Let me to introduce you to AMD's new datacenter CPU, Rome! The bane of Intel!
Everyone clap clap!
Let me to introduce you to AMD's new datacenter CPU, Rome! The bane of Intel!
Bentley Reyes
Other urls found in this thread:
phoronix.com
twitter.com
Wyatt Ramirez
i thought this was a cherrypicked avx 512 workload using more than 400 watts of power
Tyler Hall
It's real
Isaac Gutierrez
looks like a terribly optimized cherrypick if there's no benefit whatsoever from extra processors and there's actually a regression from last-gen
Julian Green
Jeremiah Long
Intcel tried his hardest, yet failed..
Michael Stewart
phoronix.com
>Cherry picked benchmark because the previous cherry picked benchmark ctx_clock can no longer be used against Intel
Keep on being salty, AYYMDPOORFAGS
Ryan Barnes
You can actually see where faggot cocksucker OP cut the text away but accidentally left a bit behind.
Robert Peterson
Well I would still call it a cherry pick even if it's the opposite in AMD's favor because a vectorized scientific workload that gets close to nothing from doubling up on throughput smells a little fishy.
Hudson Evans
>datacenter
>only using 1 cpu
You can't possibly be this retarded.
Eli Price
Less is better people... Hate these benchmarks...
Justin Roberts
>2 old gold xeons doing better than two newer platinums
wat
lower is better wouldn't. make any sense either, with a single Epyc doing better than two of the same put together
Brody Ward
Read
Owen Lopez
read
especifically the part where I say it's weird how a single Epyc does better than two of them, and even a higher end
Chase Barnes
Keep seething cause you got scammed, nigfag
Ian Nelson
400 > 225
Connor Cooper
Can someone explain to a dummy why 2x EPYC 7502 preforms less than one EPYC 7502?
James Adams
If seconds is better than why the fuck is the 7502 performing better than the higher tier chips? even the fucking 2x 7502.
Owen Taylor
mfw you cant even cherrypick benchmarks because icel gets fucked in all of them
Evan Ross
Lower is better, seconds normalized by price.
Benjamin Robinson
Not every single time
>more cores = less time to complete
With every new thread there is an overhead in both memory, synchronisation and additional steps the program has to take. Also in ryzen architecture a lot of shenigans can happen since not all cores are equal and shared cache can also made very wild results.
Sometimes its faster to do small task as single thread than doing it in 8 threads and then wait for all cores to deliver subresults and then finally sum them.
Ian Morgan
Parallelization is hard in some tasks.
For example in computer chess cores will have to communicate with each other so they aren't analysing the same nodes/moves*, this has delay. Especially if it's a multi-socket CPU. Too many cores and the cores may be doing each other's work + the added delay of having to communicate with each other.
So even if you double cores it may not actually be doing anything of use.
Seems like molecular dynamics is also such a task.
The CPU with fewer cores can be clocked higher per core. So if the program can't properly use a lot of cores the higher clocked CPU will come out ahead.
*And it happens anyway.