AMD Ryzen 9 7950X3D

[W1zzard]

because of videos like this, and threads like this.

No time to watch the videos, but people on the Reddit link are discussing "max" ? Please do let me know when you find any guidance how to make the CPU run at 5.25 all the time

No issue with scalar x10 afaik, the cores are artifically limited in max clocks anyway, 5250 iirc

 

[desiprofessor]

No time to watch the videos, but people on the Reddit link are discussing "max" ? Please do let me know when you find any guidance how to make the CPU run at 5.25 all the time

No issue with scalar x10 afaik, the cores are artifically limited in max clocks anyway, 5250 iirc

Yeah will do, if I find anything on it, I'll share that.

Thanks for clarifying about the scalar.

 

[Mussels]

Hey @W1zzard

Is the IF set incorrectly here?

The ram specs also don't mention ranks, which is as critical on DDR5 as it was on DDR4 - I lucked out buying 2x16GB recently for my ITX rig and it was single rank, and lost a good 10% performance.
Igorslab confirmed that's still an issue on AM5 (Well, it's free performance so it's a win for users - but it's an issue if you're comparing performance with a RAM setup that runs worse on that platform unknowingly)

Spoiler

Large change from just adding more ranks with the exact same setup otherwise
Igor seems unaware of 2033 being the sweet spot, but he did confirm 2167 is massively slower than even 2000 as buildzoid discusses below

Test setup says IF @ 2000, AIda64 says 1866 (And buildzoid claims a large difference at 2033, with higher values being worse - 2033 is the 1:1 equivalent at DDR5 6000)

Buildzoid has a YT video on this where there's an oddity where 6000MT/s on the RAM matches up with 2033MHz on the Fclk (NB clock in Aida64) for a large performance difference

"And the reason we're at 2033 and not 2067 or 2100, is because the infinity fabric has some weird speed spot effect especially at DDR5 6000"
He then goes on to confirm that higher speeds are not faster, and it's been confirmed with every AM5 board and CPU he's ever tested


Could you check into this, and see if it did in fact change off what you thought it was at?

 

[desiprofessor]

That error has been corrected in a recent bios update. 2000 mhz is good again. That being said I have mine set at 2133mhz, and overclocked to 6200mt/s. Latency is about 57-58 with my 32gb gskill hynix a die kit.

 

[Mussels]

That error has been corrected in a recent bios update. 2000 mhz is good again. That being said I have mine set at 2133mhz, and overclocked to 6200mt/s. Latency is about 57-58 with my 32gb gskill hynix a die kit.

I'm not sure about at 6200, but you may want to test and see if lower is faster for you, after seeing higher values be worst for both Igor and Buildzoid

 

[desiprofessor]

I'm not sure about at 6200, but you may want to test and see if lower is faster for you, after seeing higher values be worst for both Igor and Buildzoid

Yup I've tested it many times. 6200 is faster. I'm on Overclockers.net. We have some users with 6400. It's stable, and benchmarks are slightly faster as well. Read speed is also at about 92000, with write at around 92-96000.

Also have run benchmarks on Modern Warfare II, and SOTR.

 

[Mussels]

Yup I've tested it many times. 6200 is faster. I'm on Overclockers.net. We have some users with 6400. It's stable, and benchmarks are slightly faster as well. Read speed is also at about 92000, with write at around 92-96000.

Just confirming you understood my question
I'm not asking if 6200 is faster - i'm asking if 2133 is the fastest at 6200

I'm curious if theres a pattern or math to what combinations work
2000/6000 made sense if you just ran 1:3, but it seems to want to be one step above that according to Buildzoids results

 

[desiprofessor]

Just confirming you understood my question
I'm not asking if 6200 is faster - i'm asking if 2133 is the fastest at 6200

I'm curious if theres a pattern or math to what combinations work
2000/6000 made sense if you just ran 1:3, but it seems to want to be one step above that according to Buildzoids results

It truly depends on luck of the draw with your ram. I tried 2167 and the infinity fabric became unstable. Each notch I increased I would run Linpack xtreme 1.1.5, and look at the gigaflops at each run. You are looking for stability on this, and for the numbers not to drop. you are also looking for an increase in gigaflops at each uptick. Stable would be running at 700 to 800 consistently atleast for my ram. If it goes 700, 500, 600, 400, something like that, you know the IF is getting unstable.

Here are my ram timings, as well as my aida benchmark as well. This is completely stable.

A lot of the testing is based on buildzoid's video on infinity fabric.

 

[Mussels]

Ah, you're looking at the gflops - Aida64 results with the latency and other values is what i'm talking about

It's very similar to the 1:1 ratio on AM4, where it seems you can get some values (raw MB/s) higher at higher values, but other performance metrics go down when you go past the sweet spot

I just wanted to confirm if 6200 has a higher sweet spot than 6000, as if the values are consistent we can math that out and turn it into generic advice for AM5 users

 

[desiprofessor]

Ah, you're looking at the gflops - Aida64 results with the latency and other values is what i'm talking about

It's very similar to the 1:1 ratio on AM4, where it seems you can get some values (raw MB/s) higher at higher values, but other performance metrics go down when you go past the sweet spot

I just wanted to confirm if 6200 has a higher sweet spot than 6000, as if the values are consistent we can math that out and turn it into generic advice for AM5 users

I'd say 6000 is still the sweet spot. On ON, I'm still seeing plenty of people who can't hit 6200. I've tested enough on games and other apps to know atleast for me 6200 is better, I get slightly higher frames in most games and lower latency.

 

[Mussels]

I'd say 6000 is still the sweet spot. On ON, I'm still seeing plenty of people who can't hit 6200. I've tested enough on games and other apps to know atleast for me 6200 is better, I get slightly higher frames in most games and lower latency.

The combination of FCLK and ram, not either alone

6000 + 2033 is a sweet spot, but buildzoid didn't confirm if 2033 is the sweet spot for all RAM speeds, or if it varied as RAM speed goes up

I'm still on AM4 here, so i don't have access to a system to find out myself

 

[desiprofessor]

The combination of FCLK and ram, not either alone

6000 + 2033 is a sweet spot, but buildzoid didn't confirm if 2033 is the sweet spot for all RAM speeds, or if it varied as RAM speed goes up

I'm still on AM4 here, so i don't have access to a system to find out myself

Yup, I know what you are referring to. So yes, for default users, who don't want to do a lot of tweaking, the sweet spot is 6000 mhz and 2000 flock. The issue with it being 2033 instead of 2000 was corrected on a bios update. But as mentioned if your ram can handle it it can be overclocked and performance can be had.

 

[Mussels]

Yup, I know what you are referring to. So yes, for default users, who don't want to do a lot of tweaking, the sweet spot is 6000 mhz and 2000 flock. The issue with it being 2033 instead of 2000 was corrected on a bios update. But as mentioned if your ram can handle it it can be overclocked and performance can be had.

If it was just that small amount off due to an agesa math error, then it makes sense that 2:3 is the new 1:1
2000 Fclk + 3000 MHz DRAM (6000 MT/s)

Buildzoids video is a little older now, but he did experience it on several boards