Comparing Shadow of the Tomb Raider benches of different GPUs

[J7SC_Orion]

In case you're wondering about your GPU and whether to upgrade...some simple comparisons (incl. 1x, 2x 2080 Ti, 6900XT, 3090) in spoiler below using the Shadow of the Tomb Raider DX12 Benchmark at 4096x2160. 

 

FYI as background, I use three different machines for work, for play and our media room. That includes a TR2950X w/ 2x 2080 Ti, a 3950X w/ a 6900XT and a 5950X w/ a 3090 Strix OC. I upgraded the latter two over a year ago as I found great deals (below MSRP) for the 3090 Strix and 6900XT before the horrible 'price hump'. The latter two are connected to a 48 inch 4K 120 Hz OLED, while the 2080 Tis power a 55 inch 4k / 60 Hz IPS. All benches were done with max eye candy ('highest' settings), including HDR. v-sync was off to level the playing field, though I normally have it on. Each system had a mild 24/7 oc ('game' instead of 'bench') and used the stock bios instead of higher-powered ones, so its not a 'max bench result' comp. Also, all GPUs below are custom w-cooled. Re. the 6900XT at 4096x2160, it obviously did not benefit from the DLSS which is bundled w/ ray tracing in the STR benchmark. I do believe that the ray tracing was used as default by the 6900XT, but am not entirely sure.

 

3090

 

 

2x 2080 Ti

 

 

1x 2080 Ti

 

 

6900XT

 

 

[Sir Beregond]

Hmm that's an interesting result for the 6900XT.

[pio]

Doesn't DLSS operate like FSR does for AMD, in the sense that it lowers the resolution during the gameplay / bench?  Wouldn't it be more accurate to disable DLSS for the Nvidia cards since the AMD is doing actual native res?

 

I'm pretty happy with my 6900XT outside of a few RT titles at my native res.  And even those RT titles, I can get them playable in freesync range with FSR or RSR.  Which.....appears to be similar with Nvidia's DLSS anyway, so I'm fine with that.

[J7SC_Orion]

1 hour ago, pioneerisloud said:

Doesn't DLSS operate like FSR does for AMD, in the sense that it lowers the resolution during the gameplay / bench?  Wouldn't it be more accurate to disable DLSS for the Nvidia cards since the AMD is doing actual native res?

 

I'm pretty happy with my 6900XT outside of a few RT titles at my native res.  And even those RT titles, I can get them playable in freesync range with FSR or RSR.  Which.....appears to be similar with Nvidia's DLSS anyway, so I'm fine with that.

 

...it's not a perfect comparison, but the Shadow of the Tomb Raider benchmark only has ray tracing AND DLSS as a single option and I definitely wanted to test everything max, including with ray tracing, at 4096 x 2160. As stated before, I believe that the 6900XT used ray tracing but not Super Resolution 'on' in the driver. I reran the 6900XT just now with Super Resolution engaged, per below - not much of a difference. Still, quite a respectable result. The 6900XT/6950XT have a tougher time with 4K in general, due to the much narrower VRAM bus (even w/ InfinityCache), but at 1440p and below, they start to pull ahead.

 

The other thing worth noting is the peak power consumption !

 

---------- max per Shadow of the Tomb Raider bench  --- max stock bios --- max custom bios

 

Strix 3090    ---                 413 W                                                ---     450 W             ---   525W/1000W

6900 XT       ---                 298 W                                                 ---     303 W            ---   450 W / 600 W (MPT)

2080 Ti (1x) ---                  361 W                                                 ---     380 W            ---   380 W

2080 Ti (2x) ---                  718 W                                                ---      760W               ---  760 W

 

[Sir Beregond]

45 minutes ago, J7SC_Orion said:

 

...it's not a perfect comparison, but the Shadow of the Tomb Raider benchmark only has ray tracing AND DLSS as a single option and I definitely wanted to test everything max, including with ray tracing, at 4096 x 2160. As stated before, I believe that the 6900XT used ray tracing but not Super Resolution 'on' in the driver. I reran the 6900XT just now with Super Resolution engaged, per below - not much of a difference. Still, quite a respectable result. The 6900XT/6950XT have a tougher time with 4K in general, due to the much narrower VRAM bus (even w/ InfinityCache), but at 1440p and below, they start to pull ahead.

 

The other thing worth noting is the peak power consumption !

 

---------- max per Shadow of the Tomb Raider bench  --- max stock bios --- max custom bios

 

Strix 3090    ---                 413 W                                                ---     450 W             ---   525W/1000W

6900 XT       ---                 298 W                                                 ---     303 W            ---   450 W / 600 W (MPT)

2080 Ti (1x) ---                  361 W                                                 ---     380 W            ---   380 W

2080 Ti (2x) ---                  718 W                                                ---      760W               ---  760 W

 

What kind of FPS do you get on the 6900XT with RT&DLSS turned off at 4k?

 

Yeah I would have liked to see a version of the 6900XT with a 384-bit bus. I mean could have even met in the middle and done a 320-bit with 20GB GDDR6 + the Infinity cache and I'd imagine that would have really helped it at 4k.

[J7SC_Orion]

3 hours ago, Sir Beregond said:

What kind of FPS do you get on the 6900XT with RT&DLSS turned off at 4k?

 

Yeah I would have liked to see a version of the 6900XT with a 384-bit bus. I mean could have even met in the middle and done a 320-bit with 20GB GDDR6 + the Infinity cache and I'd imagine that would have really helped it at 4k.

 

...4K and ray tracing are least favoured for the 6900XT - nevertheless, it can hold its own, settling in somewhere around a RTX 3070 - 3080. In Port Royal (which is 1440, ray tracing), my 6900XT / 3950X hits 11902, which is incidentally the fastest such combo at 3D Mark. Then again, my 3090 / 5950X combo hits 15,531 in that bench...but in Time Spy (non-extreme, DX12, 1440), the 6900XT / 3950X hits over 24.1k graphics score, while the 3090 / 5950X comes in at just over 23k max...

 

...but back to Shadow of the Tomb Raider - here are two DX runs of the 6900XT. Both with the same mild oc clocks as above, but one of them has the '6950XT' PL and a few other mods at 450 W limit instead of 300 W limit (actual was 413 W for the second run):

 

 

 

I then switched both cards to DX11 / 4096x2160 and no DLSS/RTX...I also ran the ancient but fun Unigine Valley. Enhanced bios for the 6900XT but same mild oc  clocks as before...3090 also at its mild oc from above, but stock bios. The gap between the two gets a lot closer w/o ray tracing and if the 6900XT is allowed to draw similar power...the 3090 still has the edge (never mind its custom bios not used here, or full-tilt clocks), but they're in the same ballpark...the RDNA2 VRAM bottleneck shows a bit for the AMD at this resolution even with the edge in rasterization at lower resolutions. 

 

 

...and finally, Unigine Superpostion 4k w/ full clocks and stronger (but still non-XOC) bios

 

 

 

 

Edited May 17, 2022 by J7SC_Orion

[J7SC_Orion]

...quick addendum: I tried the Shadow of the Tomb Raider bench in DX12 with free sync enabled on the 6900XT (as opposed to 'disabled' above). The difference was only one FPS. That said, on a big OLED, everything is perfectly smooth and DX12/HDR/sync is a real treat to watch; big visual difference to DX11...

[damric]

I find that Shadow of the Tomb Raider is also a good benchmark for comparing cache and memory, bandwidth and latency for the whole platform, not just the graphics card.

 

I ran this the other day while testing a Ryzen 5500 with PBO +200x104 base clock. Ran the same card settings that I ran with a Ryzen 3600 for comparison (Vega64 1750MHz +50%power limit).

 

[J7SC_Orion]

1 hour ago, damric said:

I find that Shadow of the Tomb Raider is also a good benchmark for comparing cache and memory, bandwidth and latency for the whole platform, not just the graphics card.

 

I ran this the other day while testing a Ryzen 5500 with PBO +200x104 base clock. Ran the same card settings that I ran with a Ryzen 3600 for comparison (Vega64 1750MHz +50%power limit).

 

 

 

Good idea...The Shadow of the Tomb Raider bench has that little table at the bottom right which does provide some CPU and related sub-system data anyway, but for the fun of it, I ran my 3950X and 6900XT at the same 1080P / 'lowest' setting...

 

...ohh, my eyes - game looks very different  ...I'll try to run 1440 'highest' on both this setup and the 5950X / 3090 later to add some data points to the OP.

 

[damric]

I can't run in 4096 but I can run in 3840. You must have an inch wider screen or something

 

 

[J7SC_Orion]

On 18/05/2022 at 12:01, damric said:

I can't run in 4096 but I can run in 3840. You must have an inch wider screen or something

 

 

 

...screen is 48 inch C1 OLED - it's not narrow 

 

Here are some 1440 results for the 3950X / 6900XT and 5950X / 3090...at this resolution, the difference in CPUs plays more of a role, but the right-hand column in the lower right table suggests that the GPUs are fairly close, once taking the RTX/DLSS into account...apparently, there's an AMD FX CAS patch available for Shadow of the Tomb Raider, but for now at least only for the full game, not the demo...Then again, to help 'equalize', the 6900XT had a better -than-stock PL, and both cards pulled around the 410W mark for this 1440 test according to HWInfo...