NVIDIA GeForce RTX 3070 Performance In CUDA & OptiX Rendering

[PAGETITLE=NVIDIA GeForce RTX 3070 CUDA & OptiX Rendering Performance][Top Ad] The new GeForce RTX 3070 graphics card from NVIDIA doesn't hold many secrets, largely in part to the fact that the company revealed the specs for its first three Ampere GeForces from the get-go. That even includes the price, which is interesting, because it's rare when we're aware of a price more than a month in advance. Arguably, the RTX 3070 could be considered the most interesting of the three 3000 series cards released so far, because in a single generation, NVIDIA knocked a card like the $1,199 GeForce 2080 Ti down to this $499 RTX 3070. It's not often we see that kind of price-drop, but it's obviously a move that will be welcomed by many, since 2080 Ti performance (or better) is now available for far less money. As we've done with the 3090 and 3080 launches, we're going to kick off our coverage with a look at rendering performance here, exclusively in those renderers that revolve their lives around CUDA and OptiX - with the exception of Blender, which is our only test here to include Radeon GPUs. There's not too much to talk about here that hasn't been discussed ad nauseam over the last number of articles, but for those interested, here's a breakdown of NVIDIA's current lineup (including 1080 Ti for reference): NVIDIA's GeForce Gaming GPU Lineup Cores Base MHz Peak FP32 Memory Bandwidth TDP SRP RTX 3090 10,496 1,400 35.6 TFLOPS 24GB 1 936GB/s 350W $1,499 RTX 3080 8,704 1,440 29.7 TFLOPS 10GB 1 760GB/s 320W $699 RTX 3070 5,888 1,500 20.4 TFLOPS 8GB 2 512GB/s 220W $499 TITAN RTX 4,608 1,770 16.3 TFLOPS 24GB 2 672 GB/s 280W $2,499 RTX 2080 Ti 4,352 1,350 13.4 TFLOPS 11GB 2 616 GB/s 250W $1,199 RTX 2080 S 3,072 1,650 11.1 TFLOPS 8GB 2 496 GB/s 250W $699 RTX 2070 S 2,560 1,605 9.1 TFLOPS 8GB 2 448 GB/s 215W $499 RTX 2060 S 2,176 1,470 7.2 TFLOPS 8GB 2 448 GB/s 175W $399 RTX 2060 1,920 1,365 6.4 TFLOPS 6GB 2 336 GB/s 160W $299 GTX 1660 Ti 1,536 1,500 5.5 TFLOPS 6GB 2 288 GB/s 120W $279 GTX 1660 S 1,408 1,530 5.0 TFLOPS 6GB 2 336 GB/s 125W $229 GTX 1660 1,408 1,530 5 TFLOPS 6GB 4 192 GB/s 120W $219 GTX 1650 S 1,280 1,530 4.4TFLOPS 4GB 2 192 GB/s 100W $159 GTX 1650 896 1,485 3 TFLOPS 4GB 4 128 GB/s 75W $149 GTX 1080 Ti 3,584 1,480 11.3 TFLOPS 11GB 3 484 GB/s 250W $699 Notes 1 GDDR6X; 2 GDDR6; 3 GDDR5X; 4 GDDR5; 5 HBM2 GTX 1080 Ti = Pascal; GTX/RTX 2000 = Turing; RTX 3000 = Ampere Versus the 2080 Ti, the new RTX 3070 draws much less power overall - about 30W less on paper. The biggest difference beyond that is that the 3070 loses 3GB of the Ti's frame buffer, but for a card that costs much less than the Ti, we're not sure many will complain. That's especially so as the 3070 will definitively beat out the Ti in rendering in most every case, thanks to the ray tracing performance improvements seen with the Ampere architecture. It's worth noting that like the bigger GPUs, the RTX 3070 includes the new mini 12-pin power connector. Unlike those bigger cards, however, NVIDIA included a special single cable converter for this GPU, instead of the Y split adapter like the others. Even then, only half of those 12 pins are populated, so it's clear that this is not going to be a very power-hungry GPU. We'll let others evaluate whether or not the modest power implementation impacts overclocking at all. To be clear, we have gaming testing en route, and at this point, it's effectively half-finished. We're taking advantage of being able to launch with a ProViz look while the rest of the testing gets completed in the background, so if you're interested in the gaming aspect of this card, please stay tuned. For now, here's a look at the current workstation GPU test system: Techgage Workstation Test System Processor Intel Core i9-10980XE (18-core; 3.0GHz) Motherboard ASUS ROG STRIX X299-E GAMING Memory G.SKILL FlareX (F4-3200C14-8GFX) 4x8GB; DDR4-3200 14-14-14 Graphics NVIDIA RTX 3090 (24GB, GeForce 456.38) NVIDIA RTX 3080 (10GB, GeForce 456.16) NVIDIA RTX 3070 (8GB, GeForce 456.96) NVIDIA TITAN RTX (24GB, GeForce 452.06) NVIDIA GeForce RTX 2080 Ti (11GB, GeForce 452.06) NVIDIA GeForce RTX 2080 SUPER (8GB, GeForce 452.06) NVIDIA GeForce RTX 2070 SUPER (8GB, GeForce 452.06) NVIDIA GeForce RTX 2060 SUPER (8GB, GeForce 452.06) NVIDIA GeForce RTX 2060 (6GB, GeForce 452.06) NVIDIA GeForce GTX 1080 Ti (11GB, GeForce 452.06) Audio Onboard Storage Kingston KC1000 960GB M.2 SSD Power Supply Corsair 80 Plus Gold AX1200 Chassis Corsair Carbide 600C Inverted Full-Tower Cooling NZXT Kraken X62 AIO Liquid Cooler Et cetera Windows 10 Pro build 19041.329 (2004) All product links in this table are affiliated, and help support our work. OTOY OctaneRender 2020. We start with a result that might make some 2080 Ti owners scratch their head a little bit, especially if they happened to have purchased one in recent months. The lowly $499 RTX 3070 manages a score that's 16% higher over the RTX 2080 Ti. Naturally, the 8GB frame buffer might be a limitation for some, but there's no denying that some serious generational improvement has been made here. Let's see how that carries over to some real OctaneRender projects: These real-world results actually expose something interesting. In both cases, it seems clear that the 11GB frame buffer allows the 2080 Ti to get work done a little more efficiently, as the RTX Off render test crown belonged to 2080 Ti, whereas turning RTX On gave the RTX 3070 the kind of lead we expected. In this case, RTX On resulted in a score boost of 23% over the 2080 Ti - even better than what we saw with OctaneBench. Blender 2.90. The RTX 3070 continues to flaunt its capabilities well against the 2080 Ti here. Whereas the Ti managed to surpass the 3070's RTX Off performance in OctaneRender, the opposite is true here. Where the 2080 Ti rendered this BMW test in 43 seconds, the 3070 shaved that down to 36. Let's see how that carries over to Eevee: Well, well, well. What do we have here? We have the first test result that shows the new RTX 3070 lagging behind the 2080 Ti, hinting that our Eevee render might take better advantage of more memory than we realized. Eevee doesn't take advantage of ray tracing acceleration like Cycles can, or else we may have seen some stronger performance out of the 3070 here. That's not to say what we see is poor - not by any stretch. With its positioning, the RTX 3070 obliterates everything under RTX 2080 Ti. With a return to the Cycles renderer and the Classroom project, we see the RTX 3070 manage to pull ahead of the RTX 2080 Ti again, in either CUDA or OptiX (RTX On) rendering. It's OptiX where AMD will likely have a difficult time competing with its next-gen Radeons, unless we're in for some surprises. Even if AMD managed to largely catch up (or perhaps beat) NVIDIA's CUDA rendering performance a bit, the fact that RT acceleration exists will likely rain on the company's parade. Without RT acceleration, the 3070 would settle on a render time of 52 seconds, but because it exists, that's instead cut down to 34 seconds. That's hardly a small difference. Let's check out viewport performance: The 1080p test makes it obvious that we should be introducing a second more strenuous project to give us some different perspectives, because the CPU is clearly a bottleneck at that resolution. For that reason, we're interested in seeing if AMD's upcoming Zen 3 processors might break through this bottleneck a little bit, especially since AMD's chips have been favorable to Blender in the past few years (at least on the CPU side). At 4K, the RTX 3070 once again sits behind the 2080 Ti, but it manages to keep pegged to around 60 FPS, which is more than enough for this type of work. You could even get by with 30 FPS here, so it's not as though half of the stack under-performs - unless we're talking about the truly low-end parts. Chaos Group V-Ray 5. V-Ray shows off the RTX 3070 in quite a nice light, with both of the CUDA and OptiX renders showing huge performance gains over the RTX 2080 Ti. The fact that the 3070 keeps pegged to near the top of the chart in each result is really telling, because it's quite a good placement for a GPU that costs a clean $500. As before, once you enable OptiX, you gain even more of your time back. The chart above shows that if you have a powerful CPU, you could take advantage of heterogeneous rendering as well to speed things up, but as you will be able to see when comparing the two sets of results, OptiX can negate the need to introduce the CPU. Whereas our 18-core CPU combined with RTX 3090 rendered this project in 74 seconds, using OptiX instead drops that to 52 seconds. If you have a small CPU, you're likely to see even greater performance deltas. While V-Ray's standalone benchmark is based on the older 4.X version, it still scales well with current hardware, as evidenced above. Even without current-gen optimizations, the RTX 3070 topples the RTX 2080 Ti handily here, yet again placing third among the entire lineup, only behind its bigger siblings. Autodesk Arnold 6. Arnold seems to love memory, so the 8GB attached to the RTX 3070 should prove sufficient for many users. Based on previous-gen performance, had this RTX 3070 only included 6GB of memory, then these projects would struggle to render, which is why you see the RTX 2060 noted as "Way Too Long" in the Sophie render (the actual time was something around 1,500 seconds). Yet again, the RTX 3070 shines here, easily placing ahead of the 2080 Ti, and we guess it's fair to say ahead of the TITAN RTX, as well. That GPU has a bit of a memory advantage, but even so, these results continue to look great for NVIDIA's new card. Luxion KeyShot 9. We had considered testing KeyShot 10's beta for this article, but time got the best of us, so that will instead come in the near-future (although not likely the beta, since the final version is due soon.) Nonetheless, we continue to see the RTX 3070 deliver super-strong performance here, again beating out the top-end GPUs...

NVIDIA GeForce RTX 3070 Performance In CUDA & OptiX Rendering

Rob Williams