Hello !
I have my hands on new CPU week ago feed my scene and tested 2990wx with renderman and it's deadly slow .
numbers are just terrifying .. average Intel CPU just easily outperforms 32 cores
I'm looking all over the internet and it seems that memory bandwidth is an issue , I'm not sure but most likely it's how AMDs new CPU treats and lets data from memory to flow into CPU
thank god I didn't buy that CPU for rendernode month ago and decided to make some research .
So , let's do some math !
1) Take a look at that topic here :
https://www.pcworld.com/article/3298859/components-processors/how-memory-bandwidth-is-killing-amds-3...
2) I took that from renderman forum , it requires registration access so I saved it to PDF and attached here , it's addressing same issue I'm talking about .
frankly had my eyes also on epyc CPUs , and also hope on upcoming AMD Rome , but if it will continue like that I don't know .
when someone buys 10k$ dual socket PC he can expects to multiply render speed by factor of 5-10x compared to average cpu
when you're advertising your products and doing that Cinebench 15 tests , they are small , all bench scenes are small , and it flies from RAM to CPU just like in fraction of second but when you're doing (rendering) heavy content , then it bottlenecks the whole calculation .
Is that really AMDs way to handle memory bandwidth ?
Will Rome CPUs have same issues ? because if it will it means in heavy rendering 2 or 3 average intel CPUs will outperform 2 dual socket Rome
I was thinking how to ask long in short, and only this comes to my head
What is going on ?