Shadow of the Tomb Raider, DX11 Benchmark Last Change: 20/Jan/2023, 1554hrs. Whoami: Ian Mapleson http://www.sgidepot.co.uk/sgi.html http://www.sgidepot.co.uk/sgidepot/ Here I present collated results for a range of different systems for the built-in SotTR DX11 benchmark, testing with HIGH settings at different resolutions (the HIGH configuration matches the "High" settings used by Gamers Nexus for their GPU reviews, except they normally use DX12 - I won't be testing DX12 for a while yet). See after the tables for various observations. Key (all GPU clocks are given as 'Core/Shader/RAM' or 'Core/Boost/RAM'; a single number refers to just the core clock): RX 580 = PowerColor Red Devil RX 580 8GB, default 1190/2000 core/RAM. GTX 780 Ti = EVGA GTX 780 Ti 3GB (Reference cooler), stock @ 980/1046/1750 core/boost/RAM (stabilises under load at around 1125MHz). GTX 1060 = Gigabyte GTX 1060 6GB Windforce, stock @ 1557/1772/2002 core/boost/RAM (stabilises under load at around 1924MHz). GTX 1080 = PNY GTX 1080 8GB, stock @ 1607/1734/1251 core/boost/RAM (stabilises under load at around 1860MHz). GTX 1080 Ti = Palit GameRock GTX 1080 Ti 11GB, stock @ 1595/1709/1376 core/boost/RAM (stabilises under load at around 1974MHz). Tables are shown in order of AVG FPS; if there's a tie then the next priority is 1% LOW. 1080p, HIGH (GN): Total GPU GPU Cores / Clock VRAM Core RAM CF/ AVG MIN MAX 1% .1% GPU CPU Threads Speed HT GPU(s) MB MHz MHz SLI FPS FPS FPS LOW LOW Manufac. Motherboard Model Drivers PCIe RAM Configuration Notes ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Ryzen 5 5600X/75W 6 / 12 4525 ON GTX 1080 Ti 11264 1595 1376 NO 114.1 57.2 252.6 55.2 37.0 MSI B450 Tomahawk Max II 526.47 3.0 16GB DDR4/3600 CL18 Win10/Pro/64, RAM @ 18/22/22/42/1T, PBO Enabled @ 75W package limit, Boosting: 4.524GHz all-core, 4.65GHz single core] i7 6850K 6 / 12 4400 ON GTX 1080 Ti 11264 1595 1376 NO 111.5 47.4 230.9 52.5 19.4 ASUS Rampage V Extreme 441.66 3.0 32GB DDR4/3000 CL15 Win7/Pro/64 i7 870 4 / 8 4308 ON GTX 1080 Ti 11264 1595 1376 NO 66.9 25.1 143.5 33.6 14.9 ASUS P7P55 WS Supercomputer 445.87 2.0 16GB DDR3/2154 CL9 Win7/Pro/64 i7 870 4 / 8 4308 ON GTX 1080 Ti 11264 1595 1376 NO 64.5 26.7 126.2 28.6 16.6 ASUS P7P55 WS Supercomputer 445.87 2.0 16GB DDR3/2154 CL9 Win10/Pro/64 Ryzen5 3600 6 / 12 4000 ON GTX 1060 6144 1557 2002 NO 64.1 41.9 115.4 42.8 29.3 Asrock B450M Pro4 451.67 3.0 16GB DDR4/3200 CL16 Win10/Pro/64, 1c boost = 4.2GHz, all-core boost = 4GHz. i7 2700K 4 / 8 5000 ON GTX 780 Ti 3072 980 1750 NO 51.2 33.5 96.6 33.4 19.3 ASUS Maximus IV Extreme 411.70 2.0 16GB DDR3/2133 CL9 Win7/Pro/64 i7 2700K 4 / 4 5000 OFF GTX 780 Ti 3072 980 1750 NO 50.5 10.8 99.3 17.6 1.8 ASUS Maximus IV Extreme 411.70 2.0 16GB DDR3/2133 CL9 Win7/Pro/64 i7 2700K 4 / 8 4000 ON GTX 780 Ti 3072 980 1750 NO 50.2 33.4 94.9 29.5 19.2 ASUS Maximus IV Extreme 411.70 2.0 16GB DDR3/2133 CL9 Win7/Pro/64 2x XEON E5 2640 12 / 24 2800 ON GTX 780 Ti 3072 980 1750 NO 49.3 31.4 94.4 34.2 15.6 Kllisre E5 M2L-8D, Dual-S2011 441.66 2.0 64GB DDR3/1866 CL13 NUMA OFF, DX12/Win10/Pro/64 Max Turbos: 1c/2c = 3GHz, 3c/4c = 2.9GHz, 5c/6c = 2.8GHz. Idle = 1.2GHz. 2x XEON E5 2640 12 / 24 2800 ON RX 580 8192 1190 2000 NO 46.6 18.6 109.1 15.6 4.8 Kllisre E5 M2L-8D, Dual-S2011 19.12.2 2.0 64GB DDR3/1866 CL13 NUMA OFF, Win7/Pro/64 Max Turbos: 1c/2c = 3GHz, 3c/4c = 2.9GHz, 5c/6c = 2.8GHz. Idle = 1.2GHz. 2x XEON E5 2640 12 / 24 2800 ON GTX 780 Ti 3072 980 1750 NO 46.0 30.3 92.5 27.5 12.3 Kllisre E5 M2L-8D, Dual-S2011 441.66 2.0 64GB DDR3/1866 CL13 NUMA ON, DX12/Win10/Pro/64 Max Turbos: 1c/2c = 3GHz, 3c/4c = 2.9GHz, 5c/6c = 2.8GHz. Idle = 1.2GHz. 2x XEON E5 2640 12 / 24 2800 ON GTX 780 Ti 3072 980 1750 NO 45.5 19.5 104.3 23.0 12.4 Kllisre E5 M2L-8D, Dual-S2011 441.66 2.0 64GB DDR3/1866 CL13 NUMA ON, Win10/Pro/64 Max Turbos: 1c/2c = 3GHz, 3c/4c = 2.9GHz, 5c/6c = 2.8GHz. Idle = 1.2GHz. 2x XEON E5 2640 12 / 24 2800 ON GTX 780 Ti 3072 980 1750 NO 44.8 25.3 96.7 22.8 8.5 Kllisre E5 M2L-8D, Dual-S2011 411.70 2.0 32GB DDR3/1866 CL13 NUMA ON, Win7/Pro/64 Max Turbos: 1c/2c = 3GHz, 3c/4c = 2.9GHz, 5c/6c = 2.8GHz. Idle = 1.2GHz. 2x XEON E5 2640 12 / 24 2800 ON GTX 780 Ti 3072 980 1750 NO 43.9 18.6 100.9 20.9 12.8 Kllisre E5 M2L-8D, Dual-S2011 441.66 2.0 64GB DDR3/1866 CL13 NUMA OFF, Win10/Pro/64 Max Turbos: 1c/2c = 3GHz, 3c/4c = 2.9GHz, 5c/6c = 2.8GHz. Idle = 1.2GHz. XEON E5 2420 v2 6 / 12 2500 ON GTX 780 Ti 3072 980 1750 NO 43.9 17.8 93.9 19.9 9.7 Huananzhi X9D 411.70 3.0 16GB DDR3/1600 CL11 Win7/Pro/64 Max Turbos: 1c = 2.7GHz, 2c = 2.6GHz, 3+ cores = 2.5GHz. Idle = 1.2GHz. 2x XEON E5 2640 12 / 24 2800 ON GTX 780 Ti 3072 980 1750 NO 43.5 23.0 96.2 22.4 10.1 Kllisre E5 M2L-8D, Dual-S2011 411.70 2.0 64GB DDR3/1866 CL13 NUMA OFF, Win7/Pro/64 Max Turbos: 1c/2c = 3GHz, 3c/4c = 2.9GHz, 5c/6c = 2.8GHz. Idle = 1.2GHz. 2x XEON E5 2640 12 / 24 2800 ON RX 580 8192 1190 2000 NO 42.5 15.2 101.6 15.1 8.1 Kllisre E5 M2L-8D, Dual-S2011 19.12.2 2.0 64GB DDR3/1866 CL13 NUMA OFF, Win10/Pro/64 Max Turbos: 1c/2c = 3GHz, 3c/4c = 2.9GHz, 5c/6c = 2.8GHz. Idle = 1.2GHz. Core2Quad Q9550 4 / 4 2830 - RX 580 8192 1190 2000 NO 32.8 15.7 72.6 14.8 10.3 ASUS P5Q3 19.7.1 2.0 16GB DDR3/1333 CL9 Win7/Pro/64 Stock config, C-States active, CPU runs at 2GHz when idle. ******************************************************************************************************************************************************************************************************************* ******************************************************************************************************************************************************************************************************************* 1440p, HIGH (GN): Total GPU GPU Cores / Clock VRAM Core RAM CF/ AVG MIN MAX 1% .1% GPU CPU Threads Speed HT GPU(s) MB MHz MHz SLI FPS FPS FPS LOW LOW Manufac. Motherboard Model Drivers PCIe RAM Configuration Notes ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- i7 6850K 6 / 12 4400 ON GTX 1080 Ti 11264 1595 1376 NO 89.3 54.9 161.4 51.1 28.0 ASUS Rampage V Extreme 441.66 3.0 32GB DDR4/3000 CL15 Win7/Pro/64 i7 870 4 / 8 4308 ON GTX 1080 Ti 11264 1595 1376 NO 64.5 33.8 127.2 33.7 18.8 ASUS P7P55 WS Supercomputer 445.87 2.0 16GB DDR3/2154 CL9 Win7/Pro/64 i7 870 4 / 8 4308 ON GTX 1080 Ti 11264 1595 1376 NO 63.4 32.8 122.3 33.3 20.7 ASUS P7P55 WS Supercomputer 445.87 2.0 16GB DDR3/2154 CL9 Win10/Pro/64 Ryzen5 3600 6 / 12 4000 ON GTX 1060 6144 1557 2002 NO 44.1 33.3 72.5 33.4 25.2 Asrock B450M Pro4 451.67 3.0 16GB DDR4/3200 CL16 Win10/Pro/64, 1c boost = 4.2GHz, all-core boost = 4GHz. 2x XEON E5 2640 12 / 24 2800 ON RX 580 8192 1190 2000 NO 36.7 17.8 69.1 16.6 12.0 Kllisre E5 M2L-8D, Dual-S2011 19.12.2 2.0 64GB DDR3/1866 CL13 NUMA OFF, Win7/Pro/64 Max Turbos: 1c/2c = 3GHz, 3c/4c = 2.9GHz, 5c/6c = 2.8GHz. Idle = 1.2GHz. 2x XEON E5 2640 12 / 24 2800 ON RX 580 8192 1190 2000 NO 35.9 16.8 70.9 16.2 11.1 Kllisre E5 M2L-8D, Dual-S2011 19.12.2 2.0 64GB DDR3/1866 CL13 NUMA OFF, Win10/Pro/64 Max Turbos: 1c/2c = 3GHz, 3c/4c = 2.9GHz, 5c/6c = 2.8GHz. Idle = 1.2GHz. i7 2700K 4 / 8 5000 ON GTX 780 Ti 3072 980 1750 NO 35.7 25.4 61.6 25.6 10.4 ASUS Maximus IV Extreme 411.70 2.0 16GB DDR3/2133 CL9 Win7/Pro/64 i7 2700K 4 / 4 5000 OFF GTX 780 Ti 3072 980 1750 NO 35.6 16.7 61.6 22.3 1.7 ASUS Maximus IV Extreme 411.70 2.0 16GB DDR3/2133 CL9 Win7/Pro/64 i7 2700K 4 / 8 4000 ON GTX 780 Ti 3072 980 1750 NO 35.5 25.0 60.9 25.1 8.8 ASUS Maximus IV Extreme 411.70 2.0 16GB DDR3/2133 CL9 Win7/Pro/64 2x XEON E5 2640 12 / 24 2800 ON GTX 780 Ti 3072 980 1750 NO 34.7 23.3 65.9 21.3 11.9 Kllisre E5 M2L-8D, Dual-S2011 441.66 2.0 64GB DDR3/1866 CL13 NUMA ON, Win10/Pro/64 Max Turbos: 1c/2c = 3GHz, 3c/4c = 2.9GHz, 5c/6c = 2.8GHz. Idle = 1.2GHz. 2x XEON E5 2640 12 / 24 2800 ON GTX 780 Ti 3072 980 1750 NO 33.8 20.7 64.7 20.4 10.9 Kllisre E5 M2L-8D, Dual-S2011 441.66 2.0 64GB DDR3/1866 CL13 NUMA OFF, Win10/Pro/64 Max Turbos: 1c/2c = 3GHz, 3c/4c = 2.9GHz, 5c/6c = 2.8GHz. Idle = 1.2GHz. 2x XEON E5 2640 12 / 24 2800 ON GTX 780 Ti 3072 980 1750 NO 33.7 24.1 60.7 21.5 7.4 Kllisre E5 M2L-8D, Dual-S2011 411.70 2.0 32GB DDR3/1866 CL13 NUMA ON, Win7/Pro/64 Max Turbos: 1c/2c = 3GHz, 3c/4c = 2.9GHz, 5c/6c = 2.8GHz. Idle = 1.2GHz. 2x XEON E5 2640 12 / 24 2800 ON GTX 780 Ti 3072 980 1750 NO 32.9 22.3 60.1 21.6 5.4 Kllisre E5 M2L-8D, Dual-S2011 411.70 2.0 64GB DDR3/1866 CL13 NUMA OFF, Win7/Pro/64 Max Turbos: 1c/2c = 3GHz, 3c/4c = 2.9GHz, 5c/6c = 2.8GHz. Idle = 1.2GHz. XEON E5 2420 v2 6 / 12 2500 ON GTX 780 Ti 3072 980 1750 NO 32.9 19.7 60.3 17.8 6.2 Huananzhi X9D 411.70 3.0 16GB DDR3/1600 CL11 Win7/Pro/64 Max Turbos: 1c = 2.7GHz, 2c = 2.6GHz, 3+ cores = 2.5GHz. Idle = 1.2GHz. Core2Quad Q9550 4 / 4 2830 - RX 580 8192 1190 2000 NO 31.7 15.1 68.6 14.4 10.1 ASUS P5Q3 19.7.1 2.0 16GB DDR3/1333 CL9 Win7/Pro/64 Stock config, C-States active, CPU runs at 2GHz when idle. 2x XEON E5 2640 12 / 24 2800 ON GTX 780 Ti 3072 980 1750 NO 30.6 21.3 55.5 20.9 13.1 Kllisre E5 M2L-8D, Dual-S2011 441.66 2.0 64GB DDR3/1866 CL13 NUMA ON, DX12/Win10/Pro/64 Max Turbos: 1c/2c = 3GHz, 3c/4c = 2.9GHz, 5c/6c = 2.8GHz. Idle = 1.2GHz. 2x XEON E5 2640 12 / 24 2800 ON GTX 780 Ti 3072 980 1750 NO 28.9 21.0 54.7 20.7 10.9 Kllisre E5 M2L-8D, Dual-S2011 441.66 2.0 64GB DDR3/1866 CL13 NUMA OFF, DX12/Win10/Pro/64 Max Turbos: 1c/2c = 3GHz, 3c/4c = 2.9GHz, 5c/6c = 2.8GHz. Idle = 1.2GHz. ******************************************************************************************************************************************************************************************************************* ******************************************************************************************************************************************************************************************************************* NOTE: installing this game automatically updated the GPU drivers to 411.70, which is REALLY ANNOYING. As a result I had to redo some 2700K tests which had originally been done with 391.35. Oh well. The following discussion refers to the use of a 4GHz 2700K and 780 Ti at 1080p... I am not testing at other detail settings because it really doesn't affect performance that much, even though VRAM usage is greatly reduced, eg. at 1080p: LOW: 1822MB MED: 2177MB HIGH: 3055MB HIGHEST: 3053MB (the game throttling its usage? Maybe this happens at HIGH aswell? I'll check later) For example, using the LOW preset instead of HIGH only improves the average frame rate by 16%, while MEDIUM instead of HIGH was a mere 1% faster, suggesting the game is CPU-bound, as tech reviewers frequently state. Reported CPU usage at HIGH was about 52% with 7 or 8 threads loaded, though one thread tends to remain above 75%, with GPU usage at times dipping to 95%. One could infer though that it's not so much the "lack" of a powerful CPU as the use of a poorly optimised game engine which produces this behaviour, it's hard to be sure, though reviewers say the game does benefit from more than 8 threads, so it'll be interesting to see how both a 3930K behaves; certainly the 2420 v2 does ok, though its Lows suffer somewhat (DX12 may boost such XEONs more). Perhaps this is why the 2700K/5GHz numbers are barely any different to the 2700K/4GHz numbers. People often use SotTR at 1080p as a pseudo CPU benchmark because it's supposedly CPU-bound, but I can't help wondering if that's like judging a car's performance characteristics when the wheels are covered in mud. I remember similar discussions about Crysis as a GPU benchmark, because it hammered GPUs so hard, but Warhead was about 15% faster than the original Crysis engine. In the end though I suppose it is what it is. It would be ironic though if more threads was a greater benefit to the game than a higher clock, though presumably one could infer ideally one would have both, but really it shouldn't be necessary to have a CPU as powerful as 4GHz with 12 threads to run a modern game. Why are these game engines hitting the CPU so hard when, onscreen, much of the time not really that much is happening? Note that at HIGHEST, performance only dropped by 10% to 46.4fps avg, so it looks like the game works hard to avoid VRAM throttling, which is kinda cool, gotta give it creds for that. One thing I did notice was main memory usage was high, easily 8GB on a system with 16GB RAM (rising to 10GB at 1440p). I wouldn't want to play SotTR on a system with only 8GB RAM, nor with a GPU that has less than 3GB VRAM (which doesn't bode well for numerous older cards, though the LOW preset would keep usage within the 2GB limit of cards like the GTX 680). SotTR is certainly heavy on the VRAM, which is ironic for a benchmark that is so often used to test CPU performance. I would simply say that, if you want to play SotTR, then use a GPU with at least 3GB VRAM; that way you'll at least be able to select the HIGH preset regardless of how your main CPU might affect overall frame rates, assuming the GPU has enough raw performance anyway (eg. a GTX 460 2GB might be stretching things a bit). Interesting to note that, unlike RotTR, the Q9550 and RX 580 were able to run this game ok, though it's definitely CPU-limited (maxed out most of the time). Testing the dual XEON 2640 system under Win10 didn't change the avg fps much, but it did widen the Min/Max gap somewhat, though Lows improved a little. This is still with DX11 though; testing with DX12 showed no significant change at 1080p (with NUMA on) but it hurt performance a lot at 1440p (avg fps down 12%). Thus, it's a bit complicated, at least with dual-CPU systems anyway. I will have a separate table for DX12 results but I have included just those for the dual-S2011 board in order to explore NUMA issues aswell as DX11 vs. DX12 since games vary how they respond to each of these modes, separately or combined. At 1080p, turning NUMA off improved DX12 results even further, resulting in an overall speedup from DX11 to DX12 with NUMA off of 12%. At 1440p though the effect is reversed, the DX12 result is even slower (grud knows why). The entire ordering of results is flipped between 1080p and 1440p, so perhaps testing at 4K would show even greater disparity (I'm still looking into what 4K monitor to buy; it has to be something that supports sync-on-green aswell so that I can use it with SGIs too). Thus, in pure performance terms, at 1080p the best results are with Win10/DX12 and NUMA off, while at 1440p the best results are with Win10/DX11 and NUMA on (but the gap is smaller and probably within error margins), so overall I would just leave NUMA off - other games seem to favour this mode in general. ******************************************************************************************************************************************************************************************************************* CPU-Zs: i7 2700K @ 5.0: http://valid.canardpc.com/show_oc.php?id=2334534 XEON E5 2420 v2: http://valid.x86.fr/wd97e9 Q9550 @ 2.83 stock: http://valid.x86.fr/yma3su 2x XEON E5 2640: http://valid.x86.fr/4wp69g i7 870 @ 4308MHz: http://valid.x86.fr/n40mdc Ryzen5 3600 @ 4GHz: http://valid.x86.fr/n9t6wt Ryzen5 5600X @ 4650MHz (PBO 75W): http://valid.x86.fr/wftacb i7 950 @ 4.26 http://valid.x86.fr/lumtze