osecare.blogg.se

Vray render benchmark
Vray render benchmark









vray render benchmark

To me, if I lay down serious cash on a server and it's not performing as I expect out of the box, if support can't be quick about answering it, they're shooting themselves in the foot. How long have you guys had the server? Think about the support received so far, is it worth the pain to deal with this now? 2 years down the road? Is it too late to swap out for something from a server vendor like Supermicro instead? These are serious considerations now on a platform that hasn't managed to impress you out of the box. We also have Threadrippers in the office which in some cases are rendering frames faster than the Epyc! Oh and we are running 2400mhz confirmed in bios which is correct for our memory sticks.

vray render benchmark

Waiting almost a week for them to figure out whether or not we have custom fan control and how to access it. I will reach out and ask but Asus has been painful. The bios on this Asus board is extremely bare.

vray render benchmark

PS: Chip clock for EPYC is not just temperature controlled, it is also power controlled, you may have bios settings to enable/disable 'green' modes Make sure bios understands the ddr4 you have installed (eg shows the speed correctly), I have had issues with motherboards downgrading memory speed because they didn't recognize the ram on AMD systems. In theory, the EPYC has around double the cache as the Xeon (64mb vs 35mb), but you don't say what the main memory speed was on the xeon. I lean this way because rendering uses tons of memory, but benchmarks often work with small test images that use less memory, so you often see them appear much faster than reality. If the temps are staying the same, and it is downclocking, then it has become memory bound, or disk bound, I am leaning toward memory, the flaw with 'more threads/cores is faster' is they are still all fighting over the same memory bandwidth, and when you run out of cache, you drop down to a very slow main memory speed that is almost independent of core count.











Vray render benchmark