GPU benchmarks

[dmeyer]

Thanks, that answers a lot of my questions. I was planning to build 3 workstations, and was inclined to use 2x W8000 on each. This certainly clears a lot of doubt in my minds. The latest benchmarks on W8000 seem very impressive, and the price is also not too out of reach.

The current W8000 is based on the Tahiti, I was hoping the Hawaii based pro cards would come soon. Can't imagine what a 290X based FirePro can do, if the W8000 is performing like this.
http://www.anandtech.com/show/7901/a...-firepro-w9100

No timeline on them though...

w9100's are supposedly available quite soon. MSRP $4K US.

[cob]

I know that the W9100 will be out of budget for most of us, but W8100 and W7100 will surely follow in the same footsteps, and be more affordable to the masses.

[cob]

Another question: has anyone done this benchmark scene with the Quadro K5000?

Looks like, due to my office's corporate vendor restrictions, AMD is out of reach. So it's either going to be 2x Titan Black, or 2x Quadro K5000. Between those two, I like the specs on Titan Black so far. But I need to see more on the K5000 (to ponder the worth of the extra $$$ to be spent if I went K5000 instead of Titan Black).

Any suggestions???

[dmeyer]

Another question: has anyone done this benchmark scene with the Quadro K5000?

Looks like, due to my office's corporate vendor restrictions, AMD is out of reach. So it's either going to be 2x Titan Black, or 2x Quadro K5000. Between those two, I like the specs on Titan Black so far. But I need to see more on the K5000 (to ponder the worth of the extra $$$ to be spent if I went K5000 instead of Titan Black).

Any suggestions???

I haven't done this bench on K5000, but on other GPU renderers the Titan will outright murder a K5000.

[fraggle]

Exept if such GPU are for a render GPU farm, I really doubt that pro line have anything better than the gamer one.
What Nvidia called "maximus" is just for marketing folk.

You can get 3 Titan for the price of one K5000.

Another question: has anyone done this benchmark scene with the Quadro K5000?

Looks like, due to my office's corporate vendor restrictions, AMD is out of reach. So it's either going to be 2x Titan Black, or 2x Quadro K5000. Between those two, I like the specs on Titan Black so far. But I need to see more on the K5000 (to ponder the worth of the extra $$$ to be spent if I went K5000 instead of Titan Black).

Any suggestions???

[Valtiel]

Or a few 780 Ti's if you're just doing RT stuff and no double precision calculations.

[matanov]

W9100 - 2:36s / 14.20 custom drivers , it's bad comparing to Nvidia cards BUT actually it's a BIG step forward comparing to the older drivers they sent us 2 weeks ago (3 min).

[cob]

W9100 - 2:36s / 14.20 custom drivers , it's bad comparing to Nvidia cards BUT actually it's a BIG step forward comparing to the older drivers they sent us 2 weeks ago (3 min).

W9100 with 2:36 - I was expecting a bit better from those esp considering Vlado tested a W8000 with a 2:39.

Damn, these results (though not as good compared to the NVidia) plus their Adobe OpenCL image/video benchmarks keep dragging back in the FirePro territory. I just finished talking to my corporate rep a minute ago, and I'm requesting them to try my system with the W8000 again. My guts telling me to stay with W8000 over the K5000.

[vlado]

The W8000 I tested was in a Silverdraft Demon machine, I don't know if this matters... I also tested V-Ray 2.4 where the OpenCL code is a bit simpler.

Best regards,
Vlado

[Morbid Angel]

so I got 780 ti. Here are some results:

vray rt 3.05.05
cuda
2652.955 kpaths/s
2m.08s for frame

I have to note though, the light cache is being calculated on the cpu, so it takes 51 seconds to do the LC, which makes the render time incorrect for pure gpu.

[vlado]

You must switch to brute force; otherwise the results are skewed and not comparable to the other benchmarks.

Best regards,
Vlado

[Morbid Angel]

You must switch to brute force; otherwise the results are skewed and not comparable to the other benchmarks.

Best regards,
Vlado

Thanks!

new result using bf/bf:

vray rt 3.05.05
cuda
1976.967 kpaths/s
1m.46.5s for frame

[danio]

AMD r9 290x @1100mhz = 1m 59s

Pretty good! Certainly brings AMD back into the game I think.
In fact given that this card has 4gb ram and the 780 only has 3gb of ram, this is probably a better choice with the Titan being the only better choice.

What about the 280x? It looks like there are 6gb versions of that out right now for about $400. I haven't been paying super close attention to the AMD product line but I don't see dual cores that would mean split ram in the specs or anything. Is the 280x just obscenely slow compared to the 290x in compute?

Even if it was moderately slower than some, at that price and with 6gb it makes it a real possibility for exterior arch viz scenes it seems like.

Dan

[Valtiel]

The 280X is the old 7970 rebranded. Only the 290's and 290X are on their new Hawaii architecture so you can reasonably expect a significant performance difference.

I feel like the size of one's VRAM (or system ram for that matter) is often over-stated as a determinant of performance. RAM size, to me, is something that affects performance until it is enough and more past that does not benefit you at all. RAM speed is another story but even that can be a marginal factor in many cases. Somehow I doubt the test scene, or many workstation applications, would use 6GB of VRAM. 3GB was enough for the originals in most gaming applications when they launched 2 years ago. Especially without FirePro drivers, I'm fairly confident RT would have trouble using more than that original amount.

Devs: Any progress on getting RT 3.0 working with AMD hardware? I'm still getting that string of errors and crashing.

[vlado]

Devs: Any progress on getting RT 3.0 working with AMD hardware? I'm still getting that string of errors and crashing.

This has been working fine for us with the latest builds, latest drivers and latest hardware.

Best regards,
Vlado