Ryzen 3000 owners census (questions and help)

hAyO

Updated, infinity fabric theme with Ryzen 3000

Fassou

About idle voltages on Windows, on Ryzen 3: https://www.reddit.com/r/Amd/comments/cbls9g/the_final_word_on_idle_voltages_for_3rd_gen_ryzen/

Salu2!

Obione

I just ran some Cinebenches with manual OC weis. With a boost of 4.2 the PBO loses much of its appeal.

Volta Frec Temp
1.32 4125 81
1.28 4100 79
1.27 4075 77
1.25 4050 76

Rounded Vcore (CPU-Z). The first time I passed CB R20 at 1.3 it was OK at 4125, this time it has blown up at the end. It's a very fixed voltage and although it's a quick cuttest, it gives me the clue that there's no clear sweet spot although for now I put the target at 4.1 pending to see how the homeopathic adjustments of this board behave, which you touch voltage offset or different LLC levels and it ignores them. Anyway, it's reminding me of my MSI P67 which was just as messy as this Lite BIOS.

whoololon

Piece of thread. :thumbsup_tone2:
alt text

Obione

Well, I'm sticking with 4.1@1.32 both are stable, after testing Prime custom and OCCT. Let's see if they release the new Agesa because the BIOS is dumb on many boards and the turbo doesn't work well among other things. When they are more debugged, I'll get into the turbo and dynamic with a little undervolt, 4 in multi with boost to 4.2 should perform better in games that don't parallelize well like my little car simulator.

It is said that with fixed voltage you should not go over 1.32volts, with PBO the peaks of 1.4X are OK.

rul3s

Here another 3700x + Asus x470 Strix F.
I'm still testing with the RAM but for now I'm doing 3666@16-15-15-28-42 with only 1.35v (1000% hci stable and TM5)

@obione said in Census of Ryzen 3000 owners (questions and help):

Well, I'm staying with 4.1@1.32 both fixed, after testing Prime custom and OCCT. Let's see if they release the new Agesa because the BIOS is dumb on many boards and the turbo doesn't work well among other things. When they are more debugged, I'll get into the turbo and the dynamic with a little undervolt, 4 in multi with boost to 4.2 should yield more in games that don't parallelize well like my little car simulator.

It is said that with fixed voltage you should not go beyond 1.32volts, with PBO the peaks of 1.4X are OK.

My 3700x does 4.1GHz with 1.275v, tested 2h of realbench and 30minutes of prime95, although it was just a test, normally it goes with the PBO that works wonderfully.

garfield

Hello:

Another one joins the 3000. Mounted on the x470 I had before, but with an H150i and Ripjaws V memory with hynix MFR chips 3200@3466 16-18-18-18-40 at 1.39v. Curious thing with these memories: more voltage, more errors. All according to the Dram Calculator for Ryzen. Moreover, according to that program, they should not work at more than 3448 16-18-20-20-40.
Latencies of 70ns. I don't think I can get much more. At 3533 the prime fails. But I can't complain about memory.

Curious thing that several of us have noticed with MSI: before each bios change you have to reset it, because it gets stuck. According to what Rul3s said it's because the values that the bios stores for the memories may not be compatible with the new versions. I must say that the last one for my board already has agesa 1.0.0.3ab, and that as we have seen there are not many changes with the memories. Maybe it's because they will be adding new chips that are now coming out on the market.
I have also noticed that compared to the 2700x the memory latencies have increased. I have the other computer with an x370 and the 2700x I had before, with some Micron B-die Ballistic Sport memory, at 2933 cas 16-16-16-16 and they give 65ns. Less than mine at much less speed. I don't know if it has to do with the chiplets, but I think AMD has a lot to improve in that area. I suppose that in the ryzen 4000 they will already have it solved, because in these it still stumbles a lot.
Last week I got some B-die, but one of the modules came bad, so I changed them for some Micron E-die Ballistic Elite 3600 cas16. According to what I have read they go up very very well. I have seen them at 4000 cas14-14-14-32 1T. We will see how they go, and if I don't return them to amazon and get some FlareX. For tinkering that won't be the case.

I forgot. My board allows you to change the infinity fabric bus asynchronously with the memory. I don't know for what, because the normal thing is that the memory goes the same as the infinity so that the yield is optimal. I suppose it will be for those who have very fast memories can use them to the maximum and the IF to the limit in which the dividers make it work at half. Or that depending on the processor or the board the IF can work at more or less speed. The truth is that until I get some memory that exceeds the 3733 that they say is the maximum of the IF I won't be able to check it. Let's see if they arrive this week.

Greetings

Obione

Sep garfield, you can overclock IF above the sync with memos and something can be scratched. And sep, the latencies are worse but, the bigger caches compensate for it. With Ryzen 2 the impact of the memories is not so big and from 3000MHz it scratches little, but it scratches. In fact, with that of holding more RAM clocks, the geeks have equalized the response against Ryzen +. They are working with B-Die's tightening the timings as before but, now at 3600 getting the same 6Xns.

Mr rul3s: you are touching my balls. Don't rub in my face how good your x8 is I already know how well it works for you and I don't need you to remind me that your beast, is more better than mine

rul3s

@obione said in Ryzen 3000 owners census (questions and help):

Sep garfield, you can overclock IF above the sync with memos and you can scrape something. And sep, the latencies are worse but the larger caches compensate for it. With Ryzen 2, the impact of the memory is not so great and from 3000MHz you can scrape a little, but you do scrape. In fact, with the fact that they can handle more RAM clocks, the geeks have equalized the response against Ryzen +. They are working with B-Die's tightening the timings like before but now at 3600 getting the same 6Xns.

Mr rul3s: you're touching my balls. Don't rub in my face how good your x8 is because I'm already aware of how well it works for you and I don't need you to remind me that your beast is better than mine

Ahahaha, but I don't consider mine good either xD
What I will leave here is an excel with memory configurations and their results, it can be very useful for you especially if you have bdie!

https://docs.google.com/spreadsheets/d/19M_eLP7i9eIWGGXaYtrzZD41fx0jMi4wfBdQvrbFOPI/edit?usp=sharing

I'm currently running that config 3666@16-15-15-15-28-42 and it's killing it, and I only need 1.375v in ram.

P.D: I'm still testing and it seems that with 1.275 it also does 4.1GHz

rul3s

Well, I definitely conclude the tests and for now I'll leave it with the following configuration:

OC Manual 4.15GHz
1.25v vCore (llc4) + 1.075v vSoc (llc3)
RAM 3666 16-15-15-15-28-48 @ Config de la hoja de excel
FCLK 3666 1:1 Coupled mode with the RAM.

RESULTS:

Aida64: Read 54781, Write 29275, Copy 51734, Latency 65.9ns
CB15: 2206
CB20: 4949
3DMark Time Spy CPU: 10853
Realbench 2.44 hwbot: (Image) 213165, (Encoding) 346093, (OpenCL) 116437, (Heavy MultiTasking) 275377, (System Score) 214480
Shadow of the tomb raider Complete: 166 (General average), 315 (Average cpu)

Perhaps later when they finish debugging the bios and the PBO I'll give it another chance, for now I'll leave it like this since with PBO I get the same frequency in multicore, 41.5x, but it puts 0.1v more than necessary... That is, in monocore it reaches 43.5, but since I don't need it I prefer to have it like this and cooler.

Here I leave the captures of the tests:

CONFIG:
alt text