Hey everyone, I recently upgraded my 2060 KO to a shiny new 4070, so I can get a nice performance bump in blender.
I ran a benchmark test (v 3.5.0) and got a score of 4671. This lands my card, according to the open data table, closer to a 4060 Ti or a laptop GPU. My card is expected to deliver closer to 6000. I do understand it’s a median and not a mean, but a deviation of this much from what is expected? Is there something I’m missing, or is my card faulty?
FWIW, my previous card gave me a score of 1782, which lands it just short of 2060 territory - but that is also quite weird because that 2060 KO was supposed to be like a 2080 in blender, because it had a binned TU104, at least according to gamersnexus. I chalked it up to CyclesX not using this card to its full potential anymore, and I wasn’t able to find any old benchmarks for this card so I didn’t think much of it.
So - I’m consistently getting lower scores than I’m supposed to (about 20-30% being left on the table as far as I can see). Is this a setup problem or am I just misunderstanding the whole thing? TIA.
A couple of things come to mind. First would be CUDA vs OptiX. Possible that for some reason you maybe using CUDA but then comparing it to a OptiX render result. If that’s the case I’d expect a 30%+ difference.
The other thing to do is grab GPU-Z and look at the stats while running a benchmark and see what looks wrong. Like lower clock frequency or low power or something else. That will at least give some clue as to why the scores are lower.
Remember: different cards with the same GPU could score differently… also your RAM speed also do have some influence and also what’s loaded or running on you PC during the test… that’s also the reason why benchmarks shoudl be run mtuliplt times… but then again if alredy loade in th VRAM the second run can be quicker… (also the speed of HD SSD etc…)
So… I closed off all the other applications and ran the benchmark again, a couple of times. It seemed to give me a score of 5418, which is about 10% off the median … I use a KFA2/Galax card, and the temps are stellar - the card is seeing about 85-93% use during the benchmark and the temps are quite comfortable. I would assume card differences would net me about ± 300-ish points? is 600 within reason to mean I just lost the GPU lottery?
Hi, thanks so much for the pointers! I was able to get the scores to 5400-odd by closing everything else, but I’m still missing around 600 points. Bear with me here please, I’m not too versed in this.
I’ve attached a GPU-Z screenshot here as per @thetony20’s great suggestion during the benchmark. Here is the reference for my exact card. The TDP is 200W and they recommended a 550W psu, which is what I have coupled with a relatively low powered CPU.
Do you see anything off here? From what I can see, the clock speeds are higher than normal, but the TDP is not being hit. GPU usage is not maximal. I am getting perfcapped by VRel, which as far as I know is because I haven’t overclocked this card.
My processor is a i5 11400, which as far as I know has 20 PCIe4 lanes. I have 16 lanes for this GPU, and 4 for another SSD (not in use here) The benchmark and OS is running off of a gen 3 m.2 SSD. I don’t know if that SSD is on the PCH bridge or the CPU. Tangent: Should I just go ahead and move my stuff to the Gen 4 SSD, or will that cause other caps if both the GPU and SSD are under heavy load?
Yeah, the card is hitting its voltage limit before it hits anything else, which I think is a common thing for the 40 series cards.
If its a dual BIOS card, maybe double check it is switched to performance rather then quiet.
Other then that, could use MSI Afterburner to try and adjust some settings to see if it makes a difference.
I wouldn’t bother, it would make next to zero difference to actual render speed and the GPU. Once the data is loaded into VRAM, the rest of the system has very little impact.
dito… me mentioning hdd/sdd was more between some hdd and compared to sdd… they are so fast you have to have a very lousy sdd and buy a fast one to see any (slightly) difference… and 10% are at least something
…and i wouldn’t bother for every litle percent… but that’s only me two cent
…
