Intel's Larrabee Chip Needs Reality Check?

you think intel cant write a direct x/ogl compiler? you do know intel is very very very good at compilers

First of all, the compiler has nothing to do with it. The programming model is different. Intel themselves have "warned" developers that they need to change their way of thinking to program coming products (while claiming the same change is a bottleneck for other massively parallel devices, but that's another discussion). Read more.

http://blogs.intel.com/research/2008/06/unwelcome_advice.php

Second, if you think hordes of developers are going to migrate to DirectX or OpenGL to write high-performance applications, you are wrong. GPGPU computing is starting to gain momentum in the mainstream thanks to C-style parallel languages like CUDA and Brook. Before them GPGPU was a pain in the ass because the only way to program the GPU was through DirectX or OpenGL.
 
First of all, the compiler has nothing to do with it. The programming model is different. Intel themselves have "warned" developers that they need to change their way of thinking to program coming products (while claiming the same change is a bottleneck for other massively parallel devices, but that's another discussion). Read more.

http://blogs.intel.com/research/2008/06/unwelcome_advice.php

Second, if you think hordes of developers are going to migrate to DirectX or OpenGL to write high-performance applications, you are wrong. GPGPU computing is starting to gain momentum in the mainstream thanks to C-style parallel languages like CUDA and Brook. Before them GPGPU was a pain in the ass because the only way to program the GPU was through DirectX or OpenGL.

i thought u were refering to game dev

for application dev, its x86.

...
 
I agree! Competition is good for the consumer all the way around...it spurs innovation and keep pricing competitive.

This is a little bit naive to say though, because this isn't just Intel making a GPU. They're trying to change the game by changing how games are made, which will fragment an already struggling PC gaming market. Now developers have to choose whether to make an Intel larrabee based game (which has an uphill battle for marketshare, which is what developers need to consider this) or a normal DX/OGL based game for NVIDIA/AMD cards.

And yes they claim they'll support DX and OGL but performance won't be as good as a game coded natively for their arch/api. Developers have a hard enough time supporting hardware specifics for both AMD and NVIDIA to give the end-users the best experience possible.

It will be interesting to see how it plays out.
 
the thought here is intel will provide software tools capable of making the performance that good. remember, compilers again. and this is essentially compiling the hardware to fit the software.
 
i thought u were refering to game dev

Sorry, I thought Knight was talking about general purpose computing. Of course Intel would provide DirectX and OpenGL interfaces to the hardware, so it doesn't make sense to talk about the difficulty of programming a device that already supports the standard APIs. The only "difficult" model IMO would be GPGPU.

for application dev, its x86.

...

What does "it's x86" mean? If you're implying Larrabee is better for general purpose because it's x86... :rolleyes: It means nothing when you write code in a high-level language and use a compiler. Larrabee might be better due to other reasons - the cache architecture, memory interface, inter-core communication - but that's all speculation until they get it out the door.
 
i mean its x86, and intel knows how to write compilers for it. and in this case, the compilation is being done to the hardware to better fit the software
 
This is a little bit naive to say though, because this isn't just Intel making a GPU. They're trying to change the game by changing how games are made, which will fragment an already struggling PC gaming market. Now developers have to choose whether to make an Intel larrabee based game (which has an uphill battle for marketshare, which is what developers need to consider this) or a normal DX/OGL based game for NVIDIA/AMD cards.

+1

Intel will have a tough time pulling devs along if they have to make a different version of the game to get good performance on Larrabee. They'll fall back to DX/OpenGL to save development time and money, not to mention reach a larger customer base.
 
i mean its x86, and intel knows how to write compilers for it.

And NVIDIA knows how to write a compiler for it's hardware, and AMD knows how to write a compiler for it's hardware, and so on. You can pontificate all you want about compiler expertise - it means nothing until we see the same app running on an NVIDIA card, an AMD card, and Larrabee.

and in this case, the compilation is being done to the hardware to better fit the software

huh?
 
hmm i was just giving this some thought... intell is saying that it is going to be different to program for this chipset, (yet it will still support ogl/Dx) which are both rasterizer style renders... without really knowing whats up there sleeves, and speculating that they are going for massive parellel cpu style rendering, im beginning to wonder if they are going to try their hand at real time raytraceing... aside from writing a totaly new rendering method (not impossible, and both seem unlikely but what the hell), this would seem like the only other way to do things... thoughts?
 
What terrible reasoning. It doesn't matter if Nvidia had a billion years to perfect their technology. Anyone could come along with something new and better.
I'm sure you have something new and better right now then? Experience must count for nothing in your world. Lets take something easily accessible like making a pizza. Sure if I'm very lucky I might be able to make the best pizza in the world on my first try, but its unlikely. I could also iterate and work on making pizza's my entire life and never come close.. but the chance would go up with each iteration.

Take nvidia for example, when they first tried 3d graphics they worked with quads, now no one works with quadrangles. We know(or atleast experience tells us) that triangles are better to form 3d graphics with.
 
yeah but intel? is that good for the longterm competition... intel practicaly run the show in the cpu/chipset market... do you really want... intel cpu, intel ram, intel board, intel gpu, intel case, intel psu, intel mouse, intel keyboard etc etc etc??

Yes.. sort of... Intel CPU, Intel chipset, Intel GPU will be what most people will want.

Larrabee means the death of the current fixed function GPU's. Many people have not grasped that or what that means. The end result is way less complication, less bugs, many many more capabilities, much more compatibility and more balanced speed across all types of graphics applications.
 
I think most people are forgetting that Larrabee isn’t really a GPU.

It's full of x86 cores and its DirectX, OpenGL support will be based on API’s and compliers etc two things that Intel is arguably the best in the world at.

Yes, point is that even if it fails as a GPU, it could still be a great device for supercomputers, much like the Tesla cards that nVidia is now selling, and AMD has a similar card aswell (FireStream? can't recall the exact name).

Three other things I'd like to point out, regarding the 'analyst' and his comments:
1) Experience doesn't mean all that much in the land of GPUs. The whole architecture has been reinvented many times since the first 3dfx cards. As experienced as 3dfx was with the first generation of cards, they went out of business as soon as the focus shifted to other areas, where nVidia was better equipped.

2) GPUs are getting ever closer to CPUs in many respects. nVidia's Cuda support proves that a GPU is now capable of many tasks that are not directly related to 3d graphics, such as physics, folding@home or video transcoding. GPUs are moving towards CPU-like programmability and Intel intends to meet nVidia half-way. It's hard to say who's stepping on what turf really.

3) Intel never claimed they will outperform AMD or nVidia, at least not with the current Larrabee project (although obviously they do hope for this sometime in the future). They were very specific in saying their first Larrabee-based cards would be aimed at the mid-end market at the time of release.
 
Yes, point is that even if it fails as a GPU, it could still be a great device for supercomputers, much like the Tesla cards that nVidia is now selling, and AMD has a similar card aswell (FireStream? can't recall the exact name).

Three other things I'd like to point out, regarding the 'analyst' and his comments:
1) Experience doesn't mean all that much in the land of GPUs. The whole architecture has been reinvented many times since the first 3dfx cards. As experienced as 3dfx was with the first generation of cards, they went out of business as soon as the focus shifted to other areas, where nVidia was better equipped.

2) GPUs are getting ever closer to CPUs in many respects. nVidia's Cuda support proves that a GPU is now capable of many tasks that are not directly related to 3d graphics, such as physics, folding@home or video transcoding. GPUs are moving towards CPU-like programmability and Intel intends to meet nVidia half-way. It's hard to say who's stepping on what turf really.

3) Intel never claimed they will outperform AMD or nVidia, at least not with the current Larrabee project (although obviously they do hope for this sometime in the future). They were very specific in saying their first Larrabee-based cards would be aimed at the mid-end market at the time of release.

Yup, Larrabee is to get the foot in, the real push will be Larrabee II...and you just have to listen to NVIDIA's rethoric before you realize that they are worried....NVIDIA using more PR to bash Intel than AMD is a powerfull pointer...and NVIDIA can't "muscle" Intel...it's more the other way around.
 
Back
Top