Advantages and disadvantages between diffrent Workstation GPU's

[humbug]

http://streamcomputing.eu/blog/2014-04-08/accelerators-hpc-2014/

Traditionally this is a space that Nvidia have dominated.
but recently with AMD's MAC Pro Win its time to look at what AMD are doing here these days. AMD are fighting hard with Better performance per watt, better functionality and API support, especially with Open Standards.

Conclusion
Open standards

For who doesn’t know, I prefer open standards. Reason is that switching hardware is easier, giving you space to experiment. AMD, Intel and Altera support OpenCL 1.2 and will start later this year with 2.0, whereas NVIDIA lags over 2 years and only support OpenCL 1.1. Now the results are very visible: due to problems with Maxwell, you need to postpone you plans to 2015 if you code in CUDA. There is on way to pressure them, as all you can do is to port all your code to OpenCL. I have been astonished by the FUD around OpenCL, but that at least says they are afraid of the open standard.
Green 500

You might have noticed the big differences between the GFLOPS/Watt. Where this is important, is the Green 500 – the list of energy efficient supercomputers. The goal of today’s supercomputers is that they are mentioned in the top 10 of both lists. If you build an efficient cluster (say 2 CPUs + 4 GPUs), you can get to 70-80% of max DGEMM performance. Below is a list for 75%:

AMD FirePro – 7.10 GFLOPS/Watt DGEMM -> 5.33 GFLOPS/Watt @ 75%
NVIDIA Tesla – 5.65 GFLOPS/Watt DGEMM -> 4.24 GFLOPS/Watt @ 75%
Intel XeonPhi – 3.56 GFLOPS/Watt DGEMM ->2.67 GFLOPS/Watt @ 75%

Currently this list is lead by a cluster with K20X GPUs, steaming out 4.50 GFLOPS/Watt, which even has 86% of max DGEMM.

In other words: if the FirePro gets out in time, then the green 500 could be full of FirePro GPUs.
The winner

As there are only three offers, they are all winners. What matters is the order.

AMD FirePro – 16GB fast memory, clear winner in DGEMM performance. Bad: not available yet.
NVIDIA Tesla – Second to everything from FirePro (bandwidth, memory size, GFLOPS, price). Bad: outdated OpenCL-support.
Intel XeonPhi – Same as FirePro if it comes to memory, but 60% slower in DGEMM and 50% less efficient. Bad: 300 Watt for a server.

I am happy that after years of NVIDIA leading the pack, AMD is the clear winner now. As AMD is the foremost supporters of OpenCL, this could seriously democratise HPC for the times to come.

 

[Uncle Petey]

Er... Yay AMD?

 

[drunkenmaster]

Basically what a lot of people have been highlighting lately. AMD has good wide compute support, great openCL performance and with HSA will certainly be helping.

Even Fudzilla did an article pointing out some pretty damn strong AMD wins in multiple professional benchmarks.

AMD have gone from 20 to 30% market share in a pretty short space of time. One of the biggest issues AMD had in professional markets was actually picking a fight in the market. They more just made a Firepro version of their gpu's and that was about it, they've finally gotten far far more aggressive with marketing, sales and bringing in people to push them in markets and go after big wins like Mac Pro's.

Will be interesting to see where AMD professional gpu's get in the next couple years. Such a high margin segment with effectively no excess R&D costs as it's the same gpu's fundamentally so raising sales means an increase in overall margins and healthy improvement in profit. The 10% they've gained in professional could easily be 150-200mil extra in profits for the year, if they could get to 40 or even 50% in a year or two, it really does effect AMD's bottom line significantly.