Intel's Core i7-980X Six-Core Benchmarked

Ninjakicks writes "Although they won't hit store shelves for a few more weeks, today Intel has officially unveiled the new Core i7-980X Extreme processor. The Core i7-980X Extreme is based on Intel's 32nm Gulftown core, derived from their Nehalem architecture and sports six execution cores. The chip runs at a 3.33GHz clock frequency, that can jump up to 3.6GHz in Intel's Turbo Boost mode. This processor has a max TDP of 130W, which amazingly is the same as previous generation Core i7 quad-core CPUs. Of course, it's crazy fast too. Some may say that the majority of applications can't truly take advantage of the resources afforded by a six-core chip capable of processing up to 12 threads. However, the fact remains there are plenty of multi-threaded usage models and applications where the power of a CPU like this can be put to very good use."

Comments

Nice, but who has $1000 to pay on a CPU?

[Targon]

I know there are SOME people out there who have $1000 to spend on just a CPU, but until these come down a long way in terms of price, it is WAY out of my price range.

Re:Nice, but who has $1000 to pay on a CPU?

[Vectormatic]

Intel always prices their High end around $1000, never mind the fact that price/performance on those chips is horrible.

It is the price you pay for getting the bleeding edge, AMD also has some halo models, but because they cant beat intel in performance, they cant afford to charge $1000 for their high end chips.

As for this comming down, AMD is slated to release six-core phenoms to the desktop before summer iirc, it wont have the raw performance of this thing, but 6 cores for under 200 bucks sounds nice doesnt it?

Re:Nice, but who has $1000 to pay on a CPU?

[rotide]

All new bleeding edge CPUs are expensive. That's not the point of the article/submission. The point here is that a very fast 6 core, 12 thread consumer level processor is now on the market.

Price will come down in due time.

Re:Nice, but who has $1000 to pay on a CPU?

[eldavojohn]

I know there are SOME people out there who have $1000 to spend on just a CPU, but until these come down a long way in terms of price, it is WAY out of my price range.

Companies? Rendering farms? At this price, I'd imagine they're not really for the average consumer but more so for companies that can consider such a purchase an asset.

That said, you do realize that the i7-975 quad core that they compared it to is also nigh $1000, right? I think showing that the same price will buy you an entirely different beast signals that quad cores are complete. The current quad cores price will come down but why would you make a more expensive quad core at Intel? The specs here show it cannot stand up to the new six core platform.

All these prices will come down, of course. So it's fun to look forward to what I'll be using in two years (I just bought a low range quad core for $140 a week ago, almost right in time for this).

And also, who strayed from the duo- quad- naming methodology?! Are you insane!? Do you have any idea the marketing power that a sexa core chip could have?

Re:Nice, but who has $1000 to pay on a CPU?

[Pojut]

It is the price you pay for getting the bleeding edge, AMD also has some halo models, but because they cant beat intel in performance, they cant afford to charge $1000 for their high end chips.

AMDs current flagship costs $195 and is still a heck of a performer. I'll stick with AMD for now.

lol, anyome remember the horribly overpriced Athlon 64 FX-55?

Re:Nice, but who has $1000 to pay on a CPU?

[confused one]

People should have modded you differently, not "Funny". I don't believe I've ever spent more than $150 on a processor for my personal use, going all the way back to the 386sx I bought. Now, work is a different story...

Re:Nice, but who has $1000 to pay on a CPU?

[ircmaxell]

AMD also has some halo models, but because they cant beat Intel in performance, they cant afford to charge $1000 for their high end chips.

FUD, pure FUD. AMD has always been cheaper than Intel. Even back before Intel introduced the Core2 series, when the AMD K2 and Athlon series spanked everything that Intel had to offer. Heck, even back to the days when AMD first entered the mass market (80386 days IIRC), they were the less expensive product. And to date, AMD has arguably always held the performance/$$$ award. Sure, Intel has started gaining a lead (Marginal with C2 series, but significant with the i7 series) in recent times, but AMD isn't THAT far behind. And if you consider that most of the true innovations in CPU design have come from AMD (true multi-core (I mean where there are 4 physical cores on die, not 2 dual core cpus on the die), 64bit, shared L3 cache, on-die memory controller, elimination of the north bridge and hence the system bus, etc), I find it VERY funny that "It is the price you pay for getting the bleeding edge" is applied to the more expensive Intel as opposed to the innovator AMD. Now, I'm not saying that Intel hasn't innovated at all. I'm just saying that the major innovations that the i7 used to surpass the C2 series (Namely the elimination of the system bus, on-die memory controller and the tiered cache architecture) were done first by AMD...

Re:Nice, but who has $1000 to pay on a CPU?

[TheRaven64]

I wonder how many of the people who can afford a $1000 CPU want an Intel chip. At that price, something like a Power6 or T2+ looks more attractive, depending on the workload.

Re:Nice, but who has $1000 to pay on a CPU?

[Vectormatic]

hey, i never said AMD was more expensive then Intel, and i bet you that if they could charge $1000 for their top end, they would (and they should, milking the high end is the easiest way to recoup dev costs)

personally i prefer AMD because of their price/performance ratio too, and they have consistently kicked intels but there

Re:Nice, but who has $1000 to pay on a CPU?

[Anonymous Coward]

but there....what?

huh?

Re:Nice, but who has $1000 to pay on a CPU?

[Pojut]

hey, i never said AMD was more expensive then Intel, and i bet you that if they could charge $1000 for their top end, they would

They never quite hit $1000, but their Athlon 64 FX-55 went for something like $700 or $800 when it was brand new.

Re:Nice, but who has $1000 to pay on a CPU?

[Kjella]

Nice, but who has $1000 to pay on a CPU?

Everybody that makes money off the processing power of their computers? Not many hobbyists would spend 1000$ on a camera, but photographers spends thousands. Granted, that's really a workstation market more than a consumer market, but it's not special like ECC RAM, Quatro graphics cards, SAS hard drives or similar server/niche products. If you use the right apps and get a 50% speedup it'll pay for itself in many places. Overall, I don't think it's a really expensive hobby if you want to drive around in a car costing 2000$ less and blow it all on computers. I could afford this one if I wanted to, I just don't see the point. It's so much else I could spent it on and so little extra gain.

Re:Nice, but who has $1000 to pay on a CPU?

[ircmaxell]

i bet you that if they could charge $1000 for their top end, they would

Well, if by "could" you mean with a better product, then no. That was proven in the days of the Athlon (When AMD owned almost every benchmark). They were number 1, but still the least expensive of the two by a fair margin.

If by "could" you mean with market position, then yes. Intel can charge $1k, because they have two things that AMD doesn't. First, brand recognition (I'd be willing to bet the "common" person knows Intel a lot more often than they know AMD). Second is manufacturer agreements. Intel in the past has held exclusivity agreements with a number of major computer manufacturers (And was one of the reasons for the big Anti-Trust lawsuits they faced in recent times). Intel could charge whatever they wanted not because they had the better product, but because they had the major distribution channels locked up. AMD was forced to undercut their prices just to be able to remain competitive...

Re:Nice, but who has $1000 to pay on a CPU?

[Anonymous Coward]

In the past the extreme chips were typically not really worth the bang/per buck increase over the mid level chips. They had more cache and were unlocked, etc but they were not = 2x the lesser chips, especially in rendering farms. Remember these are consumer chips, not server chips. Intel's new strategy is interesting. There will only be a 6 core extreme chip on the 32nm process to go beside the 4 core 45nm chips. So now there are 2 major differences between extreme and non-extreme i7s; the underlying process technology, and core number.

I suspect Intel will keep it this way and concentrate on 32nm chips for laptops, as they are doing now, and high end desktop and servers alone. In less than a year Intel should be unveiling the "tock" cpu design, Sandybridge. At this point their focus may shift, since all factories running the 32nm process will be qualified to fab those chips, but as of now there is no indication of a high mid range consumer 6 core chip based on Nehalem.

Re:Nice, but who has $1000 to pay on a CPU?

[petermgreen]

6 cores for under 200 bucks sounds nice doesnt it?
That all depends on how those cores perform.

Personally given the choice I'd rather have a higher per core performance than more cores. there is still a lot of single threaded stuff out there and even some multithreaded stuff has single threaded stages and/or a lot of locking between threads.

The information i've seen indicates that there will also be a slightly slower non-extreme version of the i7 hex core, wikipedia claims a release price of $562 though it doesn't give a source fot that claim.

Re:Nice, but who has $1000 to pay on a CPU?

[petermgreen]

I just took a look at a toms hardware CPU chart ( http://www.tomshardware.com/charts/2009-desktop-cpu-charts-update-1/Performance-Index,1407.html ), picked out the intel CPU that came immediately above the AMD CPU you mentioned and looked up the price on newegg ( http://www.newegg.com/Product/Product.aspx?Item=N82E16819115215&cm_re=i5-750-_-19-115-215-_-Product ) and it was $5 more.

Re:Nice, but who has $1000 to pay on a CPU?

[LWATCDR]

Rendering farms maybe. It would be an interesting trade off.
Does this chip offer more processing power for dollar than using more but cheaper CPUs? You would have to look at power, cooling, space, system, and admin costs. I would give it a big maybe.

Companies? Most corporate PCs could run on Atoms these days but in some areas I agree I see this being great.
Simulation/CAD/CAM and Video editing are the two that jump to my mind. Throw in a any number of Science applications as well.

Honestly I see them going in to gaming rigs. The "hard core" gamers are just that nuts. This CPU is right in line with the cost of a high end video card set up Two 5970s will run you that much if not more. Throw in your three 30" monitors and it all fits right into same insane price bracket.

Re:Nice, but who has $1000 to pay on a CPU?

[bfree]

And also, who strayed from the duo- quad- naming methodology?! Are you insane!? Do you have any idea the marketing power that a sexa core chip could have?

The same people who decided to never release a Sexium after the Pentium.

Re:Nice, but who has $1000 to pay on a CPU?

[ircmaxell]

Personally given the choice I'd rather have a higher per core performance than more cores.

I wouldn't. Most of what I do is very multi-tasking heavy. The fact that one program can't use more than one core doesn't bother me nearly as much as that 3 or 4 programs must share the same core. Especially when you consider that I typically run more than 1 VM at a time along side my regular programs, I think (for my use case at least) the more cores, the better the computer will perform. I very rarely use a single program (I don't play games on my computers), so the whole a faster core is better than n slower cores doesn't play as much of a factor in my computing. But to each their own (That's why they produce more than 1 line of processors)...

Re:Nice, but who has $1000 to pay on a CPU?

[petermgreen]

Two questions

1: do you know of any solid comparisons between those chips and current x86-64 chips using at least the same application software? (same OS would be nice too but it's difficult to chose one that is fair to all the candidates)?

2: do you realise just how much of the computing world is tied into either wintel or lintel?

Note that the particular chip mentioned in the current article is the desktop version, apparently there will be a dual-socket version but I haven't seen any recent information on when it will be released.

Re:Nice, but who has $1000 to pay on a CPU?

[cheesybagel]

Hyperthreading and MMX were arguably new additions at the time they came out. They might be shitty, but MMX was one thing which helped dedicated sound cards become obsolete. Intel is also a leader in semiconductor manufacturing processes. Which is part of the reason for their insane profits.

Re:Nice, but who has $1000 to pay on a CPU?

[cheesybagel]

The price is going down because the 32nm manufacturing process is coming online at Intel. Smaller transistors means you can add more to the same die area, at the same manufacturing cost. Moore's law et al.

Re:Nice, but who has $1000 to pay on a CPU?

[Zebra1024]

This can really save companies on software licensing. For example, SQL Server is licensed per CPU (not per core). So now I can have a single 6-core CPU running my database with one CPU license.

Re:Nice, but who has $1000 to pay on a CPU?

[ircmaxell]

I'm not saying Intel doesn't innovate. But look at the two big ones that you mentioned. Hyperthreading was introduced on the Xeon-MP line in 2002. MMX was introduced in 1996 on the original Pentium. Not to mention that MMX had nothing to do with sound cards (Other than the fact that it enabled the CPU to natively do the vector math that DSP chips were doing at the time). If anything, AMD's improvement on MMX --named 3DNow!, which added support for floating points in MMX instructions-- helped dedicated sound cards become obsolete. Then Intel took AMD's changes, incorporated them in MMX and introduced SSE. So, sure, Intel did innovate MMX, but the success of MMX (and its successors) was due to a combined effort between Intel and AMD (Intel's original design, and AMD's improvements on it).

And don't get me wrong. I'm not trying to imply that Intel is a bad guy at all. They do make some amazing silicon. My only point, is that people love to bash AMD, when you could argue that a significant portion of Intel's key features were either developed in parallel with AMD (Virtualization technologies for example) or were developed by AMD first (x64, on-die memory controller, elimination of north bridge, etc, etc)... That's not to say that AMD doesn't owe a lot to Intel as well. One without the other, and we wouldn't be anywhere near where we are today...

Re:Nice, but who has $1000 to pay on a CPU?

[BlueFireIce]

Not really true, seeing as both chips are "EE" (Extreme Edition) most of the time the prices on these stay pretty level, just like the Core2 Extreme Edition chips were still 1k at the time the i7's had well been out, come to think of it, look up the QX9770 right now and you will still see them selling for over 1k. These chips are sold to people who don't know better and/or can't/wont/don't know how to over clock. Anyone who has ever built a PC will wait until the "normal" line comes out and pay half the price and OC it to well past the stock clock of the Extreme Edition chip.

Re:Nice, but who has $1000 to pay on a CPU?

[drooling-dog]

Right now I'm using what must be one of the humblest CPUs on Slashdot, an Athlon XP 2500+. That's 1600 MHz of single-core 32-bit goodness. It's served me loyally for years with nary a complaint, and never missed a single day of work.

It still does almost everything I ask of it, but sometimes does struggle to keep up with HD video. I could help it out by getting a video card that supports VDPAU, but my equally faithful motherboard only has PCI and AGP, so there's not much room for upgrade there.

So finally it's time to retire them, and their replacements are on the way. The new kids are still pretty humble themselves, just an Athlon II X2 and a cheap AM3 motherboard. With 2GB memory, a grand total of $180. No bragging rights around here, of course, but there's nothing I'm likely to be doing for the next few years that they won't handle easily.

But here's the thing. I should be excited about bringing in the new regime, but I really feel like I'm spending my last few days with some good old friends. Should there be some kind of ceremony? Is there a computer heaven where they'll be waiting happily for me when I reach the end of my own days, along with my old 286DX25 and AMD K2? What a joyous reunion that will be...

Re:Nice, but who has $1000 to pay on a CPU?

[toastar]

given those two processors I'd take the AMD, And I'm a huge fan of the I5 architecture.

it comes down to 4x-2.66 or 4x-3.4

I do wish AMD did some jiggling with the on die cache. I think having a small L2 with a big L3 really isn't that smart. but i can't say that as fact :(

Re:Nice, but who has $1000 to pay on a CPU?

[0123456]

My only point, is that people love to bash AMD, when you could argue that a significant portion of Intel's key features were either developed in parallel with AMD (Virtualization technologies for example) or were developed by AMD first (x64, on-die memory controller, elimination of north bridge, etc, etc)...

On-die memory controller like the Intel 4004, you mean? And I don't believe the 4004 had a north bridge equivalent, since it could talk to memory directly.

AMD certainly deserve kudos for developing x86-64, but claiming that an on-die memory controller was some huge innovation when microprocessors have had on-die memory controllers since the stone age of computing is just silly. If there was a huge advance it was separating the memory from the CPU by attaching it to the north bridge so you could use any compatible CPU with any type of RAM that the north bridge supported... the reason we went back in the other direction was because latency became more important than compatibility.

Re:Nice, but who has $1000 to pay on a CPU?

[wgaryhas]

FUD, pure FUD. The Athlon FX-57 came out at an initial price of $1031 for distributors. While the intel competitor at the time, Pentium 4 Extreme Edition 3.73 came out at $999. Then the Athlon FX-62 came out at $1031 as well, then a month later the first core 2 duo came out for $999, and AMD hasn't been able to successfully charge a thousand dollars for a desktop processor since. So, saying AMD was never more expensive is flat out wrong.

Re:Nice, but who has $1000 to pay on a CPU?

[ILuvRamen]

Well with the Core 2 quad extreme at $1000 and the core 2 quad like 0.1 GHz down from it at $550, I don't think you'll have to wait long for it to get a bit cheaper. I'm hoping this forces the wolfdale duos down though. I don't have a huge problem doing builds and upgrades with 3.0GHz Regors but the fast wolfdales are like 10-20% faster!

Re:Nice, but who has $1000 to pay on a CPU?

[PitaBred]

AMD's chips don't change sockets every 2 months. I can upgrade my AMD CPU without having to upgrade my entire machine. You can't compare the cost of the Intel chip directly to the AMD chip without taking the other costs into account as well.

Re:Nice, but who has $1000 to pay on a CPU?

[ryantmer]

I wouldn't. Most of what I do is very multi-tasking heavy.

Same here. You can also overclock if per-core performance isn't high enough for you (if you wish to do so), but I have yet to see a simple method of increasing the number of CPU cores through your BIOS...

Re:Nice, but who has $1000 to pay on a CPU?

[BlackSnake112]

AMD's chips don't change sockets every 2 months. I can upgrade my AMD CPU without having to upgrade my entire machine. You can't compare the cost of the Intel chip directly to the AMD chip without taking the other costs into account as well.

I thought I remember something similar being said about AMD switching to AM2 then AM3 sockets. Yes you can plug an AM3 CPU into an AM2 socket but there was (is?) a performance hit. Also how long did intel keep the 775 socket? But intel should have kept the 1156 socket around longer. They did jump to socket 1366 really fast when compared to how long the 775 socket was around.

Re:Nice, but who has $1000 to pay on a CPU?

[aztracker1]

They have both 1156, and 1366.. the 1366 is meant for higher end systems, and the 1156 is meant for more retail channels and OEM/ISVs. The biggest difference is the reduced architecture costs of 1156 vs. 1366. Also, I tend to replace both my motherboard and CPU on a 2-3 year upgrade cycle. My most recent upgrade to an 1156 based Core i7-860 is probably more than I really need. I really didn't feel much pain on my old system (Core 2 Duo E6600), but felt it was time. Other than maybe a GPU refresh in 18 months or so and barring hardware failure, unless something comes down the chain that is really compelling, I don't see myself upgrading anything in the next 5 years. My last system was about 3 years old at the time of upgrade.

I'm actually a big fan of AMD, and I think their 780-785G based motherboards and on-board graphics are a far better combination for a typical home user than anything on the Intel side. I know a lot of people laugh at an on board gpu, but they're (AMD ones) decent.

Re:Nice, but who has $1000 to pay on a CPU?

[hairyfeet]

But how many people are actually gonna have enough work to feed 6 cores for any length of time at all? I'm thinking that niche has to be pretty teeny tiny, considering when I do follow ups on my customer's new builds and check the logs a good 80% of the time the duals are twiddling their thumbs. I know that even with me liking to do the occasional video editing and transcoding my AMD quad sits idle probably 70%+ of the time.

So I've started to go the other way for myself and my customers, as it really is starting to feel like an ePeen thing now. Instead of getting the absolute fastest I used a 95w max quad, and I'm waiting for the parts to build a couple of the new 65w PCs for my customers. The savings in power and heat makes for a much quieter and cheaper to run PC, while still being fast enough it will be sitting idle a good 65-75% of the time, because the average Joe just doesn't have enough hardcore number crunching to keep these new monsters fed. With games now being made for console first and then being made (shudder) "multiplatform" even they just aren't pounding the CPU hard enough to justify shelling out $1000 for a CPU IMHO.

With most of my customers I've found the duals are frankly "good enough" while the triples and quads, which many get because the price for AMD multi is so cheap, is frankly extreme overkill. I can't even think of what any of my customers could do to come up with enough work to make them think they need 6 cores.

Re:Nice, but who has $1000 to pay on a CPU?

[InvisiBill]

But intel should have kept the 1156 socket around longer. They did jump to socket 1366 really fast when compared to how long the 775 socket was around.

1366 came out first for the i7. 1156 came out recently for the i5 and i7. Both are still around, and will be for at least a while. 1366 is aimed at enthusiasts and workstations, while 1156 is a "mainstream" part with some limitations compared to 1366.

Re:Nice, but who has $1000 to pay on a CPU?

[Khyber]

"Rendering farms?"

Those would be handled by massively parallel GPU clusters, not slower than crap CPUs.

Re:Nice, but who has $1000 to pay on a CPU?

[Khyber]

1. Web database servers
2. Not very much considering most actual production machines run a Unix variant (I mean PHYSICAL production, not software production) and I've had the displeasure of having to repair major industry machines, which forced me to learn EIGHT different Unix subsets.

No *REAL* production house (except maybe digital art studios) uses Wintel, and in fact Pixar is likely moving to solid-Tesla computing in the near future.

Re:Nice, but who has $1000 to pay on a CPU?

[InvisiBill]

AMD's chips don't change sockets every 2 months. I can upgrade my AMD CPU without having to upgrade my entire machine. You can't compare the cost of the Intel chip directly to the AMD chip without taking the other costs into account as well.

I actually built my system with an i7 because my AM2 board (about 16 months old and fairly high-end) apparently came out just before AM2+ and therefore couldn't be reused. If you happen to have a good AM2+ system, you may be able to drop in an AM3 CPU for a boost, but simply because you have AMD doesn't mean you suddenly have a free upgrade path. In my case, I would've had to buy the same components to go with Intel or AMD, so I ended up spending a little more on Intel to get a lot more performance.

As for the "2 month" crack, the LGA1366 socket used by the i7 980X being reviewed was first released for the i7 in November 2008. It's almost a year and a half old already. I waited for the prices to drop some before building my i7 system, and I've had it for over a year now.

I doubt I'll spend $1,000 on a new CPU but I may upgrade to one of these 6-core 32nm models when they release a more reasonable version. These new CPUs have been supported by many LGA1366 boards for a while with nothing more than a BIOS update.

Re:Nice, but who has $1000 to pay on a CPU?

[petermgreen]

But intel should have kept the 1156 socket around longer. They did jump to socket 1366 really fast
They didn't jump from 1156 to 1366 at all (1366 is actually older than 1156). They created two different sockets for different markets (and I'm pretty sure there will be a third soon for the new processors with 8 cores, 4 QPI links and seperate memory buffer chips).

1366 is a socket really designed for dual-socket workstation and server stuff but also used for some high end single processor stuff. 1156 is the mainstream socket.

Re:Nice, but who has $1000 to pay on a CPU?

[petermgreen]

Can you provide links? A quick search for xeon vs powerpc benchmarks didn't turn up much other than articles about how the new intel macs were better than the powerpc ones.

Re:Nice, but who has $1000 to pay on a CPU?

[johnw]

but until these come down a long way in terms of price, it is WAY out of my price range

This is your lucky day. I happen to know where I can get you a pallet of really cheap Intel Core i7 processors, retail boxed, complete with heatsink, fan and a booklet.

Re:Nice, but who has $1000 to pay on a CPU?

[YesIAmAScript]

First, your comment is non-responsive. He never said Intel was cheaper, he said AMD doesn't have a model with performance levels high enough to merit a $1000 price tag when compared to what you can get for $1000 from Intel.

Second, you cherry pick innovations for AMD.

What do you care if your four cores are on-die? Larger dice means lower yield. Two-dice in a package can increase yield and thus reduce price or allow you to have a lot more cache, which helps performance.

Did not Intel do L3 cache first with the Pentium 4 Extreme Edition on Gallatin in September 2003? I think AMD's largest contribution to cache technology is NUMA, not L3 cache. It took Intel at least 5 years to even approximate the advantages of NUMA.

AMD didn't eliminate the Northbridge, they just took one part of it out, the memory controller. The Northbridge was still needed for fast I/O devices, most notably PCIe slots. Intel was first to remove the Northbridge with the Core i7/8xx series.

Intel did on-chip cache first. They did hyperthreading first. They did SIMD extensions (MMX) first. They did superscalar first (with Pentium and its U and V pipes). They did on-chip floating point first (486DX). They did power/heat management first with Speedstep. Intel did 32-bit memory spaces first (and the technology is still used, even with 64-bit extensions in place). Intel pushed the bounds of memory bandwidth far beyond what AMD did. When AMD had 1GB/sec memory bandwidth (133MHz, 8-byte wide), Intel had 6.4GB/sec (800MHz, 8-byte wide). Intel went to pinless page (no more bent pins!) first.

AMD obvious did a ton of good stuff, well disproportionate to their slice of the competitive marketplace. Besides NUMA above, they did 64-bit, they also were deep into multi-package (multi-chip) MP back when Intel wasn't. This never became mainstream because the motherboards cost $300, but it was an advance. They integrated the memory controller (a feature I think was vastly overrated, given that it tied AMD to memory technologies, keeping them stuck on DDR as DDR2 dropped below DDR in price and was already well past it in performance and power).

More importantly, there is the stuff AMD didn't do. AMD didn't put CPUs on a huge cartridge package (slot 1 and the truly massive slot 2!) in order to try to bundle SRAM memory (cache) chips into your CPU purchase to increase sales. AMD didn't sign an exclusive with RAMBUS forcing everyone to use massively overpriced, underperforming RAM. AMD didn't jump on FB-DIMMs (essentially RDRAM-lite), which again are massively slow and power hungry.

AMD matters, but I think you give them too much credit on innovations.

Re:Nice, but who has $1000 to pay on a CPU?

[Mashdar]

Did you really just try to use a 70s era PDIP processor as a counterpoint to AMD's incorporation of a modern memory controller into a CPU die? Thats like saying that NASA's space shuttle's ability to be re-used was nothing new because airplanes were always reused.

Re:Nice, but who has $1000 to pay on a CPU?

[m.dillon]

People should also remember that Intel's graphics strategy is to try to do it with a general purpose cpu rather than a dedicated graphics engine. This could be part of that strategy. With so many cores available a system can simply dedicate a few to particularly important jobs such as rendering.

-Matt

Re:Nice, but who has $1000 to pay on a CPU?

[petermgreen]

Not to mention that MMX had nothing to do with sound cards (Other than the fact that it enabled the CPU to natively do the vector math that DSP chips were doing at the time).
Another thing that MMX brought that previous CPUs core stuff didn't have afaict was saturating arithmetic. Traditionally CPUs do modular arithmetic.

Saturating arithmetic is useful if you are trying to do stuff like audio mixing in software.

What really made dedicated sound cards obsolete though IMO was a combination of advancing general CPU power and the fact that intel made it really cheap and easy to hang an audio chip off thier southbridges (by integrating a lot of the functionality into the southbridge and providing relatively simple interfaces for the audio codec* chips to connect to)

*Note the term codec has at least two meanings in computing. In this context it reffers to a chip that is both an ADC and a DAC.

Re:Nice, but who has $1000 to pay on a CPU?

[Mashdar]

PS What they called RAM back then we call cache, and it never left the die.

Re:Nice, but who has $1000 to pay on a CPU?

[hairyfeet]

Dude make it into a file server or netbox. no reason to toss when KVMs are dirt cheap, I think I paid $24 for my 4 port at Newegg with cables. I am typing this on a Sempron 1.8Ghz with 1.5Gb of RAM, which makes for a whisper quiet netbox/downloader without needing to fire up my quad.

So don't toss dude, re-purpose. As long as it still runs good and doesn't throw errors there is no reason you can't still get plenty of use out of it as a file server, netbox, or a dedicated box for downloading large files. Just add a nice cheap KVM and you are good to go and you'll be glad you have it, just as I am glad to have this whisper quiet Sempron for checking my email or downloading files at 3AM.

Re:Nice, but who has $1000 to pay on a CPU?

[ircmaxell]

Did not Intel do L3 cache first with the Pentium 4 Extreme Edition on Gallatin in September 2003?

Yes and no. L3 just means level 3, and yes Intel did introduce that in 03. But when I said L3, I meant a shared L3 across cores. All of AMD's multi-core chips have a shared cache level accessible by all cores. Intel didn't with the first few rounds of the C2Duo series (Including the original C2Quad chips).

AMD didn't eliminate the Northbridge, they just took one part of it out, the memory controller.

Well, even as such, I don't think you could still call it a north bridge then. One key feature of a Northbridge is connecting to the CPU via a Front Side Bus. AMD eliminated that bus when it introduced the point to point interconnect (Hypertransport). So yes, there is still one of the functions of the Northbridge still remaining on the mobo. But because the method of interconnection is different and the primary function is gone I don't think you can still call it a Northbridge (IMHO)...

AMD matters, but I think you give them too much credit on innovations

You proved my point when you said "AMD obvious did a ton of good stuff, well disproportionate to their slice of the competitive marketplace."... I wasn't trying to imply that Intel stood still in the past few years (or ever really). My implication was that people tend to shit on AMD and say "Well, they don't perform, they don't innovate" when the fact of the matter is that a significant number of advances that led to today's i7 processors came from AMD...

Re:Nice, but who has $1000 to pay on a CPU?

[CharlieHedlin]

My first thoughts! With SQL server licenses as expensive as they are you can throw a lot of very expensive hardware at them (an array of 10 SSDs anyone) to get the most out of a single or dual CPU license.

Re:Nice, but who has $1000 to pay on a CPU?

[igxqrrl]

And to date, AMD has arguably always held the performance/$$$ award.

Baloney. That all depends on which specific slice of the performance market you look at. You can certainly find markets for which AMD offers better bang for the buck, and you can find plenty where Intel offers better value.

Sure, Intel has started gaining a lead (Marginal with C2 series, but significant with the i7 series) in recent times, but AMD isn't THAT far behind.

Right now, AMD is quite far behind in terms of performance by any subjective measure.

And if you consider that most of the true innovations in CPU design have come from AMD (true multi-core (I mean where there are 4 physical cores on die, not 2 dual core cpus on the die), 64bit, shared L3 cache, on-die memory controller, elimination of the north bridge and hence the system bus, etc), I find it VERY funny that "It is the price you pay for getting the bleeding edge" is applied to the more expensive Intel as opposed to the innovator AMD.

This is misleading. None of the items you list were first developed by AMD unless you limit yourself to the x86 space. Your multi-core example is a common but incorrect simplification. Intel suffered from this lesson. The P4 was extremely had an unconventional and aggressive microarchitecture but many of those bets did not pay off. Some such as hyperthreading, have survived and returned to the lineup. AMD's "innovations" are by comparison downright pedestrian. "Innovation" is ultimately marketing-speak for engineering tradeoffs. Perhaps it's "innovative" to over-engineer a product, but if the end result is that you're a day late and a dollar short, then that innovation was a stupid business decision.

Re:Nice, but who has $1000 to pay on a CPU?

[petermgreen]

when microprocessors have had on-die memory controllers since the stone age of computing is just silly
IMO the innovation is multiple links out of the processor.

If you look at older computers (certainly stuff like BBC micros and i'm pretty sure eartly PCs were the same) nearly everything was on one bus (with maybe the odd bus buffer chip somewhere or maybe a DRAM refresh chip if the CPU didn't have that capability built in). This worked with the tech of the time but as things started to speed up it became a liability.

Gradually the busses split apart more and more due to the facts that a mixture of slow and fast devices had to be dealt with and the fact that fast busses and large numbers of devices are not a good combination but there was still a single shared bus linking the processors to everything else and this link was becoming a bottleneck especially in multiprocessor systems.

AMD eliminated the shared FSB and replaced it with on-chip memory controllers and point to point hypertransport links.

Intel did the same (but with quickpath instead of hypertransport) much later.

Re:Nice, but who has $1000 to pay on a CPU?

[yuhong]

AMD didn't put CPUs on a huge cartridge package (slot 1 and the truly massive slot 2!) in order to try to bundle SRAM memory (cache) chips into your CPU purchase to increase sales.

They did, with the Slot A Athlons.

Re:Nice, but who has $1000 to pay on a CPU?

[yuhong]

And if you think Intel Lynnfield's integrated northbridge was new, Transmeta did that first.

Re:Nice, but who has $1000 to pay on a CPU?

[yuhong]

Yep, Intel's AC' 97 and later High Definition Audio.

Re:Nice, but who has $1000 to pay on a CPU?

[BikeHelmet]

It is the price you pay for getting the bleeding edge, AMD also has some halo models, but because they cant beat intel in performance, they cant afford to charge $1000 for their high end chips.

That's not the reason.

Even when AMD had the fastest chips, they were cheaper. AMD seems to charge less because Intel is a marketing behemoth.

Back around the time Intel released their P4 Extreme Edition chips, they were charging over $1000/cpu. AMD was charging a meagre $600-700 for their FX-55 which slaughtered it. And I picked up an Athlon XP for $80 around that time, which let me play all the new games. :P

Re:Nice, but who has $1000 to pay on a CPU?

[Zeio]

I'd buy it on sight if it supported ECC. No ECC support = unstable system. I always have an ECC system, and I always get high "3DMarks" and frame rates and I never get a BSOD or other system errors.

Without ECC its impossible to know if memory errors are occurring, and 12GB of memory at 1333/1600MHz probably has a single bit event quite often.

Re:Nice, but who has $1000 to pay on a CPU?

[drkim]

No *REAL* production house...uses Wintel

I'm not sure where you get your information, but a great number of production facilities are Wintel based.
I'm in the business - and there are too many small ones to name - but one huge, new one is the $350,000,000 Letterman Digital Arts Center (and the Lucasfilm Animation Singapore unit)

Chips for the Mac Pro refresh I believe.

[jo_ham]

I believe this is what's been holding up the Mac Pro refresh, with the top or middle Mac Pro slated to get these as an upgrade from the 4 core ones.

I think core number is the new MHz. We're not going any faster, but we can just give you more of them, which makes quite a lot of sense. All those FCP render pipelines and encodes just got a lot shorter with th3 12 core Mac Pro.

Re:Chips for the Mac Pro refresh I believe.

[Anonymous Coward]

Well, new chips are still made over a wide range of clock speeds, so it's a bit early to throw that number away. We see chips clocked at 1.x ghz (various Atoms) up to 3.x ghz at the high end. And we see chips with 1, 2, 3, 4, or 6 cores now. But three cores at 3ghz will probably outperform four cores at 2ghz, whether the task is single threaded or multithreaded.

Of course what I'd rather see as the center of advertisement would be two different numbers: 1) performance on some sort of standard single-threaded program, 2) performance on some sort of standard embarrassingly parallel program. That would probably require a competing chip architecture to get popular, though. The latest ARM designs might succeed, if they can continue the push from embedded to smartphone/netbook, but it's too soon to tell.

Re:Chips for the Mac Pro refresh I believe.

[Wesley Felter]

The Mac Pro will use Xeon 56xx, not Core i7 (although they're basically the same chip, the 56xx hasn't been announced).

Cool

[Pojut]

Now to see what AMDs 6-core offering is like. I know that Intel destroys AMD in performance benchmarks and real-world performance, but AMD is FAR less expensive. If I was pushing an Eyefinity setup or something, then sure, I would go all out and drop a few hundred dollars or more on an Intel CPU. Considering that AMDs current flagship costs $195 and is still a heck of a performer...yeah, I'll stick with AMD for now.

Re:Cool

[biryokumaru]

If the AMD chip has a higher bang for buck ratio, why not just do the sensible thing and make a beowolf cluster? It's just like have more cores, because you do.

Re:Cool

[Anonymous Coward]

Er ... FAIL

Top 500 stats show that Intel has over 80% of the top 500.

Re:Cool

[Pojut]

Sorry...my experience with late-90's Warez websites has trained me never to look at a site with "top" followed by a number in its URL -_-;;

Re:Cool

[beelsebob]

AMD's flagship chip does indeed cost $195, but then, it's about the same speed (as the benchmarks showed) as the Core i5 750, which costs $199. AMD isn't offering better bang for you're buck, they're offering high energy use CPUs with comparable performance to intel's similarly priced CPUs.

That Phenom II uses 30W more than the Core i5, so it'll cost you about $30 a year more to run, and be less upgradable.

Re:Cool

[Vectormatic]

That Phenom II uses 30W more than the Core i5, so it'll cost you about $30 a year more to run, and be less upgradable.

True, except for when you already have a AM2/AM2+/AM3 board, or a good supply of ddr2 ram. In that case the phenom is a drop in upgrade, versus a platform upgrade for the i5. Also keep in mind that AMD will be releasing 6-core CPUs this year too, and they will fit in any recent AM2+/AM3 board, while for the intel high end stuff, you are locked into their 'premium' 1366 socket.

anyway, AMD does have a slight edge in BFTB in the lower segment

plus i like rooting for the underdog

Re:Cool

[Pojut]

True, except for when you already have a AM2/AM2+/AM3 board, or a good supply of ddr2 ram. In that case the phenom is a drop in upgrade, versus a platform upgrade for the i5. Also keep in mind that AMD will be releasing 6-core CPUs this year too, and they will fit in any recent AM2+/AM3 board, while for the intel high end stuff, you are locked into their 'premium' 1366 socket.

This applies to me. I just ordered AMDs 965 Phenom II to replace the Athlon 64 X2 5400+ currently in my AM2+/AM3 Gigabyte board...when the new AMD chipset is widely released with SATA6/USB3 and the price becomes reasonable, I'll order one of those motherboards. Until then, my AMD 785 chipset board will suffice. AMD has always been pretty good about making sure their sockets are versatile, and the AM2+/AM3 boards are the most versatile yet.

plus i like rooting for the underdog

This is also a reason why I stick with AMD. They're the only ones producing CPUs that can remotely compete with Intel in the consumer space, yet they are a MUCH smaller company. I like that.

Re:Cool

[gumbi west]

Yeah, but if you count by Rmax share (instead of # of supercomputers share), Intel only has 50%.

Re:Cool

[petermgreen]

I know that Intel destroys AMD in performance benchmarks and real-world performance, but AMD is FAR less expensive.
hmm, are you aware of any good comparisions between the best AMD chips and the best intel chips available at a given price point?

I tried to do one by taking a look at http://www.tomshardware.com/charts/2009-desktop-cpu-charts-update-1/Performance-Index,1407.html, looking up prices on newegg and ingnoring pricessors that are either unavailable at newegg or are more expensive than a faster chip of the same brand and limiting myself to quad core chips I got the following in decreasing order of speed

Intel Core i7-975 Extreme Edition $969.99
Intel Core i7-950 $569.99
Intel Core i7-870 $569.99
Intel Core i7-920 $288.99
Intel Core i5-750 $199.99
AMD Phenom II X4 965 $194.99
AMD Phenom II X4 955 $160.99
AMD Phenom II X4 945 $150.99
Athlon II X4 630 $99.99

I got bored and stopped after this point

but while doing it I realised that toms hardware mostly only tests high end stuff so it isn't a very usefull comparision (in partcular there was only one i5 quad core in that list)

Re:Cool

[petermgreen]

One thing to consider is that the cost of using a CPU is not the same thing as the cost of the CPU.

Every CPU needs to be put into a socket. That socket has to be on a motherboard*. That motherboard needs a case, a PSU, ram, a switch port, something to boot off (admittedly the onboard nic may allow this). It will also need to be put in a case and those cases stored somewhere (perferablly a proper rack)

When calculating the bang per buck of a given CPU choice you have to include these support components as well as the CPU itself!

Plus not all software is friendly towards being spread accross a cluster of machines.

*Yes you can have more than one CPU on a motherboard but both motherboards and processors increase in cost hugely if you want to do that.

Re:Cool

[T-Bone-T]

Of course. You have to do that calculation for yourself because it only applies to you. The cost of the CPU is the only constant and possibly the only expense. What if you are just replacing the CPU? Those other costs don't matter.

Re:Cool

[zdzichu]

AMD is just about to ship 12-core Magny Cours to customers. That's a beast!

Re:Cool

[petermgreen]

Of course, most top supercomputers take years to plan and build
Indeed and that means most of them were probablly designed before the xeon 5500 series (when intel finally got arround to replacing the FSB with QPI) was available.

And if you want more than 2 sockets or more than 6 ram slots (not that most of these big clusters do, they preffer to just use more nodes) on a system with a point to point based architecture AMD is still the only option.

Re:Cool

[petermgreen]

True if you already have a platform that can support a chip with the performance you want it's often a better buy.

while for the intel high end stuff, you are locked into their 'premium' 1366 socket.
True, BUT amd simply don't have anything comparable to the intel high end stuff anyway. AMD high end stuff is comparable to intel midrange stuff.

Will the AMD 6-core be better than a comparablly priced LGA1156 intel quad-core available at the time? The answer probablly depends on the workload (e.g. can the workload fill all the cores all the time, most can't).

Re:Cool

[BikeHelmet]

That Phenom II uses 30W more than the Core i5, so it'll cost you about $30 a year more to run, and be less upgradable.

Haha! Interesting way of looking at things.

I bought an AM2+ board and Athlon X2 way way back. 2008, I think it was. I upgraded to a Phenom II X4 925 about a month ago when I saw the ridiculous $109.99 launch price on NCIX. (Yes, I'm Canadian) It's now overclocked to 3.5ghz. What do I use it for? H.264 encoding (x264) and games. (TF2, L4D2, anything I can buy on steam)

Looking at these benchmarks... I'm getting just under half the performance of this flagship CPU, for about 11% the price. It's an impressive chip - probably great for companies rendering or encoding stuff 24/7, since it'll shine on power efficiency - but really not worth it to me. Another important consideration for rendering is how many hours it takes before you have something to show your client, which also doesn't apply to me.

I find the game benchmarks to be a laugh. I want to see the performance difference at MAX settings, too, because that's what I play at. In reality, this new chip probably only gives a 20% fps boost. Since I'm already at 60fps, there's not much point blowing money.

I would consider the Core i5 750, but it's $218.99 on sale on NCIX, so that's literally twice the cost for an equal amount of performance. (But more overclocking room, I imagine) I'd also have to buy a more expensive board... really not worth it. Regarding power consumption - I measured with my Kill-A-Watt. This Phenom II X4 at 3.5ghz consumes 105 watts when idle, and 135 watts when x264 encoding. I have a GTS 250 in my system, two WD black HDDs, a DVD burner, etc.; PSU is an efficient Corsair HX620. Even when gaming, it spits cool air out the back. It's really not that bad. It could be significantly worse than the Core i5, but I don't have one to compare. At the current rates in BC, leaving my PC on should only cost ~$48, so I'm betting the savings are more like $8/yr.

In case you missed my point, I call shenanigans on your post. AMD is clearly more upgradeable and cheaper.

Re:Cool

[Agripa]

AMD's flagship chip does indeed cost $195, but then, it's about the same speed (as the benchmarks showed) as the Core i5 750, which costs $199. AMD isn't offering better bang for you're buck, they're offering high energy use CPUs with comparable performance to intel's similarly priced CPUs.

It is really more comparable to the Intel Xeon X3450 ($241) because the Core i5 series lacks ECC support.

Wow!

[Mad Merlin]

Just image how fast you could play Game! with that beast!

No thanks

[oodaloop]

Unless it's lead with a solid plastic fan, I'm not interested.

the new 12 core mac pro starting at $4500 with

[Joe The Dragon]

the new 12 core mac pro starting at $4500 with 6gb ram and ati 5350 512 video. Price to high you can get the $800 mini with i5 430 and Intel video with 4gb ram.

Turbo mode?

[Stenchwarrior]

After 12 years, I finally have a use for that TURBO button on the front of my case again.

Re:Turbo mode?

[TheRaven64]

The Turbo Boost mode is present on most of the newer Intel chips. It overclocks one of the cores, while underclocking the others, to give single-threaded apps a boost without exceeding the thermal envelope. It needs some extra support from the OS scheduler, because suddenly you have different cores running at different speeds, which messes up process accounting. As I recall the OS needs to specifically request turbo boost mode, which it does when one process is using all of the CPU time that it is given but other cores are idle.

Re:Turbo mode?

[Cowclops]

Thats not how it works. What it actually does is shuts power off to cores that aren't in use, and then overclocks the remaining ones. It won't/can't run different cores at different clock speeds. So if you have a 4 core processor, it might shut off 2 of the cores and then boost the clock speed by up to like 50% depending on what CPU it is, up to the thermal limits of the processor. The "breakthrough" in engineering is the part of the circuit that shuts off power to the unused cores better than anything else has in the past. This essentially gets you the best of both worlds in a single CPU - a lower clocked quad and a higher clocked single/dual core.

Re:Turbo mode?

[BikeHelmet]

So Turbo Boost is Phenom power management + automated overclocking?

Finally

[Rallias Ubernerd]

A CPU that can actually efficiently run my Artificial Intelligence program... if each core is 1.5x scaler.

No, no...

[PostPhil]

...you're not getting it. It's waay better than the AMD stuff because it's xxxxxxtreme! That's like *twice* the xxx!

Reminds me

[gaelfx]

This really reminds me of the recent Ask Slashdot article lamenting the naming schemes being implemented for most pieces of hardware. i7= 4 or 6 cores. Makes sense since the first thing I think when I hear 7 is "must be 4 or 6!" And the '980' really goes a long way towards confirming that initial suspicion. I'm really glad they put the 'extreme' in there, cause I was worried about the numbers being too low. Seriously though, can't they come up with a name that is actually descriptive of the product rather than a bunch of reassurances about the awesome-o amazingness of their processor? It seems to me that most people ask someone who knows something about computers when they need to buy a new one or replacement parts for their old one, and I don't know about the rest of you, but I really hate names that give me no real information about what the heck I'm buying. Yes, I can google the information, but the whole practice seems immature (and sometimes a little insulting).

Re:Reminds me

[L4t3r4lu5]

Of course!

"Introducing the new Intel Core 1336-32nm-3.33-6x+HT-VT-12MBSC!"

Re:Reminds me

[drinkypoo]

I'm right there with you. How the hell would anyone know the Phenom II 720 is a triple-core, 2.8 GHz processor with a K10 core? Assuming I even remember correctly. When I hear "phenom" I think of Dre, not a K10.

Re:Reminds me

[drooling-dog]

Have to say I'm disappointed too. I wanted to know whether the i{N} naming is N=3,5,7 as in odd numbers or primes. This was going to be the chip that settled that once and for all, because it would be either the i9 or the i11. The mystery lives on.

Re:Reminds me

[Rockoon]

I would love a return to uniform class-based suffixes. For example, the 80386SX vs 80386DX. Although "SX" and "DX" isn't descriptive, its UNIFORM. The SX chips were running a 16-bit bus and the DX chips were running a 32-bit bus.

Later Intel moved to Pentium vs Celeron, but Celeron itself wasnt uniformly descriptive (beyond meaning "shit") of the differences between them. Some Celerons had their cache's cut in half, others were simply a lower clock rate, still others were a combination of the two.

I strongly suspect that its Intel's (unreliable) FAB process causing many partial duds that creates the need for so many variations in their line. So many variations in fact that its just not a good idea to make most of the differences obvious.

As far as AMD's line, the number of cores is always in the full name of the product: "Phenom II x4 965" .. still, that "965" number is itself meaningless information. It doesnt tell me shit other than that its better than a "955" (at least there is that!)

Re:Reminds me

[TyFoN]

I thought the difference between DX and SX was that the SX didn't have a math co processor.. Maybe I am remembering wrong though.. It is a while ago :)

Re:Reminds me

[Rockoon]

80386's never had integrated math coprocessors. The 80387 was a separate chip (I had one!), socket compatible with the 80287.

The 80486 was the processor where DX meant integrated mathco. Not well know is that the 80487 was a full blow 80486DX chip, and when placed in the mathco socket, it completely disabled the "main" socket.

Re:Reminds me

[bdenton42]

i7= 4 or 6 cores. Makes sense since the first thing I think when I hear 7 is "must be 4 or 6!"

Some of the i7 models for mobile use only have 2 cores, just to confuse things even further.

Re:Reminds me

[ricky-road-flats]

Really, it's not as bad as you make out. Firstly and most importantly, as you've already pointed out you *can* Google for it. 20 seconds on the wikipedia page for Intel's processors can tell you what you need to know about the model number in front of you.

You really think they haven't thought about how to differentiate their products and make things clear? In reality they have a lot of different product ranges to cover, from multi-socket servers down to netbooks and PDAs.

Then they have variables like number of cores, HT or not, VT or not, integrated graphics or not, 45 or 32 nm, which socket it's for, TDP, Speedstep, bus/QPI speeds, OEM or retail, which SSE/ANI extensions they have, L1 cache size(s), L2 cache size(s), L3 cache size, form factor... the list goes on.

Then remember the VAST majority of their customers don't give a flying fuck about *any* of that - they're buying a whole computer to a price point, and the reality is almost any modern CPU will do for them. A vanishingly small number of their sales are individual CPUs to geeks, and for those geeks the information is out there.

In summary, how would you suggest they do it? Is there an industry where such a complex product range is all simple, clear and straightforward? The car-analogy /. fetish gives an example of similar confusion - my BMW 323i is not a 2.3 litre engine as the model number implies, it's a 2.5, same as the 325i. All customers care about is that the 325i is MOAR - it's faster than a 323i. The 318i is a 2 litre, same as the 320i. But the bigger number is MOAR. If you have more money, you go for the bigger number. And that's the level of detail most people want.

Re:Reminds me

[jadin]

No clue if it's true or not, but I was told back then that the move from x86 to pentium was because they could not trademark a sequence of numbers. Their competitors could sell 80486 no problemo, but could not sell a pentium.

Re:Reminds me

[gaelfx]

My point was that NONE of the relevant information can be gathered simply from examining the model name, and yes, I realize that the vast majority of the information about the processor is irrelevant to most people buying them, but the most essential information could easily replace the senseless numbers they currently use to name their processors. Number of cores, Clock-speed (whether multiplied or not) and power usage seem to be the most important information, FSB I could take or leave and socket type doesn't matter since sometimes MB's aren't compatible with procs of the appropriate socket (I ran into this when I recently built a box with a Gigabyte MB).

Re:Reminds me

[NervousNerd]

First, you seem to have forgotten that the processor is called the Phenom II X3 720. That right there should tell you that it's a triple core. Secondly, how do you know the Athlon XP is the K7 core? Or the Pentium 4 is the Netburst core? Your average joe doesn't buy a processor. They buy a cheap Dell with a Celeron in it. They turn it on and it works. They don't care what happens, as long as it works. The people that buy processors usually look into them.

Re:Reminds me

[petermgreen]

Actually there is quite a bit of information in that part number if you know how to read it. I agree they could be less cryptic though

i3/i5/i7 seem to be intels indication of the target market. i3 is economy, i5 is mainstream and i7 is high performance.

The first digit of the number is the series. In general higher series are better in some respect than lower series.

The final pair of digits indicates where in that series it is, the higher the better.

The x says it's an "extreme edition" which means you get an unlocked multiplier, generally the extreme editions are also the fastest chips in thier family.

Unfortunately intel then go and screw up thier own scheme by putting the new hex core processor in the same series as it's quad core predescessors.

Re:Reminds me

[petermgreen]

well, bigger numbers are better.
Not nessacerally. According to tom's hardware an 870 outperforms a 920.

Does it still work on yield?

[GuyFawkes]

I know years ago Intel did not, for example, make a 3 GHz P4, they made shed loads of P4's and then gave each one a clock speed that it would handle, so you had a distribution curve from each batch that ran from maybe 1.6 to 3.2 GHz, and priced accordingly.

I can't imagine any recent changes in chip production per se that would mean an end to this distribution curve out of each batch.

Rather this is a case of a new process finally coming on line with the production bugs mainly worked out, which shifts the distribution curve up to higher clock speeds, the analogy here is the LHC or any other complex system.

Certainly when building systems myself you always went for near the peak but on the high side of the curve, to get most bang for your buck and also most reliability.

Re:Does it still work on yield?

[petermgreen]

Normally I would agree with you, the top models traditionally tend to carry an insane markup and yet only a modest increase in performance.

The complication here is that for the right workloads going to 6-core brings a large leap in performance over four core. The non-extreme single socket versions of this chip aren't coming out until Q3 2010 and while afaict you can use a dual socket capable version in a single socket setup it doesn't make much sense to do so.

Re:Cool, but why?

[Vectormatic]

Hold on, let me fire up eclipse and i'll get right on it!

Holy cow!

[L4t3r4lu5]

New 6-core processor is super-fast in synthetic benchmarks and when coupled with applications which are specifically coded for multi-threaded execution!

I SO DID NOT EXPECT THAT!

My Q6600 from 2007 runs every game I have on top settings (last game I bought was Prototype). I just don't see any benefit to the consumer.

Re:Holy cow!

[dfghjk]

...because consumers only play games.

Re:Holy cow!

[L4t3r4lu5]

... and they totally run POV-Ray.

This is obviously not aimed at mom-and-pop checking their Bookface and watching a little iPlayer. Don't troll so obviously.

Re:Holy cow!

[Rallias Ubernerd]

The major advantage to the consumer buying a 6 core processor is that games capible of multi-threading will have an exponentially more advanced Artificial Intelligence. So, the games should be more of a challenge to everyone, not just the person who plays it all the time.

Re:Holy cow!

[kimvette]

Editing images is faster, especially with the megapixel race in the consumer, prosumer, and professional camera markets.

Editing video is not only becoming easy for home users, but is becoming almost realtime in terms of seeing results.

Embroidery and illustration packages improve a LOT.

oblig:
Also, on the typical home machine, this will reap a huge net gain in every day computing, Now four cores can be dedicated to botnet daemons/services and other malware, and one can run Internet Explorer and one core could run the totally ineffective Norton Internet Security suite. ;)

"and be less upgradable." is not true

[Visaris]

"and be less upgradable."

Not true. AMD's platform is much more forward compatible. AMD chips can now run DDR2 or DDR3 depending on what board it's in (Socket AM2/AM2+/AM3). That means that new AMD chips are compatible with 3 socket generations. Intel boards have nowhere near this broad socket and memory compatibility. Even in the same socket, a new chipset is typically required by Intel for new CPUs. This allows Intel to fake that their socket+platform had a compatibility life of 6+ years, when really, it was more like 1 and a half because they released 4 different chipsets with different support in that time frame.

If you're building your own box, or just want to upgrade later, AMD really gives you a much more flexible route. Here's an example of Intel's mess on their _current_ generation lineup: Core i7 runs on Socket 1156, while a different Core i7 runs on Socket 1366. Socket 1156 is not future-proof and will be dropped in the future. People buying those boards and CPUs might not even notice and will be s.o.l. after the very next generation. That's just silly. AMD's platform is the one with the sane upgrade path. And it's cheaper. I don't get all the AMD hate going around.

Re:"and be less upgradable." is not true

[Pojut]

And it's cheaper. I don't get all the AMD hate going around.

It isn't 1337 enough. Screw that, it's 1337 enough to run the games I want to play with my current configuration, that's all that matters.

Re:"and be less upgradable." is not true

[beelsebob]

Not true.
True

AMD2/2+/3 may have been compatible with each other, but then, so were Pentium 4s, CoreDuos and Core2Duos, all living on socket 775. At some point, a socket gets too old to support new CPUs, 1156, being a new socket still has some legs in it, it'll support *at least* the next generation of Core is, and probably the CPU design following Core i. AM2/2+/3 by contrast are coming to the end of their run. It's unlikely that AM3 will support more than the next one upgrade of the Phenom.

For reference, suggesting that a Phenom II on an old AM2 board with DDR2 is going to be anywhere near as fast as in i7 with DDR3 is a joke, especially considering *both* these CPUs *live* on memory bandwidth.

Re:"and be less upgradable." is not true

[0123456]

People buying those boards and CPUs might not even notice and will be s.o.l. after the very next generation.

How many computer buyers ever actually upgrade their CPU? 1%?

AMD's platform is the one with the sane upgrade path. And it's cheaper.

And it would cripple a 6-core or 8-core CPU with limited memory bandwidth on a motherboard originally designed for older CPUs with 2 or 4 cores. It also means that new AMD CPUs have to support both DDR2 and DDR3, which apparently limits their memory bandwidth even further (from what I've read, the DDR2 support in the memory controller prevents it from running at optimum performance with DDR3).

Seriously, I've never understood this 'but I can run AMD's top end new CPU on my six year old motherboard' stuff: sure, you can do it, but why bother?

Re:"and be less upgradable." is not true

[fucket]

You've got it all wrong, Socket 1337 isn't due out for two years.

Re:"and be less upgradable." is not true

[cynyr]

because memry bandwidth isn't much of an issue on AMD, the controller is on die, all it is a bunch of wires. Maybe my motherboard has enough on it for me, but i'm waiting for a board with USB3/(insert other feature here) on it. so i can make a 3/4 upgrade now, and then when a board comes out buy it and drop it in. instead of going hmm i need more CPU power, ooo that C2D looks nice, and it's 775 just like my P4, ohh wait, doesn't work in my current 775 board.

30 frames in FreeCiv

[managementboy]

with this one we could get 30 frames out of the html5/javascript version of freeciv!

Re:30 frames in FreeCiv

[Rallias Ubernerd]

Really? Really? You made an html5/javascript version of freeciv? A better goal would be to make freeciv for the GPU. There it would suffer an extreme boost of power. GPU's these days have something like 256 cores to do all there processing, and never use more then like 4 (are my numbers correct?) plus most have on-board RAM, allowing them to process data more often, reducing the impact that the speed of light has.

Re:30 frames in FreeCiv

[Jeremy Erwin]

Really? Really? You made an html5/javascript version of freeciv?

It exists.

Future CPU gain

[moteyalpha]

There was an article on connecting circuits with a solder that could be de-soldered with magnetic fields and it seems that the obvious future gain is reuse. If all they are doing is packing more cores in a package, the CPU should retain its value and if an effective method could be created to allow me to add new cores or delete cores that fail, then it would be just like memory. If somebody came up with a machine that could plug cores up to even 64x, it would seem that it would allow stability for twelve years. I helped design motherboard chip sets long ago and it seems that this isn't that big a trick. With current FPGA technology the interface logic could be reused with a reprogram.
It seems like a walk in the park compared to what we did 25 years ago with masked ASICs at 50K a pop. Just like the use of RAID, this should provide a market for the technical implementer.
It seems like opportunity for the manufactures too, as that need for speed could be translated into constant desire. If I had just one more core I could compile in 3,374 seconds instead of 4,809 and that would save me more time to read those really important /. articles I missed.

Re:Future CPU gain

[kimvette]

Then, processors had 250K or fewer transistors (not to mention other components) and 8 to 24 pins to communicate through, and ran somewhere between 700kHz and 8MHz (or 33MHz at the end of the 80s, which is later than the time you indicate), chip tracings were what, several microns, and were high voltage and low current compared to today's components. Also, heat sinks and fans were not required by most processors at the time. Even the mighty Motorola 68K and the i386 usually didn't sport heat sinks.

Now, processors and chipsets feature over one billion transistors (not to mention other components), with various components running between 300MHz and 4GHz, with many sections of chips running at 1.5v or less but requiring much more current, feature nm-scale circuit tracings, require careful design to prevent RFI, and will literally burn up within seconds if run without significant cooling in place.

Things are a bit more complex now. It's not like you can slap everything onto a simple, passive altair bus backplane and end up with a reliable AND high speed computer. Sure, there are blade computer options, as well as supercomputers which allow you to plug in additional CPU modules into a bus, but then you're talking a computer costing tens of thousands to millions of dollars; not something you'd see in a medium sized business or even many enterprises, and certainly not in the home. Heck, in most businesses you're sometimes lucky to get a PHB to approve a dual socket server with 4 core processors (8 way processing) or four-socket system, let alone big iron. Even in 25 user companies it's often difficult to get management to agree to a proper server. People always want cheap, cheaper, and cheapest.

To say
"it isn't that big a trick" was true back in the day. Now components need to be closer together, run at lower voltages, communicate at much higher speeds (frequencies), and be cheap.

Re:Future CPU gain

[moteyalpha]

Obviously you have some knowledge of the business. "Isn't that big a trick" is relative to the skill set. I designed with later technology and Gatorade ( Gate arrays ) was just the start. I am quite aware of what goes into the designs as I worked in semiconductor wafer fabrication, stuffing and somebody has to design the equipment and process that instantiates that hardware.
I am just speculating on what might be useful in the future. As a programmer also, it is now necessary to think about threads and interlock. The CPU mfg can't design in the software to create independent threads. That has to take place at a higher level.
I would like to see a more open hardware environment as I believe it allows the technology to advance more quickly. I would like to see some USB OpenGL, or USB CPUs and some means to have an expandable system if I am required to do threading in the software, I would like to do more threads as it becomes easier to sort out projects. It would be nice to have a fiber optic core CPU with a satellite bus, to resolve dependencies between the threads.
I also do bioinformatics and Blue Gene is the machine we use. I have used hardware from the first IBM in BAL ( and COBOL ) with page thrashing and job control language. I like RISC and liked the 680x0 series and find CISC and SIMD or MIMD a bit cumbersome.
I use what I can get, it just would like a little more flexibility in my design as I have the skill to use it and this has always been the case.
CPUs based on DNA should be available and perhaps I should focus effort to see that design is more flexible from the start.
It shocks me that video is generated as SVGA then converted to digital with a DAC to display on new monitors. A digital interface would save a lot of waste, but its all just sand anyway so either way is fine.
In the final analysis you are correct, it is what you can sell.

Just for Bragging Rights

[MrTripps]

If someone wants to blow $1000 on their e-peen, more power to them. My little i5 system ($600 total system cost) can run anything just fine.

can't take advantage?

[l3v1]

Some may say that the majority of applications can't truly take advantage of the resources afforded by a six-core chip capable of processing up to 12 threads.

Well, those "some" don't code complex stuff. Give it to me, I can put it to good use. I'd take a motherboard with 4 of these popped in any day as my work desktop (I'm dealing with massively parallel and highly computationally intensive stuff every day).

Re:can't take advantage?

[cynyr]

get a nvidia gpu and use CUDA?

Re:can't take advantage?

[petermgreen]

BTW the CPU reviewed here is a single socket device though there is a dual socket version (which afaict is even more expensive).

If you want more than two sockets you either have to wait for the 8-core chips with more QPI links supposed to be released soon or make do with core 2 generation stuff.

Parallel apps aren't everything...

[Angst Badger]

Every time this comes up, someone makes the observation that most apps aren't written to take advantage of multiple cores. That is, indeed true, but unless you're running MS-DOS, there's more to it. On the average Windows and Linux desktop installations, there can easily be twenty or so processes running before you start your first end-user application, and most users tend to have more than one app running at a time. While there is no substitute for purpose-built multi-threaded programs, it's not like those six cores will be sitting idle, especially under Windows, where you could throw an entire core or two at the OS and another couple at the two or three resident antivirus/malware programs that you need to have running to compensate for Windows' broken security model.

Granted, a lot of end-user apps spend most of their time sleeping, waiting for user input, but a sleeping process runs just as well on one core as on six. For users whose programs are actually doing something most of the time, multiple applications can take advantage of the additional cores even if they are themselves not multithreaded.

Re:Parallel apps aren't everything...

[Malc]

CPU usage on my two year old dual core laptop:

Processes: 118
System Idle Process: 81%
firefox.exe: 12%
X1.exe: 4%
OUTLOOK.EXE: 1%
System: 1%
vmware-authd.exe: 1%

Most things don't do much, including services. They just sit there. If I close Firefox, my system barely uses any CPU (and gains about 750MB of memory back).

What I would like this CPU for is AVC encoding...

Re:Parallel apps aren't everything...

[91degrees]

Really the point is that this isn't aimed at a typical desktop user. A lot of the applications that this will be used for will easily use 12 threads. I know our 4 core i7 is great for compiling and our project is relatively small. Probably pretty good for rendering as well.

Re:Parallel apps aren't everything...

[T-Bone-T]

I just upgraded from a 2.4GHz Pentium 4 to a Core i3-530. The difference is astounding. A typical DVD movie image would take my P4 about 8 hours to convert to an iPhone-compatible format whereas the i3 only takes 20 minutes. Sure, I got the cheapest Core series processor but I don't even know what do do with all that power. I could convert all my movies but that would only take the afternoon.

Re:Parallel apps aren't everything...

[cynyr]

I'd like a 4 or 6 core machine for doing things like play wow, libx264 encode, gentoo emerge, firefox with a flash based music player all at the same time without noticing. granted i can do a good chunk of that on my Athlon X2 but it requires some nicely set priories.

Additional Benchmarks

[Low Ranked Craig]

I think these kind of tests should start to include virtualization benchmarks. I'd really like to know, for example, how do VMWare, Virtul Box, Parallels, etc. benefit from these new processors?

Bang For The Buck Gaming Conclusion

[u64]

Bang For The Buck Conclusion, QuakeWars and Crysis fps

i7-975 . . . . (175.2/969.99 * 187.43/969.99) /2 = 0.017 FpsForEachBuck
i7-980X. . . .(196.9/999.99 * 234.24/999.99) /2 = 0.023
i7-870 . .. . (172.3/569.99 * 167.33/569.99) /2 = 0.044
Phe2X4-965 (128.8/194.99 * 112.33/194.99) /2 = 0.190
i5-750 . . . . (150.3/199.99 * 154.62/199.99) /2 = 0.290

As we go lower in cost we can get even more for each buck. The key is
to discover a sweet spot between too-high-costs and performance-too-slow-for-our-need.

(i7-980X 999.99$ is just a guess)

The internet is missing a site that combine a comprehensive cpu-chart real-world benchmarks
with current prices...

Re:Bang For The Buck Gaming Conclusion

[petermgreen]

I maintain that bang per buck of processors alone isn't what matters.

Either you have a job that has to run on one machine in which case it's a matter of whether time saving/better experiance is sufficiant to make up for the cost difference or you are looking at a number of machines working together in which case you want to look at bang per buck of the entire node (including the costs of power and rackspace).

We all know top end Intel processors have a horribly low bang per buck (even when you take whole system bang per buck) but that isn't what matters when comparing intel to AMD. What matters is who can give a bettter price for a given performance and/or who can give a better performance for a given price.

The internet is missing a site that combine a comprehensive cpu-chart real-world benchmarks
with current prices...
Agreed

Re:Bang For The Buck Gaming Conclusion

[ak3ldama]

I maintain that bang per buck of processors alone isn't what matters.

That's absolutely true... a site with user submissions for machine cost, or at least particular part cost, and performance given would be awesome. You would think that with just CPU/Motherboard/Ram/Video given you could find some amazing sweet spots.

Taking advantage of the resources ...

[Lemming Mark]

"Some may say that the majority of applications can't truly take advantage of the resources afforded by a six-core chip capable of processing up to 12 threads."

Well, before switching to Click-to-Flash mode I would quite happily have used 11 of those threads for Flash banner adverts spinning in a CPU-hogging mode and 1 thread for my useful applications. So I expect it will make things go faster even on the desktop! But that's not a good reason for wanting / needing more hardware threads!

Re:Fuck Multiple Cores

[zero0ne]

Why do you think these things are being released?

Just because?

If you release the multiprocessor hardware, after a few generations the majority of users will have multiprocessor PCs...

Once that is the case, THEN the programmers will follow.

Intel holds all the cards, but we're playing chess

[mykos]

All the computation I do is orders of magnitude faster on GPUs than CPUs. Furthermore, my graphics card also handles a lot of non-parallel tasks better than a CPU. I think we're seeing the waning days of Intel's processor dominance, unless they evolve their processor business into something else.

Virtualization Performance Questions

[multiplexo]

Has anyone here worked with KVM using libvirt to associate virtual cpus with physical cpus? I've always wanted to try this to see what the performance would be like.

Make Them Servers

[Gazzonyx]

My old parts still serve me well (usually better than their younger siblings, in fact!). My network router is an old Emachine PIII 600 MHz with a (you'll love this) 20Gig 5.25 Quantum BigFoot drive running PfSense. It sports 180MB of old PC100 RAM that was lying around. All three PCI slots hold a NIC (2 gigabit and 1 100mbit), and except for the lights dimming when the BigFoot spins up, it sips juice from a ~200 watt PSU. Not that I ever get to see the lights dim: current uptime 312 days.

My other old box, serving on its fifth or sixth tour of duty is a 900 MHz Duron (Socket A, 'cause I was able to put three CPUs through that socket duing upgrades) is my main server. It runs Subversion, Samba, JBoss, and anything else that I need to stay up reliably. Aptly named 'Slacker', it runs Slackware Linux and is of modest, but acceptable, speed for serving files, web pages and NTP. Ironically, this box's CMOS is shot and it loses time whenever it doesn't have power... therefore, it controls the UPS for itself and the router. Current uptime 291 days.

Both boxes are reliable enough that they've been sitting in a home made rack headless for over a year now without incident. The storage server that sits at the bottom of the rack is a much newer 64 bit Opteron. It has been nothing but trouble. Regardless of which OS runs on the hardware, and I've tried many over the last three years, (Slackware, Fedora, Sabayon, OpenSolaris) it will simply fall over. From chipset issues, overheating, blowing power supplies (currently on number 3, I keep an extra on the shelf these days), to dropping perfectly good drives from RAID because it suddenly can't see them, the box has been nothing but misery with a very expensive price tag.

Keep your old boxes, they'll outlast your new gear and enjoy the new rig they inherit when that new gear fails and you've got another empty chassis and another new motherboard that won't fit in it. The last days of one technology are always better than the first days of a new technology.

Fail.

[Gazzonyx]

He didn't say the first N prime or odd numbers greater than zero. If you're going to be clever, don't be stupid.

CPU power is never enough.

[Finite9]

No matter how fast your CPU is, it's never enough. Until my CPU can transcode a 2hr video from any format to any other in less than 5 minutes, then it won't be enough.