AMD RX 480
 

This card could be awesome for cheap trial-factoring.

It's supposed to retail for $200 US, and is supposed to deliver 5.5 TFLOPS. Not only that, but it's power connections max out at 150 watts, and analysis by others comparing its performance to the GTX 1080 suggest it will consume around 90 to 100 watts. Even at 150 watts, this is a very efficient card.

It's got 2304 cores, supposedly at a 1266 MHz base clock. It should be slightly faster than an R9 290X. 600+ GHzd/d for $200 consuming only 150 watts is fantastic.

I've been slammed with regular work lately, but we're set to get in some of the next-gen consumer GPUs to evaluate in the next few weeks.

I've got a Founders Edition 1080 coming in. Will report on TF performance once I get it installed.

Pascal is basically a shrunk Maxwell, so the performance can already be estimated clock-for-clock. It will be interesting to see what it can achieve with overclocking and watercooling. Apparently the default fan profile doesn't keep a good job keeping the GPU cool enough to avoid thermal throttling.

[QUOTE=Mark Rose;435566]Pascal is basically a shrunk Maxwell, so the performance can already be estimated clock-for-clock. It will be interesting to see what it can achieve with overclocking and watercooling. Apparently the default fan profile doesn't keep a good job keeping the GPU cool enough to avoid thermal throttling.[/QUOTE]

It is a shame that GPU manufacturers cut corners on cooling. I guess that the calculus is that no 'reasonable' person would drive a (consumer level) GPU to the wall. Of course, the same is true of "stock" CPU coolers.

Still, at the risk of sounding like a shill, my MSI GTX 580 runs far beyond its factory OC, with its 'stock' cooler, and stays in the upper-mid 70s C, unless the room gets really warm. [STRIKE]Right now, with the ambient about 24 C, it is running 918 MHz at 75°, or less.[/STRIKE] It stabilized at 74 after a reboot.:smile:
EDIT: Sorry. totally OT tangent.

[QUOTE=kladner;435579]It is a shame that GPU manufacturers cut corners on cooling. I guess that the calculus is that no 'reasonable' person would drive a (consumer level) GPU to the wall. Of course, the same is true of "stock" CPU coolers.

Still, at the risk of sounding like a shill, my MSI GTX 580 runs far beyond its factory OC, with its 'stock' cooler, and stays in the upper-mid 70s C, unless the room gets really warm. [STRIKE]Right now, with the ambient about 24 C, it is running 918 MHz at 75°, or less.[/STRIKE] It stabilized at 74 after a reboot.:smile:[/QUOTE]
[B]EDIT 2.5: [/B]@ 74 C with P95 running, and the fans all revved up. :whee:
EDIT 2: Sorry. totally OT tangent.

[QUOTE=Mark Rose;435548]This card could be awesome for cheap trial-factoring.
.[/QUOTE]

Also the crowd that use GPUs to mine crypto currencies are frothing at the mouth to get their hands on this card.

[QUOTE=airsquirrels;435565]

I've got a Founders Edition 1080 coming in. Will report on TF performance once I get it installed.[/QUOTE]

If you want to run Cuda work on the 1080 Airsquirrels, you might want to sign up as a Cuda developer ( I think it's free ) so you can download Cuda 8.0 pre-release. I believe that's the only Cuda version that will work with Pascal, as one of it features is "Pascal support".

[QUOTE=tServo;435711]If you want to run Cuda work on the 1080 Airsquirrels, you might want to sign up as a Cuda developer ( I think it's free ) so you can download Cuda 8.0 pre-release. I believe that's the only Cuda version that will work with Pascal, as one of it features is "Pascal support".[/QUOTE]

I'll give it a shot. I wonder if the Maxwell/ Cuda 7.5 bug that affected mfaktc has been fixed?

[QUOTE=airsquirrels;435718]I'll give it a shot. I wonder if the Maxwell/ Cuda 7.5 bug that affected mfaktc has been fixed?[/QUOTE]

I would hope so by now, but ...
Did anyone figure out the exact problem and file a bug report with Nvidia?
However, sometimes bugs just mysteriously disappear when a new version is released.

I'll be most interested to hear your experiences.

Ok I got the 1080 Founders Edition running.

It is still failing 43/107 self tests using 8.0 and compute 6.1, so it seems the mfaktc issue is still present. I skipped the self test to get performance numbers and I'm seeing 1071 GhzDay/Day up to 75 bits, which is 23% faster than my Fury X results.

Also compiled CudaLucas. M49 (4M FFT) runs at 3.4ms/iter, 70 hour ETA

James' estimate is 914.2 GHzDay/d. So that's good news.