RTX 4090 & 53 Games: Ryzen 7 5800X vs Ryzen 7 5800X3D

[ColdVergil]

That combination isn't terrible - it's going to be easier to run than 8 ranks (4xDR) while performance is going to get that uplift of 4+ ranks

Yeah for now I gotta settle with this, everything runs great but the heaviest stuff might drop some. Should I ask here or in that other thread which set of ram would be most ideal?

 

[Mussels]

Yeah for now I gotta settle with this, everything runs great but the heaviest stuff might drop some. Should I ask here or in that other thread which set of ram would be most ideal?

Ask in that thread for help optimising your settings for the RAM - changing the number of ranks will only lower your performance, not help it.

four ranks has about an equal uplift to running 400MHz higher, on AM4.

As an example from one of the DDR4 reviews here on TPU, theres a reason 2x16GB and 4x8GB are at the top of the charts, even with slower speeds and looser timings
AM4 runs worse past 3800 due to infinity fabric going out of sync, but when 3200 (4x8 and 2x16) C16 is beating 3200 C14 (2x8) you know it's the ranks

The top result is 3600 C14 2x16GB, which is the dream - but so expensive compared to C16 RAM it's not worth the 0.5% FPS gain from it

The blue line shows two 3600 C14 kits with minor differences in favour of the SLOWER kit, with capacity being the big change - at the time that article was written, 16GB sinkle rank sticks didnt exist so that performance uplift is purely from having two extra ranks

 

[ColdVergil]

Ask in that thread for help optimising your settings for the RAM - changing the number of ranks will only lower your performance, not help it.

four ranks has about an equal uplift to running 400MHz higher, on AM4.

As an example from one of the DDR4 reviews here on TPU, theres a reason 2x16GB and 4x8GB are at the top of the charts, even with slower speeds and looser timings
AM4 runs worse past 3800 due to infinity fabric going out of sync, but when 3200 (4x8 and 2x16) C16 is beating 3200 C14 (2x8) you know it's the ranks

The top result is 3600 C14 2x16GB, which is the dream - but so expensive compared to C16 RAM it's not worth the 0.5% FPS gain from it
View attachment 312622
The blue line shows two 3600 C14 kits with minor differences in favour of the SLOWER kit, with capacity being the big change - at the time that article was written, 16GB sinkle rank sticks didnt exist so that performance uplift is purely from having two extra ranks

Ok that took me a few re-reads but I think I'm getting the hang of it. Once again thanks for taking the time, I'll head up to that thread and post my specs. So if I got it right, single rank is best?

 

[Mussels]

Ok that took me a few re-reads but I think I'm getting the hang of it. Once again thanks for taking the time, I'll head up to that thread and post my specs. So if I got it right, single rank is best?

Single rank is best for higher clocks - but not highest performance.
Four ranks (2x dual rank or 4x single) is best on AM4, but that's not saying you cant work with 6 or 8 ranks - it just needs more SoC voltage


Look at the blue line between the two scores a 5.2% performance increase just for adding those two extra ranks.
The only reason people say NOT to do it is because they think the easier overclocking makes single ranks the better choice, when it's not.

 

[ColdVergil]

Single rank is best for higher clocks - but not highest performance.
Four ranks (2x dual rank or 4x single) is best on AM4, but that's not saying you cant work with 6 or 8 ranks - it just needs more SoC voltage


Look at the blue line between the two scores a 5.2% performance increase just for adding those two extra ranks.
The only reason people say NOT to do it is because they think the easier overclocking makes single ranks the better choice, when it's not.

Okay that's fair enough. SoC voltage though?

I also have another question, when doing the AIDA benchmark, I get 69. ns latency. Now, my brother who has the same exact ram, he gets 65 ns, is that a substanstial difference or how could I lower it?

 

[Mussels]

Okay that's fair enough. SoC voltage though?

I also have another question, when doing the AIDA benchmark, I get 69. ns latency. Now, my brother who has the same exact ram, he gets 65 ns, is that a substanstial difference or how could I lower it?

SoC voltage is the memory controller voltage on AM4.
Variance is normal to an extent since even the slightest difference will change those values, but that's exactly the value you want to work on lowering.

Better off discussing it in the Zen Garden thread rather than here, however.