Best AMD OpenCL/OpenGL Budget Development Box

[mmstick]

[link]http://hsafoundation.com/[/link] The one thing that I note is that Neither Intel or Nvidia are members of HSA; and, since they are huge players in the financial, scientific and entertainment compute markets, I question what HSA will amount to. And, it seems to be significantly led my Microsoft and their track record is not encouraging.

Another thing that I am not clear about is what HSA provides -- in real terms -- to open source developers. From what I can tell, it is simply a shim layer that distributes ready-compiled compute tasks to various hardware devices. Which seems like a solution in search of a problem. Or a replacement for an integrated MMU.

Add to that that newly released chip families are always marked up to take advantage of price-insensitive early adopters, I wonder what the specific advantage is.

It solves a very real problem -- memory bottlenecks. Especially, it serves the biggest problem holding GPGPU back -- memory bottlenecks. It therefore also makes it possible for GPUs to be 'smarter' since it shares the same memory pool as the CPU, which means it will be able to run a wider variety of different instructions that OpenCL currently cannot do or does very inefficiently. What this means to open source developers is that it will be possible to use GPGPU in a wider variety of programs to speed up processing faster than it would have been with CPU alone. What this means for PC gaming is it will be possible to get highly effecient AI, physics, raytracing, etc. in games. Looking at it right now, because of the memory bottleneck, the only things that run exceedingly well in GPGPU is simple math like crunching prime numbers. This will have a profound effect on the medical, biological, scientific, business, entertainment, gaming, and more industries. What Microsoft does is their business; they are probably looking for the next big thing to keep people from switching to Linux.

[justinzane]

HSA Defines a Hardware Spec?

It solves a very real problem -- memory bottlenecks. Especially, it serves the biggest problem holding GPGPU back -- memory bottlenecks. It therefore also makes it possible for GPUs to be 'smarter' since it shares the same memory pool as the CPU, which means it will be able to run a wider variety of different instructions that OpenCL currently cannot do or does very inefficiently. What this means to open source developers is that it will be possible to use GPGPU in a wider variety of programs to speed up processing faster than it would have been with CPU alone. What this means for PC gaming is it will be possible to get highly effecient AI, physics, raytracing, etc. in games. Looking at it right now, because of the memory bottleneck, the only things that run exceedingly well in GPGPU is simple math like crunching prime numbers. This will have a profound effect on the medical, biological, scientific, business, entertainment, gaming, and more industries. What Microsoft does is their business; they are probably looking for the next big thing to keep people from switching to Linux.

Perhaps you can tell that I have developed a serious allergy to the output of PR and Marketing folks. So, I did not get very far wading through the HSA docs which leaves me excessively ignorant...

But, your statement seems to imply that HSA defines and mandates a certain level of multi-compute-device memory-management hardware functionality. While I agree that this would be excellent for the performance of compliant multi-device implementations, it seems to be strongly restrictive in terms of vendor participation. Without the participation of Intel CPUs, GPUs and MICs and Nvidia GPUs, and assuming the continued incredibly poor open-source support for ARM-oriented chipsets and GPUs, what is the value?

Please know I'm not criticizing you, just trying to be skeptical.

[mmstick]

Perhaps you can tell that I have developed a serious allergy to the output of PR and Marketing folks. So, I did not get very far wading through the HSA docs which leaves me excessively ignorant...

But, your statement seems to imply that HSA defines and mandates a certain level of multi-compute-device memory-management hardware functionality. While I agree that this would be excellent for the performance of compliant multi-device implementations, it seems to be strongly restrictive in terms of vendor participation. Without the participation of Intel CPUs, GPUs and MICs and Nvidia GPUs, and assuming the continued incredibly poor open-source support for ARM-oriented chipsets and GPUs, what is the value?

Please know I'm not criticizing you, just trying to be skeptical.

I knew about HSA before it was a PR thing to begin with. AMD has been headed in that direction ever since they bought ATI. This was their end-goal, essentially. There's nothing stopping Intel or NVIDIA from joining the HSA foundation or competing against it, except for the fact that neither NVIDIA or Intel have proper HSA-capable hardware in development that I know of.

The problem with NVIDIA and Intel is that they aren't so great at designing chips they have no experience in designing. Neither has a strong suit in both graphics and processor designs like AMD. And especially for NVIDIA, HSA implies a design that uses OpenCL 2.0+ when NVIDIA wishes to keep using their proprietary CUDA.

As for ARM, I can't say much about that since ARM has a variety of different designs from a variety of different companies and I know even less about the mobile GPU designs. Someone would just have to design a chip that has the GPU/CPU integrated as one that uses the same shared memory.

In other words, only AMD cares about development of an HSA solution right now, but maybe Intel will change their minds once they see AMD chips knocking them out of benchmarks regularly for a cheaper price and running some cool applications that resulted from the new design. From what I've seen, AMD has strong connections with Microsoft right now, which is kind of confusing when Windows/Intel was originally the big thing, so it's not surprising that Microsoft would be heavily backing it. I'm sure they want some Windows Phones using the mobile HSA APU chips next year.

[Dandel]

Perhaps you can tell that I have developed a serious allergy to the output of PR and Marketing folks. So, I did not get very far wading through the HSA docs which leaves me excessively ignorant...

But, your statement seems to imply that HSA defines and mandates a certain level of multi-compute-device memory-management hardware functionality. While I agree that this would be excellent for the performance of compliant multi-device implementations, it seems to be strongly restrictive in terms of vendor participation. Without the participation of Intel CPUs, GPUs and MICs and Nvidia GPUs, and assuming the continued incredibly poor open-source support for ARM-oriented chipsets and GPUs, what is the value?

Please know I'm not criticizing you, just trying to be skeptical.

I can see why you'd be overly cautious about PR and Marketing folks. I personally would look into either the AMD A10-7850K or the AMD A10-7700K APU that have an expected release date as Jan 2014. This is because of the highly versatile memory architecture that has been mentioned multiple times. The Kaveri series of chips has three main features that are worth while...

• Integrated GCN based GPU ( R7 series w/ Trueaudio and mantle support )
• Integrated Cortex A5 coprocessor ( used for the arm security features )
• Steamroller based processor. ( This has a lot of performance improvements over the trinity and Richland based processors )
• PCIe 3.0 support ( this is a feature for FM2+ boards )

[leonmaxx]

I'd recommend this:
AMD A8-6500 (3,5GHz, 65W)
ASUS A88XM-A (Micro-ATX, FM2+, with Kaveri support)
2 x 4GB DDR3-1866 or faster (I use Crucial Ballistix DDR3-1600 but with low timings, 8-8-8-24)