Vega series GPU performance

Hi,

Wanted to start a thread about the new Radeon RX Vega GPU-series performance in DCS world.

I personally just put a waterblock/custom loop on my Vega 64 and overclocked it to its limits. It goes beyond GTX 1080 on Firestrike benchmarks, but the performance seems to be way below GTX 1070 in DCS.

My initial tests are mostly first impressions from flying 10 minutes both in 1.5 Caucasus and 2.0 Nevada singleplayer using M2000C in A-A instant actions. Both started from medium/high altitudes where FPS was still below 45 using VR, Oculus Rift CV1 to be exact. On GTX 1070 I mostly had 90 FPS (Rift maxed) at high altitudes with fairly low traffic. I tracked the GPU clock/memclock graphs using MSI afterburner and the clocks seemed to be only 300-1200 Mhz on core while max setting was at 1700Mhz, GPU load was low ~50%. Memory performed the same where memclocks were a step over idle clocks at 500 Mhz, while max was 1100 Mhz.

I wonder if ED could get a hands on with some Vega cards and try to see if you find the same scenario, and if you could do something to increase GPU usage. The issue might be in AMD drivers too, and generally AMD lacks in performance using DX 11 or older APIs.

I'm eager to hear if some of you have also bought Vega GPU and how it performs.

On which CPU did you made this test? AMD driver loves overclocked and powerful CPUs.

I would also add that Firestrike isn't a good representative of simulators kind of workload

If the clocks for gpu and vram fluctuate that much while you are flying DCS then that is not good.

The review I have in mind had the card running between 1400-1600 in all apps, DX11/12/Vulcan, tho it was custom watercooled and the range might be 1300-1500 on air or such. I just doubt it should ever fall below the 1300/1400 lower threshold, regardless what the CPU does. That's how my Nvidia works in full OC and 3D mode.

What's your CPU RAM SSD ??

it's in german, but the graphs speak any language and are labeled in english: http://www.tomshardware.de/vega64-wasserkuhlung-benchmarks-temperaturen-ubertaktung,testberichte-242383.html

3d mark and most synthetic GPU Benchmarks only test the GPU and not the GPU+CPU at the same time.

While Firestrike IS a DX11 Benchmark, it's Purely a Shader/Post Processing Benchmark and not a Geometry and Object Count Benchmark.

Hands on w/ Vega Cards is a high cost idea right now, due to availability and mining surges, the GPU's quickly sell out then come back 1.5x or higher their MSRP.

I was scheduled to get 2 cards to review in mGPU Modes for AMD Gaming website, that job was cancelled by the management staff due to cost and short supply.

My CPU is 1800X @ Stock settings (4.0 / 4.1 GHz on single core) and 3200 MHz DDR4 RAM. When I changed from 3570k @ 4,5 GHz to 1800X @ 4.0 I didn't notice any jumps in performance. Maybe the 1800X is a little slower which was smoothed out by switching to better memory. Can't tell.

I haven't had time to test a lot but I did another fast test with the Mirage where FPS jumped to over 90 when flying straight at low altitudes even if I moved my head in Rift. But when I started to pull G's FPS instantly dropped under 45. I didn't really see any other amounts with DCS's FPS counter. It was good to at least see >90 FPS and maybe the performance can improve in later drivers, though I think AMD doesn't really do anything specifically for DCS.

With GTX 1070 FPS transitions seemed a lot smoother and I can't remember it being tied up with how I moved the aircraft, but moving my head had some impact. This seems logical as your view changes and GPU should load new textures stress its memory. But with Vega rapidly moving your head didn't seem to drop FPS nearly as much as pulling Gs and slowly moving the aircraft around.

I know the synthetic 3dmark isn't the best point of reference for a single game, but it should show some area of how the card could potentially behave. Every game has its own preferences.