Sign in with
Sign up | Sign in
Your question

GPU 375 GFLOPs superpc. Who needs a (fast) 24 GFLOPs CPU....

Last response: in CPUs
Share
September 30, 2006 12:55:31 PM

This is what I'm wondering about...howcome that some software being created to make use of the GPU calculating power always (except for peakstream appz) go up in smoke and dissapears from the face of the earth? Are companies like Intel and Amd giving these guys a lot of money to not make their project go public? If it IS possible which Peakstream allready proves with their software used also for medicale purpose would it not be smart for programmers to start developing software that turns every pc into a super computer and get a lot of money from Intel and AMD for not releasing it? I mean...Let's be honoust....we are still talking about DUAL CORE/QUAD CORE while in fact we can go way faster. We are being fooled I guess cause it's allready possible to have your own super pc at home. You dont think so? Check this link and please show me your point of view....

http://www.tgdaily.com/2006/09/18/peakstream_supercompu...
September 30, 2006 1:33:47 PM

This is why ATI gave R520+ hardware prefetch and is their biggest advantage over Nvidia.
September 30, 2006 2:23:02 PM

Quote:
This is what I'm wondering about...howcome that some software being created to make use of the GPU calculating power always (except for peakstream appz) go up in smoke and dissapears from the face of the earth? Are companies like Intel and Amd giving these guys a lot of money to not make their project go public? If it IS possible which Peakstream allready proves with their software used also for medicale purpose would it not be smart for programmers to start developing software that turns every pc into a super computer and get a lot of money from Intel and AMD for not releasing it? I mean...Let's be honoust....we are still talking about DUAL CORE/QUAD CORE while in fact we can go way faster. We are being fooled I guess cause it's allready possible to have your own super pc at home. You dont think so? Check this link and please show me your point of view....

http://www.tgdaily.com/2006/09/18/peakstream_supercompu...



The real advantage of GPUs is that they are designed as massively parallel FP units. CPUs are designed to run general purpose code (an OS) and fit into a socket with external RAM. When you include that overhead, the CPU is at a distinct disadvantage. If AMD or Intel were to line up the same amount of relative FP units they would catch up to the GPU.

That's why Intel is touting their massive core chips (80) would easily do physics and game code and background OS threads. I'm surprised they are not looking at HW threading at this point. Both AMD and Intel are at a fine enough size to implement it.

The first Windows CPU to get HW threading will be the fastest for awhile. Imagine two threads on Core 2 Quad or Barcelona.
Related resources
September 30, 2006 2:48:11 PM

Thx for telling me this. You are right indeed about the fp difference.......and the funny part is.....the gpu does not produce that much heat as well to do so while many people complain about cpu heat problems. ;)  I'm wondering when these 80 core cpu's will come and shine. What's the time they need to make this a reality. And what's the time it takes to invent a quad core instead of a dual core. Adding 2 or 80 cores more to the allready know how multicore concept. What's the difference? :)  It's just the little part in the cpu who controls these cores like a network right? So should we still keep talking about what cpu is best while 80 core multicpu's are allready possible? The time we need to wait for such cpu's to show must be a money problem instead of a development problem, right? ;)  So if this is true...why still mumbling about who is better....AMD or Intel....it's just a big joke that way. while we are happy with dual core they can allready produce 80 cores allready if they want to.
Anonymous
a b à CPUs
September 30, 2006 4:04:06 PM

TO answer your question about building a multi-core I suggest you read up on the Cell processor from sony/ibm/toshiba, you will see that it's not easy as it sounds and that they had to do quiet a few sacrifice. That's why AMD and intel are taking a slowing approach 2 - 4 - 16 and so on. Basically they want to get there with sacrifying single threaded performance/yield/etc.

Anand's article was pretty good, but theres more around the net
September 30, 2006 4:31:50 PM

Quote:
This is what I'm wondering about...howcome that some software being created to make use of the GPU calculating power always (except for peakstream appz) go up in smoke and dissapears from the face of the earth? Are companies like Intel and Amd giving these guys a lot of money to not make their project go public? If it IS possible which Peakstream allready proves with their software used also for medicale purpose would it not be smart for programmers to start developing software that turns every pc into a super computer and get a lot of money from Intel and AMD for not releasing it? I mean...Let's be honoust....we are still talking about DUAL CORE/QUAD CORE while in fact we can go way faster. We are being fooled I guess cause it's allready possible to have your own super pc at home. You dont think so? Check this link and please show me your point of view....

http://www.tgdaily.com/2006/09/18/peakstream_supercompu...


Real world performance is what matters, and frankly I don't see the numbers being anywhere near their "hyped" numbers.
September 30, 2006 6:56:20 PM

The Folding@Home project will soon benefit from a new client designed to run on ATI's X1800 & X1900 GPUs, they mention a 20x to 40x increase over CPU code in the High performance client FAQ.
October 1, 2006 1:29:38 AM

Quote:
This is what I'm wondering about...howcome that some software being created to make use of the GPU calculating power always (except for peakstream appz) go up in smoke and dissapears from the face of the earth? Are companies like Intel and Amd giving these guys a lot of money to not make their project go public? If it IS possible which Peakstream allready proves with their software used also for medicale purpose would it not be smart for programmers to start developing software that turns every pc into a super computer and get a lot of money from Intel and AMD for not releasing it? I mean...Let's be honoust....we are still talking about DUAL CORE/QUAD CORE while in fact we can go way faster. We are being fooled I guess cause it's allready possible to have your own super pc at home. You dont think so? Check this link and please show me your point of view....

http://www.tgdaily.com/2006/09/18/peakstream_supercompu...


Real world performance is what matters, and frankly I don't see the numbers being anywhere near their "hyped" numbers.

100 GFLOPS sustained seems pretty good to me for a 8 month old video card. Imagine what R600 will do in jan/feb 8O
October 1, 2006 3:27:20 PM

Quote:
This is what I'm wondering about...howcome that some software being created to make use of the GPU calculating power always (except for peakstream appz) go up in smoke and dissapears from the face of the earth? Are companies like Intel and Amd giving these guys a lot of money to not make their project go public? If it IS possible which Peakstream allready proves with their software used also for medicale purpose would it not be smart for programmers to start developing software that turns every pc into a super computer and get a lot of money from Intel and AMD for not releasing it? I mean...Let's be honoust....we are still talking about DUAL CORE/QUAD CORE while in fact we can go way faster. We are being fooled I guess cause it's allready possible to have your own super pc at home. You dont think so? Check this link and please show me your point of view....

http://www.tgdaily.com/2006/09/18/peakstream_supercompu...


Real world performance is what matters, and frankly I don't see the numbers being anywhere near their "hyped" numbers.

100 GFLOPS sustained seems pretty good to me for a 8 month old video card. Imagine what R600 will do in jan/feb 8O

Well see if they can build a decent compiler that can even get 40% of that.
October 1, 2006 6:15:23 PM

You probably don't hear about them since the ones with the biggest gains are all scientfic and not gammer related. It is becoming more common in my field to do 3D FFT and matrix inversions on the GPU. The biggest problem is updating all the backend machines with high performance GPU's (memory is what holds us back).
October 1, 2006 7:50:42 PM

Quote:
The first Windows CPU to get HW threading will be the fastest for awhile.

What is a "Windows CPU"?
!