, since I always like to better myself, I went and looked up the definitions of price and cost, and it turned out both of my usages were correct. I invite you to definition 4 ofn ary&va=price&x=0&y=0
http://www.m-w.com/cgi-bin/dictionary?book=Dictio
"Who is the Journal of Quantum Physics going to believe?" --Stephen Hawking
I thought we could already play .MPG on almost all modern CPUs!
"There's no clear-cut price/performance leader, unless you simply believe that the lowest-cost CPU will offer the most bang for the buck. We were impressed with how the Athlon 64 3800+ placed, generally offering a slightly better position on the overall curve than most other CPUs. On the other hand, the Athlon 64 3700+, one notch below the 3700+ in price, looks to be the odd duck, and we'd recommend you avoid this one if possible."
Its like they can't figure out their own results..
And you spend 100 times the difference in cooling gear :-)
It's better to be the foot on the boot than the face on the pavement. ~~ tkx Kadin2048
This would be a wonderful, wonderful article if they had posted a scattergraph, that shows price on one axis, and performance on the other. As is, the only reason the given processors won is because they didn't compare cheaper ones. A $30 C3 running at 1GHz will still beat out, in terms of price/performance, a $150 2-3GHz chip. Cheaper/slower always gives better price/performance if you just divide. It changes when you calculate Total System Cost (including hard drive, case, etc.). Scattergraphs let you do post-processing mentally, and figure out "Total system cost/performance" or "Productivity gain/cost change" or "Fastest CPU per price point" or "Cheapest CPU per performance.
They did 95% of the work for a beautiful article, with 0% of the rewards. Suck.
Agreed. I have an XP-M 2600+ too, currently oc'd on stock cooling to 2.2 Ghz 200 fsb (equivilent to 3200+), and works at 2.4 Ghz fsb (3400+), though I want to get a copper hsf before I run it that high. Btw, for all those not familiar, the XP-M 2600+ runs at 2Ghz 166Hz fsb stock. Really wished they at least would have included this chip on the benchmark.
Is Extreme Tech for spending the time making fancy graphs of things we already knew.
A: The eight hundred dollar gap between the bottom and the top in the same class of processors only buy's you 25 percent on a good day, with the wind at your back.
B: For games, your video card negates much of the speed difference of the processors.
C: Manufacturers price competatively with respect to others products.
Platform advocacy is like choosing a favorite severely developmentally disabled child.
I'm really looking forward to truly scaleable multiprocessing. So the price:performance ratios can really be used to decide what to buy, rather than mostly support trivia or occasional insights into vendors' economies. My bet is on FPGA/DSP banks processing dataflow.
--
make install -not war
When you buy your computer (or motherboard), if you want to keep it useful for a while - buy the one with the socket that you think will be supported in the future. Otherwise, you have limited upgradability: Videocard/Memory only.
Specifically: AMD has said something along the lines of Socket 939 will support x2 chips. Socket 754 will not. Thus, You will be able to buy a low end 939 chip now, and a new one in two years. You will only be able to purchase whatever socket 754 chip they manufacture at the end of the year, as they will likely be phased out. Intel is the same deal: Whatever you buy now, they will change the socket format in a year and stop producing it.
Thus the wise advice would be to buy a socket 939 computer with a slower chip, and upgrade it in a year or two with a dual core.
Sigh, too bad none of the major manufacturers seem to make $700/$800 computers with Socket 939. (New Egg has one for ~$600, but it might as well be white box). My customers still need support if I get hit by a bus; you only get that by purchasing corporate...
My personal favorite is still the vast.
Exam 4/C again. Maybe I'll do better this time.
Just take them and the other crap out;)
The 3700+ runs at 2.2GHz, the 3800+ at 2.4GHz.
s px
http://www.amdcompare.com/us-en/desktop/Default.a
More L2 cache will not hurt performance. It's just that some of their benchmarks see much greater benefit than others when the L2 cache is increased. In the Farcry example, it happened to be enough of a benefit to overshadow the 200MHz difference.
4 : the cost at which something is obtained
Is that not what I said? Price is the price of obtaining the item, but does not include the costs incurred with its use. The purchase price is a subset of the total cost of ownership. A CPU may cost significantly more than its price, but it's price stays the same.
Anyone with any know how whatsoever buys their CPUs this way.
I would further state that this article is next to useless. It offers no insight, no numbers, nothing that makes a point. Given the graphs, I'd say buy the cheapest CPU, as there is nothing else to base your choice off of. Now, had they included $/task graphs, that would have been interesting, and maybe yielded a different set of daten a fps/a. Ev$ graph would have been neat, with axes like fps vs $ at several resolutions, which would create a discrete set of data so you could say for 800x600, CPU x gives best value, but @ 1280x1024, CPU x fails, and CPU y is best.
Basically, the tests presented are meaningless and add to the noise.
The cesspool just got a check and balance.
Obviously, you don't have a Pentium 4.
Weaselmancer
rediculous.
What is the cost at which a year of computing is obtained? Answer: the price. Does that price include what you pay to power your computer for that year? By definition.
"Who is the Journal of Quantum Physics going to believe?" --Stephen Hawking
replying have said these metrics don't have a lot of value to them (in one way or another)
I disagree
But I'm a rendering geek
I was VERY happy to see the POVRAY price/performance (technically performance/price) breakdown... and will definitely be getting an Athlon 64 XP when I build my system... the 3000+ model if these numbers are still valid when I get the loot
I am disrespectful to dirt! Can you see that I am serious?!
Really silly artical, they don't know about overclocking Semprons:n -3100e.html
Because the latest e-core process runs so cool Sempron 3100 can be overclocked to the speed of a Athlon64 4000+ without needing any special cooling:
http://www.xbitlabs.com/articles/cpu/print/sempro
The cheapest processor clearly won, notwithstanding the hedging of the conclusion. It might have been even more embarassing for the chip-makers if Semprons and Celerons, with prices well under $100, had been in the bake-off.
I think for most folks casual home use, an older CPU is fine, just throw lots more RAM at it and upgrade the video card. Unil a few months ago that's all I was using was a PP200, it worked fine with modern linux, at least, for my purposes it did. I'd still be using it if it wasn't for extremely crappy local grid power, fried two machines so I just snagged a cheap barebones system and swapped my drives out.
They left out overclockability
Yes, I know not everyone overclocks, but with chips that clock as easy as the A64's, you almost have to consider it. For example, the 3200+ came out as second place for performance/$ in every test, beat only by the 3000+. However, my 3200+ is currently running stable and cool at 2.6 Ghz and has a 512K cache.
This puts it between the 4000+ and the FX-55. And my OC is very typical. As a matter of fact, it is low. Just about any venice 3200+ will hit that speed easily, and many will reach 2.7+Ghz. This puts the peformance/$ WAY up there.
The 3000+ would probably also beat it. For some reason when I purchased my CPU, I forgot that I could run my RAM on dividers, so I ordered the 3200+ for its higher multiplier, which is completely useless with any modern motherboard, since RAM speed and CPU speed are independant of each other.
So basically, I am saying get a 3000+, since it is the best chip out there for for performance/$, and almost matches the top of the heap for raw performance as well.
the electricity has a price. the processor has a price. the combined amount being the cost of using the processor for a year... i see that you cannot be convinced, and neither can i, so let's call the whole thing off :)
You can't compute price/performance just based on an incremental cost. A CPU is chained to a $1000-2000 box full of other components.
If you re-work these numbers assuming about a $1500 system cost, the curves level out quite a lot. At $2500 system cost, some of them turn the other way (e.g., the Pentium D 830 has a higher rating than the Pentium D 820).
A (CPU + system price) / performance metric would be much better to see. You know that Intel and AMD have people crunching numbers like this -- it's the only way they could justify the prices for high-end processors.
One is DDR, the other DDR2. The pentium 4's 800mhz bus is shared and since the memory controller is off the CPU, the throughput is much lower than on the athlon 64. And the performance difference between ddr ram at cas 2 and cas 3 is small, 5-7% at most at 1T, and we're comparing apples and oranges here.
To elaborate on what the other AC said, you're trying to mate an AGP graphics card with a PCI-express motherboard. You're better off getting a PCI-express video card, AGP is fading fast anyways.
Also you might consider the separate tuner thing, like someone else mentioned. You can upgrade the main card down the road for performance, but avoid paying the "tv tuner" premium twice by sticking with your old card.
It was an interesting article, with tons of good data (and, to their credit, they include the raw data without comment in the appendix (ok, it would have been a lot nicer if they included it in a spreadsheet-friendly format, but ...))
Unfortunately, you can't do anything with a bare processor. You need a system to plug it in to, and that system costs money.
If you assume that the disk/video/case/fans/power-supply/motherboard/OS package would cost $600 or so, then that would have the effect of adding $600 to the cost of each processor for a system that can do actual work. For example, in the 3Ds Max 7 Rendering Test, their calculated best performer was the Intel Pentium 4 630 or Intel Pentium D 820 -- relatively cheap processors.
But, adding the $600 to the cost makes the best performer the Athlon 64 X2 3800 (the cheapest of the Athlon dual proc chips.) The other X2 chips round comprise four of the next five places as well.
I think that adding a minimal system cost makes for a far more useful comparison -- and it does show the value of the new dual-proc systems. Not too surprisingly, the Athlon 64 FX chips still the worst price-performance solution -- they're just too expensive for what you get.
Thad Beier
I love Mondays. On a Monday, anything is possible.
Also, if you plan to upgrade later you are essentially buying 2 processors...and for that price you might have been able to buy the better processor in the first place and get more use out of it.
Bottom line, planning to upgrade is not always worth it. Most of us probably love to tinker and will take the upgrade route anyway...even though it might not be the most cost effective.
Since everyone seems so eager to do benchmarking, what I'd like to see in the price-performance area is:
Cheapest system that can do X tests, where X is some fixed performace metric.
For example, what is the cheapest mobo/CPU/graphics/memory combo that gets me 40fps in 1600x1200 Doom3? Especially for game benchmarks, this would be interesting. And super-cool would be some sort of a vendor site with a benchmarking database where you can input your desired performance on several applications that matter to you, and then the database would spit out the cheapest combination of parts that would meet your performance requirements. AMD (especially) should be pushing for this, because it would make very vivid to people the fact that they are ahead right now on price/performance.
As far as I can tell they used the 32-bit version of Windows XP on both 32-bit Intel processors and 64-bit AMD processors. So they're not taking advantage of the performance gains of 64-bit code on the x64 processors.
They also shortchanged the Intel side a bit by not really taking advantage of the dual-core processors in the benchmarks, as mentioned in the article.
It would be interesting to see the results if they could recompile all the programs to use the full capabilities of the x64 CPUs... If they were benchmarking open-source programs it would be easy O:-) I'd love to see a big benchmark comparing things like GCC, the Gimp, OpenOffice running natively on Intel and AMD64 processors...
My bicyles
This has probably already been mentioned, but... I like how the graph titles say "price / performance" while the graphs themselves display performance / price (example at framerate/$).
Why no socket 939 Opteron 144 in the test?
The $130 s939 Opteron 144 would kill all competition if it was in the test, it even overclocks to over 3GHz with standard cooling and default vcore, performs like a FX-57.Read more here http://eclipseoc.com/index.php?id=3,25,0,0,1,0
After going through all the graphs and making up my own decision as to what seems like the best price/performance ratio CPU, the authors determine that they can't determine anything!!
Final Thoughts: No Definitive Winner
However their results seem to indicate that either the P4 630 or the Athlon64 3000+ offer the best performance at this point in time in the CPU market. I actually suspected this since higher model CPU's are way overpriced. So since the choice comes down to the P4 630 vs. the Athlon64 3000+, the only question remaining is, which one uses more power when idle and at 100% load? In any case the winner seems to be the AMD since even if the P4 has EMT64, I've heard it's a rather poor implementation of x86_64.
"So it appears that AMD and Intel are racing to see who can offer the most confusing product line."
So very true. Acronyms and strange naming conventions are the reason I try to stay away from hardware. Not that it's much better when programming ( OOP, PHP, etc ), but it's better than some of the strange things you see when browsing a hardware catalog.
God is dead -- Nietzsche
Nietzsche is dead -- God
Zombie Nietzsche lives! -- Zombie Nietzsche
... you should probably consider power costs as well.
An Athlon 64 idles at around 10W, and under full load pulls (according to the rated Max Thermal Dissipation Power on my 3400+) 62W.
The Pentium 4's idle at ~50-80, so I've heard, and under full load pull 120-150.
The P4 is dead. Both Intel and AMD have viable products on the market (Pentium M / Athlon 64), but the P4 isn't one of them.
Buying top-of-the-range CPUs is just a waste of money. Gamers are the biggest fools of the bunch with their obsession to have the latest and greatest.
Just like people who buy luxury cars or anything else that exchanges cost for a little additional benifit.
But fool seems a little harsh.
Quack, quack.
It's true that the ancilliaries count, but you are not considering the Net Future Value (NFV) of your advice. Not only do you have to purchase everyting in the *replace* case, but the extra cost now, is extra cost. Also, some componants will not speed up significantly: Disks won's spin 2x as fast, and even if they do, they won't saturate a SATA2 bus. IMHO, the last time this oppourtunity showed was early in the Athlon (700MHz - 1.4GHz boards), and in the middle of the Athlon (1.4GHz - 2.8Ghz boards)
Processor-bump Cost-of-bump 2Yr-NFV-of-cost
a64 3800 $180 $215
a64 3800x2 $200 $240
a64 4800x2 $650 $780
2.5 years is the typical life of a business pc: Projecting 2 years out we can expect the low end chip to be $150 and an A64 6000, or 5500x2. Of course, we omit the disk issue as you will fill whatever you have with music and video so the question is moot. You will have to upgrade the memory, but rather than purchasing a new box for $600, you can buy it with your left-over $30 from not pruchasing the a64/3800. (Or the left-over $630 from not getting the 4800)
At any rate, we can expect that a socket 939 machine can be maintained for twice the normal lifespan of the typical PC, simply because the socket will be maintained for a while. The Disk speeding up issue is irrellivant because it is unlikely to saturate a SATA 150 interface, and definately won't fill a SATA2 interface. Memory will speed up, but you will still be able to get more DDR400 for a lower price & dump it in.
That said, for high-power applications, you should avoid the latest Pentium 4/D/Xeon chips in favor of the latest Athlon64/Opterons, but it's still only a few percent of the TCO.
It's odd that they've not included any Semprons or Celerons in there...
When I buy a new CPU, I use a slightly different metric. Bang-for-buck is important to me, but so is raw performance. So I multiply the two together: (units of performance) * (units of performance per dollar) = (perf units ^ 2 / dollars). This tends to yield a maximum at a couple of speed grades below the highest available, which is the point at which the prices really start to take off.
Downmodding is the refuge of the weak. Don't downmod, make a better argument!
Why not the Jargon File?
As energy prices climb the effect will only become more pronounced. Selecting a processor which is cheaper and faster but also happens to consume as much power as a small city is NOT a cost effective solution. Why blind ourselves to this?
Frames per eurosecond.
...mans' libido could be measured like we do CPU.
'in '92 Joe was 200 () as in free'
'that compared with 3600 () in 2005'
'I say all CPUs are not equal'
Toms Hardware ran an article recently that showed that the Pentium-M rated at ~ 2.1 Ghz could easily be overclocked to at least 2.6Ghz with the ASUS conversion adapter - and thus giving the most processing power of any desktop processor.
.. http://www.techpowerup.com/reviews/ASUS/CT-479
Curious that they'd leave it out of this lineup..
Unable to find the toms article again this late night, but it's the one reviewed here as well
Your analysis doesn't make much since either when figuring cpu value.
100 units / $600 = 0.667 units/$
150 units / $900 = 0.667 units/$
If I add my $1000 monitor to the equation that shouldn't change the proportion of cpu value between the two, right? But it does.
100 units / $1600 = 0.063 units/$
150 units / $1900 = 0.079 units/$
Why would the CPU value increase more for the 150 unit machine than the 100 unit machine just by adding a monitor?
Thanks, because I don't know what I'm talking about and never claimed I did...
Thanks, because I don't know what I'm talking about and never claimed I did...
I'd like to see the comparison with power consumption (24x365) factored in. I've got a Cobalt Qube2 as my server (250MHz mips), but I'd like to upgrade for running more CPU intensive stuff. The thing is, I haven't found anything that I'd like to pay to run 24x365. My cobalt's power supply is only rated 36W continuous. Just the CPU on intel stuff seems to run double that. I guess the stuff that Apple is waiting for might be useful, or the Mini-ITX, but everyone claims the performance on the ITX stuff is crap.
Can anyone suggest a good replacement for a Qube2? (preferably moving to a 1U rack format (not too deep either
Awesome furniture, accessories and cabinetry in Santa Rosa, CA: http://humanity-home.com/
There are a lot of players for the office range, not on the field. I would have loved a comparison between those :
...
Semperons, Celerons, Transmeta, Via, Geode
It's not like they wouldn't make any sense. It's just a different league.
I got a little job for you guys,
10 benchmark
20 choose other CPU's
30 goto 10
40 don't worry be happy
but down here those "silent air cooling dothans" are as expensive as the processor you are trying to cool with them to overclock.
It's better to be the foot on the boot than the face on the pavement. ~~ tkx Kadin2048
What you really need to consider is how frustrated you become and how much time over the lifespan of the processor you waste waiting for the slower processor to do its work vs. the faster one. By this metric, buy the fastest thing you can afford at the time (keeping in mind that you need a new video card, the babies are hungry, the rent is due, etc.). You'll find that's what you do anyway. Do we need an article to state the obvious? Yes, yes we do.
That's it. I just thought someone ought to point out the obvious. Once again.
To do list for Windows
Shouldn't the real measure of price and performance be what the marginal utility of buying one processor over another is? Simply, is paying an additional $X for a performance increase worth it? I don't understand why people get excited over a 20 second increase in launching photoshop when it will cost them $2000 more. Or getting 20 more frames per second. It doesn't really matter...and it doesn't make sense to me.
Most people want to maximize how well their computer runs and spend a reasonable amount to make their computer experience worthwhile. This isn't rocket science.
Where it gets confusing is that this is an individual preference. So, if you are used to working on a slow computer, all of the newer computers are going to appear much faster to you. Imagine going from a PII 300 computer to an Athlon 64 FX 55. Do you really wanna shell out the extra money for a FX-57 which is 10% faster but costs much more?
http://en.wikipedia.org/wiki/Marginal_utility
... heavy-duty rendering and long-term CPU processes are really where the benchmark tests are at.
...) then you can compare each of the test sets with features, letting you know how much the feature contributes to the type of work that is critical path for your applications.
If you have extensive data sets of tests, together with a variety of major architectural features (GHz, FSB speed, L2 Cache, RAM size, disk speeds,
For example, compilations often cut sharply into programming productivity, and so are very expensive in wasting user time. Typically, they depend as much on good disk I/O as they do CPU speed. But headline CPU speed (or maybe, is usually the attribute chosen.
Deconstructing speeds can give you better insight into getting the best performance for your application mix. By homogenizing the non-CPU info, this test makes it impossible to determine critical paths besides the CPU.
If you DID get such a set of runs, the comparison is OLS regression... best done with time as the objective, and features converted into some unit that approximates, "how long will a chunk of productive work be tied up in this part of the chain?"
"Inquiring Minds Want to Know!"
Title should be:
"Which CPU is bottoms for price/performance"
or
"Which CPU is tops for performance/price"
Additionally, the second subtitle should be:
from the buck-for-your-zing department
I'll be your candy shop of infinite deliciousity if you'll be my discotheque of endless rump-shaking.
As previous stated, you must consider total cost of ownership. Pentium M desktops are not common yet but if you check the benchmark sites, you will see that the 2.1GHz Pentium M out performs the P4 running at 3.0GHz. And at 1/3 the power consumption. Intel Engineers knew the P4 was a dog when they ran the first simulation, but upper management didn't want to wait 2 months and spend millions to rework the core, so they used marketing to push the product and their marketing department should all get raises, because it worked. They used the lame excuse to techies that the pipeline is designed to work better at higher speeds, so the clock speed race had begun. When the PM are available I'm scrapping my P4, getting more perfomance, lower electric bills, lower medical bills (later in life from EMI)and leaving the dog behind. As for AMD, I am a big fan of the underdog with the superior product, AMD64, but it still is a little too pricey and sucks a lot of power also. It would be my fallback if the Pentium M does not make it to market soon. The Notebook manufactures want all of the PM production and are fighting to keep it. We will see.
This comparison seems to show things a bit differently. Gee, why is that?
? i=2249&p=12
Do you suppose Intel is sucking dirt-star and needs some numbers *bad* ?
Nah, money just can't buy integrity these days.
http://www.anandtech.com/cpuchipsets/showdoc.aspx
Join the Slashcott! Feb 10 thru Feb 17!
Depends where you live. 9 months of the year my desktop is busy keeping my room warm, and I don't run it much in summer because of the heat.
I quit!
Cooling gear (fanless, especially) is Expensive. Often, Good (as in really safe) cooling gear is more expensive than the price difference saved in OCing.
It's better to be the foot on the boot than the face on the pavement. ~~ tkx Kadin2048
A typical desktop machine migth consume something like 150W (I'm not talking a tricked-out gamer-machine here, those guys don't care about cost anyway). It migth be used 3 years long, and perhaps 1000 hours a year (that's like 3 hours a day, obviously some machines are used a lot more or a lot less)
That's in the ballpark of 500 Kwh over the lifetime of the machine. If I got a more power-saving machine that got away with 100W, I'd be down to 350Kwh. I'd have saved like $20 at my current power-prices.
That probably ain't enough to make much difference.
If the machine is used for a larger part of the day, like in the extreme case a computer that is on 24/7, the numbers change. Saving $200 on power is more of an argument than saving $20.