- Saturday, 28 November 2009
- INQ Mobile
- RSS
General Garcia is dead now, but there are other Garcias - Elbert Hubbard
GPGPU "to accelerate operating systems"
NVIDIA RECKONS Open CL and Direct X are key to operating system acceleration on the GPU and claims GPGPU (general-purpose graphics processing units) will allow for serious speed-ups of Microsoft's Windows and Apple's Snow Leopard.
The green goblin's Tesla products manager, Sumit Gupta, told CNet that OpenCL and DirectX were the tools which would allow OSs to take advantage of a machine's graphics chips, speeding up the software and taking pressure off the CPU.
Now, with OpenCL forming part of Apple's Snow Leopard and DirectX included in Windows 7, Gupta said punters with GPUs would be able to, "run the operating system faster because the operating system will essentially see two processors in the system."
"For the first time, the operating system is going to see the GPU both as a graphics chip and as a compute engine," he added.
Gupta said the increasing trend towards GPU acceleration didn't signal the death of the CPU, explaining that for unpredictable tasks, the CPU was still "the jack of all trades". He added, however, that the GPU was particularly adept at performing highly parallel tasks quickly.
Taking Apple's OS as an example, Gupta noted its interface actually contained, "more visual content than there is sequential (CPU) content," meaning there was more available for the GPU to work on and accelerate. Google's Picassa picture editing tool was trumpeted as an example of software which could easily be accelerated on the computer's graphics chip.
It certainly hasn't been an easy road to tapping the GPU's power, with Gupta saying the most difficult aspect for programmers had been having to use a graphics language to squeeze any use whatsoever out of it. Now, however, with OpenCL and Nvidia's much-hyped Cuda, programmers have access to a language conveniently based on C, making programming for the GPU that much easier.
"The CPU," concluded Gupta, "is one aspect but not necessarily the most important aspect anymore". [Cough! Ed.] µ
Share this:
ShareBusiness Intelligence Software
Customer Relationship Management (CRM)
Enterprise Accounting Software
Enterprise Resource Planning (ERP)
Enterprise Systems Management
Graphics processing is still relates with PCI Express BUS. PCI Express bus can't keep up their total transfer rates per second without high bandwidth that provided from northbridges/memory controller and processor. People have seen that GTX 295 would be wasted their performance potential if did not paired with fastest processor with hight clock rates quad core processor. If this GPU paired with only a single core like sempron, the buyer of this GPU could not achieves high frame rates when they are gaming.
Don't see how it helps MacOS and Linux...
your OS.
You should be able to add an Arm co-processor card for less than cost of the packaging on an NVidia
Funny, but all the additional uses proposed for the GPU all had something to do with graphics, and wow, does osx *really* do more work updating the UI than maintaining the computer?! I'm shocked, shocked!
And now, a dedicated graphics processing unit can even accelerate: a picture editing software. This is supposed to be progress? I had an Amiga. Way ahead of you people.
As a GPGPU programmer I don't see how an OS can be optimised *significantly* with the gpu. Sure GPGPU is good for something that can be parralised like a image editing program, but not the OS itself.
This Nvidia guy is full of shit. :(
"run the operating system faster because the operating system will essentially see two processors in the system" What? Does he not realise that a completely different architecture requires totally different code frameworks!? It will never just slot in and work as part of the OS, if anything ittle just slow down most common tasks with the overhead it incurs.
pARRELLEL means that multiple data streams arrive at point X at same useable time. Games are fast, yet refresh every 1/60th of second leaves plenty of room for accuracy.
We know computer isn't that fast at anything, yet Now its Much faster than MyTurtle, Whom Died Being too slow to move out of way of traffic. I often wished little critter has roller blades at time, alas. Not To Be.
Nvidia Time IS NOW & GpGpu Is HOT, Hot item. Remember just two years ago, nvidia had to redo all NF4 chipsets, REdo bottom core(s),just getting back to 300 & Now Must Be Preparing for ultee'7, Or Why Deeper InstSets? TS drashek
It's pretty much the holy grail of PC gaming, have every PC require a dedicated GPU on some level.
Not every PC will be used for gaming; but it would do wonders for potential install base.
Enough of this BS, everyone claims their product is used better doing (X) than (Y), and (Z) is the new trend because we've got a new way to do it.
When I see real-time raytracing, I'll be impressed and amused.
Just goes to show how full of crap NVidia "executives" are.
GUI processing is only ONE part of system; not even part of OS, depending on architecture, but seeing as Microsoft has screwed the pooch by making GUI processing kernel-based. But how a GPGPU helps with what an OS really does, which is manage computer resources is beyond me. Most likely beyond those dick-heads also.
But then desperate people will cling to anything that appears to float.
Firstly, let me LMAO...
...ok, now that I can breath again, I can say that like other persons commented, that guy is FULL of BS. Totally. Where in the Earth could DirectX accelerate an OS?...
I think he needs to study some OS basics again... I can't even keep on commenting on this...
uh, your obviously not a programmer. DirectDraw is already accelerating OS functions. Why couldn't DirectX accelerate future 3D functions in the OS.
Let's hope YOUR software or OS vendor has the latest compiler for a multi-CPU with a Multi-GPU on the die.
Programmers should not be expected to optimize for specific AIBs or chipsets.
Hardware
Hardware
Hardware
The best GPGPU in my pants!
Beneath the sign that said "U-haul"
She left angels hangin round for more
As "b" already mentioned, a "General Purpose GPU" is indeed an oxymoron. Indeed, the abbreviation "GPGPU" stands for "General-purpose computing on graphics processing units". It means to use GPUs for tasks they were originally not meant to be used for, for example database processing or image processing.
Nevertheless, a processor can't be good at "general purpose" and at the same time be good at "special purpose" computations (i.e. graphics processing). Finally, GPUs are special purpose and CPUs are not special purpose.
I also doubt that a GPU is really useful for accelerating operating system tasks. A GPU is only good at tasks where you have hundreds of very similar small tasks running in parallel.
Programmers are expected to do what their managers tell them to do.
Man,
As a skilled (let say GPGPU-skilled, but I hate this) programmer, I can confirm that running ANY code on OpenCL-based gear is possible, and with careful optimizations, profitable. However, porting ANY code for achieving little or no performance gain is just waste of the resources (time AND / OR money), esp. if done only for proving the ability.
A better approach is to optimize your algo for parallel systems. When you finish optimizing, you will realize that GPUs are more general-computing friendly, and CPUs are heavily equipped with SIMD engines. Win-win situation.
Up until the last paragraph, Drashek made a bit of sense. Unfortunately, half of that involved talking about his dead pet turtle that would've survived if it had been a rollerblading turtle.
I would love to see an actual rollerblading turtle. If I can't get that, how about an accidentally injured turtle with either a prosthetic device, or a tiny wheelchair. LMAO
Being handicapped isn't fun or funny, but the technology being invented to help handicapped people and animals is way cool.
Now that there's not one but two incipient standard GPU compute APIs, Cuda's days may be numbered. The NV guy didn't even mention it as the language of choice!
BTW NV has been hubriciously trying to make Intel obsolete for years.
"DirectX included in Windows 7"
DirectX is in Windows now !?
Holy Crap! When did that sneak up on me?
CUDA is an architecture, like .NET, that can compile code that will run on the GPU. Currently C is the only supported language but other languages could be supported.
And I believe nVidia's OpenCL implementation is actually built upon CUDA. Kind of like how many new windows apps today are built on .NET.
At least this is my understanding.
You guys are just so damn cynical.
Anything that smells like progress is waiting to be trashed by you.
Maybe it's your old anti-Nvidia thoughts clouding your reason?
A GPU can output at 10x the rate of a prossie,a little offloading isn't going to make a lot of difference there,but utilizing that extra unused power can only be a good thing.
As someone said,if it helps bring forward that power to do ray tracing,amonst other things,then that sounds like progress to me.
CUDA, is still in it's infancy,it's potential,like full GPGPU will become useful to pc users in time.