Review - AMD Radeon R9 290X
#41
hm, yeah i definitely have the minclock set the same as the maxclock. i just can't get any additional performance out of them through overclocking. definitely get more heat, just no additional performance. then again, i have't flipped the silly switches yet. so that might have something to do with it. i'll try some switch flipping and see what happens.

no, i haven't gotten amdconfig to /really/ work with the release driver. it doesn't seem to do anything at all with these cards.
#42
looks like the uber mode switch was the ticket, thanks for the tip Smile
#43
So, in view of the recent performance boost, would it be a fair assumption to say that a rig with 8xR9 290x cards (or 7 if you put the InfiniBand as well) has become the ultimate (as of December 2013) number crunching rig?
#44
yessir
#45
How's your power supplies holding up in that box? Isn't the tyan's ps output 2400 watt total? I see the 290x's are rumored to use around 295 watts per card at load, 8 of them appears as it would be a huge load.
#46
Any results on R9 290X with the new drivers yet?
#47
(01-02-2014, 07:28 PM)powderspecial600 Wrote: How's your power supplies holding up in that box? Isn't the tyan's ps output 2400 watt total? I see the 290x's are rumored to use around 295 watts per card at load, 8 of them appears as it would be a huge load.

we have actually only seen ~ 240W actual power draw, as measured by a Kill-a-Watt.

the power supplies are Delta DPS-1200RB, which are 80plus Platinum certified and can reliably produce 1200W of output.
#48
(01-03-2014, 12:39 AM)Genbox Wrote: Any results on R9 290X with the new drivers yet?

i have already made several edits to the first post, and a new thread will be posted shortly.
#49
I was wondering if you have measured the performance of the 290X with a real job and what numbers you've got.

I have tested the NTLM algo with oclhashcat 1.01, in comparison to the 6990. The O/S is linux 64-bit, driver 13.12, GPU switch set to performance, used your tool with the autofix option. If I run oclhashcat with the benchmark option (-b -m 1000), the results are consistent with your values:

290X=18602.5 MH/s
6990=20016.5 MH/s

If, however, I do similar measurements with a set of hashes, in hopeless brute force mode (-m1000 -a3 -i --increment-min=14), I get different performance ratios:

1 hash
290X=5000.0 MH/s [erratic, between 4400-6800 MH/s]
6990=17714.8 MH/s

100 hashes
290X=4800.0 MH/s [erratic, between 3700-6800 MH/s]
6990=8702.8 MH/s

10000 hashes
290X=4800.0 MH/s [erratic, between 3700-6800 MH/s]
6990=7852.6 MH/s

1000000 hashes
290X=4800.0 MH/s [erratic, between 3700-6800 MH/s]
6990=3579.6 MH/s

When using rules with large amounts of hashes, the 290X is about twice as fast as the 6990 with the same input.

Am I doing something wrong, or are the results consistent with your observations?

Cheers,
ecos
#50
You need to tweak powertune, fan and clock settings so the card doesn't throttle (otherwise it does so even when not hot at all). Specifically, the trick is to have the hardware switch in "über" mode and set min clock == max clock.

If you're on Linux, try eg. https://github.com/epixoip/od6config

Having done that you can also OC the card to 1100/1350 or so, with no problems.

Edit: I see now you did use the od6config tool, so my bet is you did not flick that switch on the card. It needs to be set "away from the rear".

Edit2: If the card still throttles despite everything done right, it really gets too hot (>94C) and you should be thankful it saves itself by throttling. Fix cooling if this happens.