NVIDIA throws another punch, sez Larrabee would've been hot in 2006

You've got to love it, don't you? The prolonged hatefest between Intel and NVIDIA is continuing on today, with Andy Keane, general manager of the company's GPU computing group, delivering the latest blow. While speaking to reporters at the outfit's headquarters in Sunnyvale, California, he was quoted as saying that there is still "an incredible amount about Larrabee that's undefined." Furthermore, John Mottram chimed in by suggesting that "as [blogger and CPU architect] Peter Glaskowsky said, the 'large' Larrabee in 2010 will have roughly the same performance as a 2006 GPU from NVIDIA or ATI." We're beginning to wonder if these guys aren't just passing disses while sharing a cold one afterwards just to get attention, but being that it's more fun to envision suits from rival firms intensely angry with one another, we'll just keep believing this actually isn't a joke.

[Via Slashdot]

[substance90 @ Aug 25th 2008 6:28AM]

Nothing surprising here. Everybody with little hardware knowledge is aware that the raw calculational power of a modern GPU is far greater than a CPU`s.

[KarlW @ Aug 25th 2008 9:37AM]

It's not 'raw computational power' - GPUs beat CPUs at vector code, but CPUs still rock at scalar code.

Larrabee looks awesome - very parallel computing (initial models expected to have 24 or 32 x86 cores), and the flexibility and familiarity with the x86 architecture to call upon. And since it has so many cores, it has excellent parallelisation, and will rock at vector code, with each core being powerful enough that it can still do well at scalar code.

And real-world test prove it. It has a near linear FPS to number of cores relationship. Why is that cool? Well, because Intel's fabrication plants are some of the best. They're constantly moving to smaller and smaller chips, with less power consumption. So they'll be able to put more and more cores on each card. And when you look at the P4 -> Core2 -> Nehalem jumps, you can see how much Intel can improve each single core. More cores X Better cores = Awesome.

[egotman @ Aug 25th 2008 10:14AM]

Would this mean that the ray-tracing gut-blasting pre-rendered eye-candy that we see from hollywood (read scalar graphics) might actually be a reality in realtime form on a platform like larabee? Or at least a later successor?

You can do quite amazing things on vector graphics with GPUs these days, but some serious problems emerge that are done easily with ray-tracing and CPUs.

I have not read into the larabee architecture if it's just many CPUs doing vector work, but how I love to dream of that day where polygons can go the way of the dodo....

[honster @ Aug 25th 2008 12:55PM]

Nvidia headquarters is in Santa Clara, California... unless AMD sold to NV.....

[vcx @ Aug 25th 2008 4:05PM]

Btw... its John Montrym and not "Mottram". Did anyone even bother to check the comments on that slashdot article before posting this?
Here is what "really" happened:
http://news.cnet.com/8301-13512_3-10024280-23.html?tag=mncol;txt
I am beginning to believe that Engadget is the Fox news of the internet.

[loosely_coupled @ Aug 26th 2008 1:35AM]

That's a very shortsighted suggestion. Although your statement is true when comparing a modern GPU to a modern x86 CPU, Larabee is not a standard CPU. We are talking about 32-48 x86 cores running 512-bit SIMD instructions. Don't get me wrong, I'm certainly not an Intel fanboy, and I am particularly skeptical when it comes to anything they have to say about graphics processing given their long history of overpromising and under-delivering. But Intel is a very powerful, capable company and they are certainly NOT stupid. I don't think they would even bring the product to market if it was only competitive with 2006-era GPUs.

The bottom line is that this is a brand new, uncharted area in graphics processing, and no one really knows what to expect from it. If Intel put together a crack team of engineers and has been able to produce excellent D3D/OpenGL drivers, this product could really be a monster. It's real benefits will be in GPGPU capabilities aka stream processing. It should prove easier, or at least more flexible and adaptable to take advantage of the parallel processing capabilities of a many-core x86 CPU than modern GPU architectures, although both Nvidia with CUDA and ATI with Apple's OpenCL will be working their hardest to negate any inherent advantage.

[Cash @ Aug 25th 2008 6:44AM]

oh no he didn't. *snap*

Seriously, will you bunch just kiss and make up do I can stop worrying whether or not the next nehalem mobi will support SLI? You two keep fucking around, and people are really gonna start buying products from both of your competitor.

[Crayola @ Aug 25th 2008 7:15AM]

nVidia can talk dirty all they want, can't hide the fact that Intel is a fab process tech supremo, all they have to do is turn the knob shrink the process, do it brute force with plenty of cash. Intel after Athlon is a renewed company with extremely good focus. Whereas nVidia can only partner with TSMC on a generic process, keep their fingers cross, hedge for the right bet.

[rudefyet @ Aug 25th 2008 7:19AM]

I'd be willing to bet Intel and the Larrabee wouldn't leave me with a laptop with a defective GPU.

[Matt @ Aug 25th 2008 9:19AM]

Are you telling me you DON'T love running your fan 24x7?

Seriously, those in glass houses should not throw stones. I like Nvidia GPUs as much as the next guy, but when an entire model of chip is defective, you shouldn't be slinging insults. I've never had an Intel proc die because they used the wrong material for ALL of them.

[waffles @ Aug 25th 2008 10:50AM]

not like intel hasn't been down the path before. you're in the game long enough, you're bound to screw up somewhere

http://en.wikipedia.org/wiki/Pentium_FDIV_bug

[Dr Zoidberg @ Aug 25th 2008 7:25AM]

Gotta love the hate.

[Beastage @ Aug 25th 2008 7:35AM]

nVidia are doing what AMD tried to do by talking too much before Intel bitchslapped them.

nVidia should shut up because they have inferior products in their entire range, their chipsets are unreliable, their gpus cost too much while being outdone by radeon counterparts...

So nVidia shut up or back up your claims .

[Zeus.:God @ Aug 25th 2008 10:31AM]

What's this crap about their chipsets being unreliable? Every Nvidia based board I have ever owned has been fantastic, and I've never had a single problem with them.

I also don't see where you're getting this crap about their GPU's being weak in comparison to ATI's. Nvidia still has the highest powered, single GPU cards out, and while they were priced high initially, their prices are steadily dropping. The only thing ATI has on the big N is the new 4870X2, which will likely be outdone in the near future... Now, I'm not saying that ATI cards aren't good or anything, because they are- they're great, and they are a great value, but to say that they're outdoing Nvidia is a plain lie.

Nvidia, aside from their defective laptop chips, have more than enough room to talk with all that they have done for the computer world in recent years, and whether or not their claims are just, they still have every right, just as any other company, to make them.

[Ike turner @ Aug 25th 2008 7:36AM]

Yeah, just like Unified Shaders, when ATI introduced them right ? Oh What do you know! Unified shaders is what's hot 2 years later when nVidia comes out with a unified shader GPU. They always do this when They are geting their ass gaped.

[chansthename @ Aug 25th 2008 7:37AM]

Well if this talk drives prices down thats ok with me. But first I think Nvidia should fix their own GPUs before attacking Intel (who
haven't got a whole series defective and arn't willing to back up their customers)

[Tired_ @ Aug 25th 2008 7:38AM]

Of course there's a lot that's undefined to nvidia. It's not out yet. Were they expecting Intel to share with their competitors?

[broli @ Aug 25th 2008 7:40AM]

This is going to bite nvidia back hard. In the short run nvidia is right but in the long run nvidia is screwed. Full x86 support on a gpu is a dream for game developers, artists and researchers. Game developers can write their own custom render pipeline where as artists can fully use the cores in rendering processes and researchers can use any language they like in their heavily threaded research. Nvidia's cuda is a joke compared to x86 as with the latter you're free to choose any language you're already used to.

This seems for like nvidia is trying to brush off its intimidation by trying to convince itself it's all oke.

[DarkSi @ Aug 25th 2008 8:49AM]

When is someone going to make Nvidia shut up with a comment like "Well, at least Larrabee doesn't use over 140w when idle, require 2 discreet (and proprietary) power connections on top of the power it draws from the board, and have a cooling system akin to a Raptor fighter jet."

Seriously? Do they believe the BS they spit out? They're doing what every company does when they're scared - bashing the competition. Grow up, Nvidia, or you're going down the toilet... Well, that may happen anyway, but at least leave some people to respect you until the very end.

[whakojacko @ Feb 23rd 2009 9:34PM]

wtf?
Rumors put Larabee's power draw at well over anything made by NV, and would thus equire an even more absurd cooling system. Nvidia's modern gpu's dont use anywhere near 140W idle, and the pci express power connection is the same non-proprietary one that ATI uses.

Now I agree that NV is talking a lot of hype and best shut up, but at leas get your facts right.

[Brzt @ Aug 25th 2008 8:55AM]

Is this their way of trying to get the x86 license from Intel? :)

[Mam00th @ Aug 25th 2008 9:07AM]

Having read Peter Glaskowsky's blog on Cnet, I can tell you he is speaking out of his ass... He base everything he says on random assumption.

I'm just wondering one friggin thing: how the fuck do Nvidia architech know about Larrabee's performances when they haven't even been disclosed?

[Jin Saotome @ Aug 25th 2008 9:30AM]

Alright nVidia, instead of throwing insults around, how about releasing a new line of GPUs that doesn't have absurdly high power consumption and good performance? Or, at least durable products with a great price/performance ratio...

[Ihar `Philips` Filipau @ Aug 25th 2008 10:14AM]

If anything, Intel has enough influence on market to establish universal standard for GPU/CPU integration.

nVidia is obviously tries to keep all its eggs in its own basket.

Intel is obviously sending signal that GPU interfaces are going to be commodity. It doesn't matter that tech is from 2006. What matters is that now more companies would be able to play in the market. Intel, after making something ubiquitous, traditionally "outsources" most of work on keeping platform up to other manufacturers. All what is important to Intel is to set a direction - the direction in which Intel can play role and have share of future profits.

[black @ Aug 25th 2008 10:48AM]

Didn't NVIDIA say they wanted to get in the CPU market? Anyway, NVIDIA shouldn't be the ones talking since Intel actually has working processors at this point. Honestly I just want a CPU/GPU combo that I can use so that way I wont have to constantly have to update the graphics card every time a game comes out, and Larrabee is looking to be it...I hope

[Zeus.:God @ Aug 25th 2008 10:49AM]

What are you talking about? They already do all of those.

[Zeus.:God @ Aug 25th 2008 10:50AM]

That was in response to Jin Saotome.

[Ian Kennedy @ Aug 25th 2008 11:37AM]

Intel graphics have been notoriously lame for years. They have to prove themselves. Every new integrated solution they come out with is claimed by them to enable all these cool scenarios, but in the end are just garbage you replace with a dedicated GPU from AMD or NVIDIA.

Basically the Intel needs to put up or shut up and the guys form AMD and Nvidia are basically saying "STFU newb."

[Ram @ Aug 25th 2008 5:05PM]

Companies who cannot deliver on their products, or do their products with flaws and get their butt kicked in the stock market should be first looking upon themselves as to what is wrong, instead of commenting on Intel. Nvidia, when was the last time u checked ur own stock price?

[futurepastnow @ Aug 25th 2008 7:00PM]

So Nvidia had a many-core, multi-teraflop x86 GPU two years ago? Man. They should have sold that thing.

[wonkydonkydotnet @ Aug 25th 2008 7:31PM]

I guess, but isn't Nvidia's gambit here a bit disingenuous?

I mean, a GPU is closer to a DSP than a CPU.
DSPs do ~1 thing very very well and CPUs do Many things not so fast. -There's a tradeoff.

[Cray @ Aug 26th 2008 12:08AM]

Thank you Engadget for being juvenile and petty.