New GPU Testing Methodology Puts Multi-GPU Solutions In Question

Vigile writes "A big shift in the way graphics cards and gaming performance are tested has been occurring over the last few months, with many review sites now using frame times rather than just average frame rates to compare products. Another unique testing methodology called Frame Rating has been started by PC Perspective that uses video capture equipment capable of recording uncompressed high resolution output direct from the graphics card, a colored bar overlay system and post-processing on that recorded video to evaluate performance as it is seen by the end user. The benefit is that there is literally no software interference between the data points and what the user sees, making it is as close to an 'experience metric' as any developed. Interestingly, multi-GPU solutions like SLI and CrossFire have very different results when viewed in this light, with AMD's offering clearly presenting a poorer, and more stuttery, animation."

Comments

Re:You use GPUs for video games?

[amanicdroid]

This is the explanation I've been given for the disparity between Nvidia and AMD:
https://en.bitcoin.it/wiki/Why_a_GPU_mines_faster_than_a_CPU#Why_are_AMD_GPUs_faster_than_Nvidia_GPUs.3F

Specifically:

Secondly, another difference favoring Bitcoin mining on AMD GPUs instead of Nvidia's is that the mining algorithm is based on SHA-256, which makes heavy use of the 32-bit integer right rotate operation. This operation can be implemented as a single hardware instruction on AMD GPUs (BIT_ALIGN_INT), but requires three separate hardware instructions to be emulated on Nvidia GPUs (2 shifts + 1 add). This alone gives AMD another 1.7x performance advantage (~1900 instructions instead of ~3250 to execute the SHA-256 compression function).

For GPU programming I've enjoyed Nvidia's CUDA package greatly over wrangling OpenCL that Radeon relies on.

Re:You use GPUs for video games?

[pclminion]

Out of curiosity, what's your break even point?

I don't know where the break even point is, but once you pass it, you can be very profitable. One of my friends built a custom "supercomputer" out of cheap motherboards and graphics cards for about $80k -- along with completely custom software to automatically tune clock speeds and fan rates in real time (all of which was written in bash script). At peak performance, his machine generated about $20k worth of bitcoin every month, which easily paid for the $12k monthly electric bill.

After a couple of difficulty-doublings, and the imminent arrival of the ASIC miners, this lost its profitability, and he went back to being a DBA... The machine is still out at the farm, cranking away. I think he'll disassemble it and part it out for cash in a month or two.