VRMark now released

[mackerel]

http://www.futuremark.com/benchmarks/vrmark

Start your benching! Downloaded the demo and run it on my Xeon box so far. No idea how this compares with anything, and will do my other systems shortly.

Score 5430, average frame rate 118.37. System is E5 v3 keep forgetting model number but the 14 core one, and GTX 970 (1317/1753).
Score 8989, average frame rate 195.97. System is i5-6600k @ 4.2, 980 Ti (1076/1801)

 

[mackerel]

Ok, not sure the detail is worth it but my other 980Ti system, and a 1070 system also both score around 9000 (6700k @4.2). A 1060 3GB + i3-4360 HT off gave 3521 but I'm not sure who is limiting there. I can turn on HT and see if that helps.

I also noticed the score is proportional to the average fps, scaled so that the reported target 109fps is 5000 points.

 

[EarthDog]

Try the blue one.

 

[mackerel]

I'd have to buy it to do that...

Oh, I did have a look inside the orange one in VR. It doesn't have controller-move so I couldn't actually walk around it due to space limitations, but what I saw was a low resolution mess. Current VR doesn't have the resolution we're used to in monitors, and the amount of detail they packed into the scene turned into blocky mess. Kinda like the Nintendo 3DS in 3D mode.

I re-ran the 1060 3GB system with the i3-4360 HT on, and score was much improved at 5768, compared to 3521 with HT off. This puts it a few hundred points ahead of the 970/Xeon system. I'd caution the Xeon is 14 cores and only turbos to 2.3 GHz all cores active, and I don't know at this point if the bench scales with cores or prefers fewer faster cores.

 

[EarthDog]

You've just seen a huge increase with threads in your testing.

 

[mackerel]

I'm being cautious based on the limited info here... 2 cores was probably starving it, so anything extra through HT was a bonus. But how many threads can it usefully make use of? 4? 8? More? The Xeon is low clock, but has 14 cores and I forgot to mention it has HT off for now. I might do more in depth testing on the weekend.

 

[EarthDog]

Not sure... test it. You have a slew of systems, all of which have HT you can enable and disable. If I remember tonight, I'll test it on 6950x with HT on/off.

 

[NeonCancerCandy]

rig in my sig got

7580 - 165 fps average. stock gpu clocks

i was able to mash out an 8310 @ 181fps with overclocking the gpu +155MHz
http://www.3dmark.com/vrm/15845606

 

[mackerel]

I gave in an bought the full version, since they stacked two different discounts.

I'm going through some CPU scaling tests right now.

Orange Room is showing CPU scaling effects I'm still examining now. For now I can say HT will gain you score on a 6700k, so it seems to have some capability of using more than 4 threads. It doesn't fit into any HT behaviour I've seen before though... I haven't tried observing the CPU activity during a run yet.

For fun I ran 2 cores at 4 GHz, with and without HT. Without HT, it really hurts performance scoring 4738. Turning HT on goes up to 7413, which is actually higher than 4x 3GHz HT off scoring 7366. HT on in that case was 7560. With 4 cores at >4 GHz it's over 9000. If we make the big assumption this benchmark is representative of real world VR apps, you might get away with a high clock i3.

Blue Room is purely GPU limited. I get around 1800 points regardless if I'm running 2 cores with HT off, or 4 cores with HT on. Big assumption: if it scales well with multi-GPU, usable framerates might be achieved with dual 1080 or higher GPUs.

 

[mackerel]

I looked at my 4 core results with HT, and also CPU usage while the benchmark is running (at 4.2 GHz). I'm not sure I can draw a solid conclusion other than HT does give a small benefit. The benefit to score is of the ball park 2.5% with HT on. I take back my earlier speculation it might be using more than 4 threads. With HT enabled, the total CPU usage during a benchmark never went above 50%, so implying it is doing the work of up to 4 threads equivalent. If I turn off HT, the benchmark was using around 65% CPU, with up to another 5% used by non benchmark tasks.

I wondered if ram could be an influencing factor here. With HT on, I ran at 2666 and 2133. I noted benchmarks results had run to run variation here, so this time I ran it 3 times at each ram speed and took the average. The difference between runs was smaller than the difference due to ram speed anyway, but the average was just over 2% faster for the 25% faster ram speed. So again, there is a small influence here but not that significant.

If this becomes a competitive benchmark, it seems the old tricks of fast CPU and fast ram can gain you that fractional advantage, but it is not really significant in day to day use. Implication is this is still more GPU bound.

I still don't have any explanation for the behaviour I was seeing running 2 cores with and without HT, and as it is such a niche condition I don't think I'll keep digging.