I have been benchmarking a little bit, and I got some questions.

SamiiRoss

Gawd
Joined
Jan 8, 2012
Messages
708
I'm trying to break all the 4080 records in 3dmark on my CPU class.
My graphics score is way higher than any heavily OCd 4080 but when it comes to Physics and Combined scores i rank lower. that could be the CPU? or my ram is too slow? "3600mhz 64GB"

bandicam 2023-01-03 03-44-03-447.jpg





bandicam 2022-12-31 08-51-56-347.jpg
 
I doubt it's the ram - I suspect that's CPU, but hard to tell. It's only threaded to a point so you've got the right idea - how much does physics change if you crank the core a bit?
 
I'm trying to break all the 4080 records in 3dmark on my CPU class.
My graphics score is way higher than any heavily OCd 4080 but when it comes to Physics and Combined scores i rank lower. that could be the CPU? or my ram is too slow? "3600mhz 64GB"

View attachment 539103




View attachment 539104
The people with higher scores may have curve optimizer working for their 5800X3D, giving slightly better performance. And they also probably have their RAM very carefully tweaked. Some people are able to get DDR4 3800 or 3900 working, with 1:1 infinity fabric. And their RAM timings are probably really low.

Take out half of your RAM, to lower the stress on your memory controller. 64GB is too much for a benchmarking system.

And then tweak your RAM speed and timings. Also, see if your bios has an update for curve optimizer or at least undervolting.
 
thank you, I will be trying to get a good 2x8GB set, and start from there.
 
thank you, I will be trying to get a good 2x8GB set, and start from there.
Personally, I would try a 2x16gb set of dual rank dimms. Dual rank gives a performance boost over single rank. And virtually all 8gb DIMMS are single rank. Since you are limited in RAM speed by the infinity fabric, a single rank system won't be able to increase RAM speed enough, to catch a dual rank system. In theory. Sub and tertiary timings are also very important:



 
CPU Temps, Curve and RAM tweaks all contribute to performance.
  • 2x dual rank or 4x single rank DIMMS give more performance than 2x single rank (most 16gb dimms are dual rank and 8bg dimms are generally single rank). It does put more pressure on the memory controller though with 4x single rank, or 2x dual rank with 4x single rank being the most difficult. Timings also contribute a bit in benchmarks. How much depends on the benchmark.
  • The CPU boosts higher when the temp is lower. E.g. I get around 2-3% more performance out of my CPU at stock by keeping it at 55 degrees instead of 70 degrees in benchmarks (oversized custom loop instead of AIO).
  • Curve lets the CPU boost higher at the same voltage both giving higher clocks and lower temps. Curve will be a few percent better than straight PBO on the regular x CPUs. Effectiveness of curve highly depends on how good your sample is. Would expect the 5800x3D to gain even more than the normal x CPUs due to lower max voltage.
 
The people at the top on 3dmark are always running settings that are only stable for going through the benchmark. They are probably running 3800mhz cl 12 or something. You can also OC the 5800x3d using bclk
 
Back
Top