Sign in with
Sign up | Sign in
Your question
Closed

Nvidia to Hit the x86 CPUs With CUDA Capability

Last response: in News comments
Share
September 22, 2010 12:11:18 PM

Whats the point of CUDA without a GPU? Multi core or not.
September 22, 2010 12:13:08 PM

Quote:
Whats the point of CUDA without a GPU? Multi core or not.


Well now you can run CUDA code WITHOUT the need for a Nvidia graphics card and will also allow them to compete with OpenCL and Directcompute.

And maybe it's cheaper to use what you have already (example in this case, a Supercomputer with say.. 100 CPU's), and would be cheaper to simply use those 100 CPU's instead of spending more cash on GPU's.

it's all in the text.
Related resources
Can't find your answer ? Ask !
September 22, 2010 12:21:20 PM

They've got the best hardware, now they need the whole software community to be able to lend a helping hand. It's okay, they plan to have a GPU that can run without a CPU two steps ahead. A greater CUDA crowd will be beneficial.
September 22, 2010 12:22:06 PM

joytech22 allow them to compete with OpenCL and Directcompute.


Except that OpenCL and Directcompute are compatible with all GPUs. CUDA is useless without GPU acceleration.
September 22, 2010 12:22:08 PM

Quote:
Well now you can run CUDA code WITHOUT the need for a Nvidia graphics card and will also allow them to compete with OpenCL and Directcompute. Think about it.. it's all in the text.


CUDA was toted as being nvidias answer to give exceptional processing power over x86 after Jen-Hsun Huang bashed it for so long.

Now that port THERE pride and joy to the thing they bagged for so long?
September 22, 2010 12:26:20 PM

What i meant was, (When) CUDA is (now) supported on CPU's as well as GPU's it allows more people to use the language without the need to spend thousands, it's just a money saver for some and allows others to mess around with CUDA without needing strict requirements, it should work, just not as fast.
September 22, 2010 12:44:57 PM

CUDA in also happens to be the best. The bees knees. The ducks nuts. in AU anyways
September 22, 2010 12:45:39 PM

joytech22Well now you can run CUDA code WITHOUT the need for a Nvidia graphics card and will also allow them to compete with OpenCL and Directcompute.And maybe it's cheaper to use what you have already (example in this case, a Supercomputer with say.. 100 CPU's), and would be cheaper to simply use those 100 CPU's instead of spending more cash on GPU's.it's all in the text.


4chan made something called tripper for cuda. it runds tripcodes in cuda, arguably the best way to get trip codes you want. now a single core cpu can do i believe 1-5 million trips a second, an i7 920 can do i think 22 million, a gtx285 is capable of almost 2 billion and if its not faked i have seen numbers up to 15 billion but i know for a fact that this is the BEST use of the gpu in practice as a gpgpu. without gou slow, with gpu fast.

point being that 100cpus are outdone by a quad sli, if you have the ability to get 100cpus, just get 4 cpus.
September 22, 2010 12:46:24 PM

Is it me or what they are saying is that is posible to run CUDA on ATI/AMD also ? :) )
a c 106 à CPUs
a b Î Nvidia
September 22, 2010 12:46:47 PM

Will it run if an AMD GPU is in the system?
September 22, 2010 12:49:18 PM

For a developer this is good news, instead of a platform specific use it can be deployed everywhere rather than just on nvidia gpu's. Hardware wise its also a great idea, accelerate the code if you want to by adding a gpu, if not it will run but not at peek performance!
September 22, 2010 12:52:04 PM

alidan4chan made something called tripper for cuda. it runds tripcodes in cuda, arguably the best way to get trip codes you want. now a single core cpu can do i believe 1-5 million trips a second, an i7 920 can do i think 22 million, a gtx285 is capable of almost 2 billion and if its not faked i have seen numbers up to 15 billion but i know for a fact that this is the BEST use of the gpu in practice as a gpgpu. without gou slow, with gpu fast. point being that 100cpus are outdone by a quad sli, if you have the ability to get 100cpus, just get 4 cpus.


But it says CPU's with CUDA capability.... I'm not getting it.

If it's faster on the CPU, it renders the GPU pointless.

If it's slower on the CPU, why use it?
a c 131 à CPUs
September 22, 2010 12:59:50 PM

jtt283Will it run if an AMD GPU is in the system?

You're thinking of PhysX. Wrong pun.
September 22, 2010 1:04:53 PM

stridervmBut it says CPU's with CUDA capability.... I'm not getting it.If it's faster on the CPU, it renders the GPU pointless.If it's slower on the CPU, why use it?


exactly my point, for the cpu we can use normal code, and either Microsoft or opencl and get all gpus to help. it still makes cuda over all pointless.
September 22, 2010 1:05:16 PM

bmaddWhats the point of CUDA without a GPU? Multi core or not.


OpenCL is meant to make programming parallel programs easier, whether that means a graphics card or a bunch of traditional processors. I'd imagine that nVidia is looking to compete with that model now, rather than focusing only on programming parallel programs for graphics cards.
September 22, 2010 1:10:45 PM

bmaddWhats the point of CUDA without a GPU? Multi core or not.


So you are not programming 2 ways in 1 application to add the CPU's power to the raw CUDA crunching.

It's also a way to learn CUDA without needing Nvidia hardware or a way to develop apps without needing Nvidia hardware. (I would assume you would deploy the finished app to CUDA hardware though)

OpenCL is owned by apple.
Honestly I trust Nvidia more from a "Nazi control freak over our IP" perspective.
September 22, 2010 1:11:15 PM

All the arguing with Intel and now CEO Jen-Hsun Huang is enjoying intel Kool-aid.
a c 106 à CPUs
a b Î Nvidia
September 22, 2010 1:19:11 PM

Enzo MatrixYou're thinking of PhysX. Wrong pun.

Actually, I was thinking of nVidia's practice of not getting along. Disabling features would be consistent with their past practices of doing that (e.g. with PhysX, as you point out) if certain competitors' equipment is in your system.
a b à CPUs
a b Î Nvidia
September 22, 2010 1:19:55 PM

Hmmmm I hope that it will run on IA64.
September 22, 2010 1:33:59 PM

here goes, where is PhysX?
September 22, 2010 1:39:32 PM

nforce4maxHmmmm I hope that it will run on IA64.


Do you still use IA64? I remember that netburst era stuff was awful on cpu intensive workloads. Hell, I can`t even find the ia64 version of 2003 server anymore.
September 22, 2010 2:02:29 PM

I would like to see this mean proper cpu optimizations for physx so you do not need a gpu to run it efficiently. Like getting it off the ancient x86 instructions and running it using the much faster SSE properly
Anonymous
a b à CPUs
September 22, 2010 3:14:38 PM

for crying out loud will they just kill CUDA off and adopt OpenCL, then we can truly see nvidia vs AMD GPGPU
September 22, 2010 3:24:08 PM

Maybe this is just the easiest way to kill OpenCL, so ARM doesn't get too keen on selling everything non-Tegra...
September 22, 2010 3:42:37 PM

Aren't today's CPUs mostly based on the x64 architecture? Pure x86 CPUs died off a few years ago.The x86 instructions will run but I don't think efficiently on todays CPUs (from both AMD and Intel, maybe even VIA).
September 22, 2010 4:02:33 PM

LORD_ORIONOpenCL is owned by apple.

Key word here being "open". Hence, no one company may "own" it. It is for all purposes, a collective effort by many of the top companies in the field to further create something that should be standardized rather than owned, licensed, controlled, and profited by one entity only. If I'm not mistaken though, Nvidia is already a participant anyway and has been, strange if they are in fact trying to compete against it now.
Anonymous
a b à CPUs
September 22, 2010 4:18:18 PM

Could it possibly increase performance of a netbook with intel graphics?
Or is the program too heavy to load on these machines?
a b à CPUs
a b Î Nvidia
September 22, 2010 4:42:35 PM

Oh, Cuda can be recompiled to run on different architectures? How exciting, wait, we already have something like that

IT'S CALLED C++!!!!!!!!!!!!
September 22, 2010 4:45:03 PM

this is probably a great move for nvidia, now it's cuda programming language will become more widespread since it can be run on any system. Then with more programs using cuda, it will really give someone a reason to buy an nvidia gpu
September 22, 2010 5:00:24 PM

tu_illegalamigoDo you still use IA64? I remember that netburst era stuff was awful on cpu intensive workloads. Hell, I can`t even find the ia64 version of 2003 server anymore.

Microsoft just released Windows HPC Server 2008, but they no longer support Itanium. It's an executive decision, if they are wildly successful with HPC Server they may decide to try to aim for the bigger servers market again in the future.
September 22, 2010 5:03:31 PM

A great way of getting a bunch of Indian programmer w/o GeForce 8+ to make stuff, even though they can't use it very well on what they've got. As for OpenCL & NVIDIA not having anything against something they get to share with ARM, Apple, Ati, etc, etc... Why not help to let others help them with CUDA instead?
September 22, 2010 5:13:36 PM

... when?
September 22, 2010 5:19:59 PM

Possibly, but Ia64 in my opinion is at the bottom with x86. ARM is a better choice for most server workloads now, except for cpu intensive, then it`s Opteron, or if you have the money Xeon.
September 22, 2010 5:28:18 PM

I don't see this being all that useful unless x86 processor manufacturer star implementing CUDA on the cpu. Given that would cut into Nvidia's market I don't see it happening which means this won't make a big impact on DirectCompute or OpenCL.
September 22, 2010 6:42:15 PM

This sounds great, unless it is not a trick to make people believe that CPUs work slower on CUDA than GPUs (which they can do by making the software run slower on CPUs).
September 22, 2010 6:44:43 PM

bmaddWhats the point of CUDA without a GPU? Multi core or not.


Portability.
Anonymous
a b à CPUs
September 22, 2010 8:59:38 PM

Forget the point about performance on CPUs. This move is very strategic for Nvidia embrace more from the gpgpu plataform, putting ATI (opencl) aside. How many apps do you see using OpenCL? And using CUDA? They're very serious about that.

Anyway, as CUDA can be ported to CPUs, CUDA can be ported to ATIs too. ATI could at least contract a 'outside' developer to create this driver, and give us this feature without need to give in.
September 22, 2010 11:53:32 PM

Physx running on CPU....properly!!! Seems completely backward from what Nvidia have been preaching but Im glad to see it. CUDA has a lot of potential and with CPU and GPU's being integrated this could be Nvidia's attempt at making CUDA survive until (if ever) they get a licence to build x86 processors.
September 23, 2010 12:43:15 AM

Awesome, Nvidia can spread CUDA out more, and hopefully they will improve it more often as well.
September 23, 2010 2:35:48 AM

It's quite a great chip for mission critical applications.
September 23, 2010 6:02:38 AM

GPU running a computer good idea that will send us in to the nest are of IT,im talking about running tare flops with a good os , like a cpu running windows xp where it is in the giga flops ,we need to invalve to a higher speed then maby we can work on that stupid lan line crap we by something thats a 10/100/1000 speed Ethernet that only gos maby 50 mgs top talk about over kill,i wish them luck, i wont something that can type faster then what im typing now ,lol more speed yes
September 23, 2010 6:23:26 AM

Afraid, the implementation will be same as with PhysX - just to compromise the idea and give more PR to the GPU-based solution
September 23, 2010 6:24:40 AM

Afraid, the implementation will be same as with PhysX - just to compromise the idea and give more PR to the GPU-based solution
September 23, 2010 5:24:46 PM

agree, this support is not more than a move to compete with opencl. programming with cuda is easier than with opencl. so, with this support that can run in cpu side, why not use cuda?
a b à CPUs
September 25, 2010 12:49:38 AM

I think this is a great move by NVidia.
September 26, 2010 12:50:05 PM

They want to popularize CUDA as much as possible by giving support to CPU as well. There is no advantage doing CUDA on CPU they just want to popularize that platform. So that there GPUs get performance advantage over ATI. NVIDIA is the most Wicked company in the world.
March 9, 2011 5:44:42 AM

nVidia already supports DirectCompute and OpenCL on thier hardware. They have whole sections of thier website and documentation for that. You can already choose to use either of those, or the CUDA extensions for C++ or Fortran as well, on all nVidia CUDA enabled hardware. It's your choice which you prefer to use, and which is best suited for your application (and skills). Obviously a 4 or 6 core cpu will never give any significant performance compare to hundreds of cores and tons of dedicated memory on a specialized board, or even a consumer class GTX graphics board. But, as previous comments have said, cpu support gives access to developers that otherwise wouldn't be able to use the tech, and allows use in some hardware situations that, while not ideal, at least would enable it. Thier hardware is not exclusive to CUDA, and now CUDA won't be exclusive to thier hardware. A lot also depends on what type of application you're developing and what type of calculations are being performed. Like anything, different things are better suited. But, if you're working with the ideally suited types of software, there is no comparison in performance when designed well. There are many examples where the same type of application and calulations are compiled and run on different hardware and development languages and the difference in speeds is sometimes simply staggering. You just need to choose what best suits your needs.
!