GT300 (NV60) vs Larrabee

[AirRaid]

GT300 (I call it NV60) will be a next generation DirectX 11 / Shader Model 5 GPU, the first completely new architecture since the NV50 / G80 / GeForce 8800. The GT300 is due out in Q4 2009. This will be Nvidia's answer to Intel's Larrabee GPGPU which will use 12/16/24/32/48 cores, (low-end to high-end) in the first generation (2009-2010).

I expect GT300 to out-perform Larrabee in current DX10 and future DX11 games that use traditional rasterization techniques.

Larrabee will be capable of going far beyond DX11 in capability/features. If and when developers ever take advantage of that, writting directly to Larrabee, or mapping future next-next gen DX12 functions to Larrabee, that's when Larrabee should start to outshine Nvidia (and AMD) DX11 hardware including GT300.

I don't expect even highend Larrabee (32 ~ 48 cores) to match the performance of Nvidia GT300 or AMD R8xx in games. With that said, I do expect Larrabee to offer solid gaming performance on all DX9, DX10 and future DX11 games, at least an order of magnitude better than Intel Integrated Graphics which cannot even run many popular games. I'd expect the mainstream Larrabees (16-24 cores) to run close to the performance of current midrange Nvidia/AMD products, and the 32-48 core Larrabee to be close to Nvidia/AMD upper midrange/performance products, but fall short of the highend, nevermind the extreme highend, and nevermind Nvidia/AMD DX11 offerings.

All of Nvidia's DX11 midrange & mainstream products derived from GT300 should outperform any version of Larrabee. I don't think Larrabee will touch anything except the lowest-end/value/budget DX11 products. I might be wrong, but that's how I see it.

Of course things would change once developers start to take real advantage of Larrabee, not writting to DX10/DX11 but writting "to the metal" for Larrabee. Yet that's only going to happen if Larrabee takes massive marketshare away from Nvidia/AMD, which is not likely in 2010.

Larrabee gets more & more interesting (and more dangerous for Nvidia and AMD) as more time goes by.

If Larrabee is even somewhat successful, then Nvidia and AMD should worry about Larrabee2 which is in development now.

[NaitoSan]

sweet. i've been looking to upgrade my ol 8800 gtx. i'm not so true about intel's larrabee but i'm going to wait and see. i'm more curious about price / performance ratio for both, especially larrabee.

[josiahsuarez]

if anything the first version of larrabee should offer good price/performance ratio, as intel has to establish themselves in this market. at least I hope so! I wouldn't be willing to pay a premium for unproven technology. if they can offer a compelling value proposition though I might bite.

[walterman]

Intel needs somebody like John Carmack (or Michael Abrash :)) to write a 3D engine to exploit all the Larrabee features.

I also think that Intel wants to fight CUDA. 32 cores x 16 single-precision float SIMD per core x 2 FLOPS (fused multiply-add) x 2GHz per core = 2 TFlops.

[Bman212121]

I don't think GT300 and Larrabee need even be in the same sentence. :lol:

[nekrosoft13]

i hope larabee fails. with dx and opengl we don't really need another api.

[Ninja Prime]

i hope larabee fails. with dx and opengl we don't really need another api.

I'm not sure you know what you're talking about, it uses DX/openGL.

[Ninja Prime]

Intel needs somebody like John Carmack (or Michael Abrash :)) to write a 3D engine to exploit all the Larrabee features.

I also think that Intel wants to fight CUDA. 32 cores x 16 single-precision float SIMD per core x 2 FLOPS (fused multiply-add) x 2GHz per core = 2 TFlops.

Wish granted:

"The team working on Larrabee is separate from the Intel GMA team. The hardware is being designed by Intel's Hillsboro, Oregon design team, whose last major design was the Nehalem. The software and drivers are being written by a newly-formed team. The 3D stack specifically is being written by developers at RAD Game Tools (including Michael Abrash)."

http://en.wikipedia.org/wiki/Larrabee_(GPU)

;)

[nekrosoft13]

I'm not sure you know what you're talking about, it uses DX/openGL.

yes it does, and it can also use diffrent api, that is if developers start using it

[bob saget]

i thought larabee was for extremely low end.
Oh well, you learn something new every day.

[CaptNKILL]

i thought larabee was for extremely low end.
Oh well, you learn something new every day.

Yeah, I just figured that out a couple weeks ago.

Larrabee is basically a multi-core x86 CPU built to run 3D rendering software. The more cores they add, the faster the software will run, meaning a faster GPU. They could integrate one or two GPU cores anywhere they want (motherboards or even CPUs) and then sell dedicated cards with dozens of them onboard to compete with the high end.

Its a pretty neat idea, but its so radically different from other GPUs that it seems unlikely that it'll set any performance records the first time around. Knowing Intel though, it'll at least be decent. They've been in this business a long time, I don't think they would have let it get this far if it wasn't a realistic alternative to a standard GPU. If a lesser known company was developing this I wouldn't be too optimistic.

[josiahsuarez]

I wonder what they will end up calling Larrabee once they drop the code name. "Graphics Media Accelerator" always sounded kind of boring.

[XDanger]

I wonder what they will end up calling Larrabee once they drop the code name. "Graphics Media Accelerator" always sounded kind of boring.

"Deliverance Engine"

[josiahsuarez]

how about PeForce (cuz of all the Pentium cores)

[AirRaid]

i thought larabee was for extremely low end.
Oh well, you learn something new every day.



No, Intel's current Integrated Graphics are for extremely low-end and Intel's future Integrated Graphics chipsets, in motherboards and in CPUs will also be for extremely low end.


The 'manycore' Larrabee chips, with 10 or more cores, upto 64 cores, will be discrete add-on cards for the mainstream/midrange, performance mainstream, highend and perhaps even extreme highend markets--The same markets that Nvidia and AMD target with their ranges of GPUs.


Probably the only Larrabee chips that'll targetthe very lowend will be the ones that have fewer than 10 cores (thus, they won't be manycore). Those will probably be integrated on motherboards and later, directly into future CPUs including the next-gen Sandy Bridge. This will probably represent the next generation replacement for Intel's crappy Integrated Graphics.

[AirRaid]

how about PeForce (cuz of all the Pentium cores)

hahah, funny.

[XMAN52373]

Wish granted:

"The team working on Larrabee is separate from the Intel GMA team. The hardware is being designed by Intel's Hillsboro, Oregon design team, whose last major design was the Nehalem. The software and drivers are being written by a newly-formed team. The 3D stack specifically is being written by developers at RAD Game Tools (including Michael Abrash)."

http://en.wikipedia.org/wiki/Larrabee_(GPU)

;)

Thats wierd, I didn't know Intel was into makes game engines. Also didn't know John Carmack worked on Intel design team.
:confused:

[Logical]

I just hope Intel price's Larrabee reasonably, maybe then we will see Nvidia drop there prices for competition.

Everyones a winner.

[ViN86]

You all honestly believe Larrabee will outperform Nvidia's offering? :wtf:

[CaptNKILL]

You all honestly believe Larrabee will outperform Nvidia's offering? :wtf:

Who are you talking to? No one said that. :lol:

[ViN86]

Who are you talking to? No one said that. :lol:
/me thinks I misinterpreted this statement...

GT300 (I call it NV60) will be a next generation DirectX 11 / Shader Model 5 GPU, the first completely new architecture since the NV50 / G80 / GeForce 8800. The GT300 is due out in Q4 2009. This will be Nvidia's answer to Intel's Larrabee GPGPU which will use 12/16/24/32/48 cores, (low-end to high-end) in the first generation (2009-2010).

I expect GT300 to out-perform Larrabee in current DX10 and future DX11 games that use traditional rasterization techniques.

Larrabee will be capable of going far beyond DX11 in capability/features. If and when developers ever take advantage of that, writting directly to Larrabee, or mapping future next-next gen DX12 functions to Larrabee, that's when Larrabee should start to outshine Nvidia (and AMD) DX11 hardware including GT300.

I don't expect even highend Larrabee (32 ~ 48 cores) to match the performance of Nvidia GT300 or AMD R8xx in games. With that said, I do expect Larrabee to offer solid gaming performance on all DX9, DX10 and future DX11 games, at least an order of magnitude better than Intel Integrated Graphics which cannot even run many popular games. I'd expect the mainstream Larrabees (16-24 cores) to run close to the performance of current midrange Nvidia/AMD products, and the 32-48 core Larrabee to be close to Nvidia/AMD upper midrange/performance products, but fall short of the highend, nevermind the extreme highend, and nevermind Nvidia/AMD DX11 offerings.

All of Nvidia's DX11 midrange & mainstream products derived from GT300 should outperform any version of Larrabee. I don't think Larrabee will touch anything except the lowest-end/value/budget DX11 products. I might be wrong, but that's how I see it.

Of course things would change once developers start to take real advantage of Larrabee, not writting to DX10/DX11 but writting "to the metal" for Larrabee. Yet that's only going to happen if Larrabee takes massive marketshare away from Nvidia/AMD, which is not likely in 2010.

Larrabee gets more & more interesting (and more dangerous for Nvidia and AMD) as more time goes by.

If Larrabee is even somewhat successful, then Nvidia and AMD should worry about Larrabee2 which is in development now.
oops :o

[hell_of_doom227]

Is it possible that NV60 is Dual Core solution?

[CaptNKILL]

Is it possible that NV60 is Dual Core solution?

You mean like two GPUs on a single PCB? I don't know why they'd bother. Unless they totally change the way SLI works it'd still be GX2-style card, just thinner. It'd pump out twice as much heat and require twice as much memory and bandwidth compared to a single GPU card. Not exactly cost effective.

[pakotlar]

You mean like two GPUs on a single PCB? I don't know why they'd bother. Unless they totally change the way SLI works it'd still be GX2-style card, just thinner. It'd pump out twice as much heat and require twice as much memory and bandwidth compared to a single GPU card. Not exactly cost effective.

no it wouldn't. it would only pump out twice as much heat if it was 2x the single gpu config. but in the case that it is not cost effective to do one 600mm^2 die then two die's of half the performance is a good solution.

also even in current dual-gpu solutions, bandwidth is aggregate... its actually one of the biggest "freebies" of sli. the 2x as much memory is true, but all fingers point to shared, pooled memory configurations for next gen dual-gpu configs. also two smaller dies can actually be more yield-effective, because the chances that one 1.4B (for example) die is going to have life-threatening problems is significantly greater than the chances that one 750M transistor die will have (assuming same architecture). and each die is cheaper, so one goes bad, either resell it as a lower-grade part (redundancy) or ditch it, but at least you won't have to ditch the entire thing.

really badly worded but I hope it makes sense. i feel like i'm on drugs or something. you ever have one of those days youre totally in a fog? i took a stats quiz today, and felt like it was a dream.

[CaptNKILL]

no it wouldn't. it would only pump out twice as much heat if it was 2x the single gpu config. but in the case that it is not cost effective to do one 600mm^2 die then two die's of half the performance is a good solution.
Well yeah, it'd only pump out twice as much heat as a single GPU config using the same GPU. That's what I said. :lol:

It could be more cost effective but having to use twice as much memory and build a card that is twice as complex probably hurts that a bit.


also even in current dual-gpu solutions, bandwidth is aggregate... its actually one of the biggest "freebies" of sli. the 2x as much memory is true, but all fingers point to shared, pooled memory configurations for next gen dual-gpu configs. also two smaller dies can actually be more yield-effective, because the chances that one 1.4B (for example) die is going to have life-threatening problems is significantly greater than the chances that one 750M transistor die will have (assuming same architecture). and each die is cheaper, so one goes bad, either resell it as a lower-grade part (redundancy) or ditch it, but at least you won't have to ditch the entire thing.

really badly worded but I hope it makes sense. i feel like i'm on drugs or something. you ever have one of those days youre totally in a fog? i took a stats quiz today, and felt like it was a dream.
I'll admit, I don't know squat about how a GPU is designed but how can the bandwidth be effectively doubled if the same data has to be sent to two sets of memory? If it is doubled, then so is the amount of data being sent over each memory bus, which negates any possible benefits.

I was always under the impression that the only thing SLI was meant for was doubling the processing capabilities, not necessarily anything related to memory.