What is your highest stable ramspeed/IF on Ryzen 5000?

[tabascosauz]

Idk. Only so much reliability and validity to a forum poll. I mean there are definitely some of us who are actually 4000/2000 stable but it took a lot of work. I’m wondering if we did a poll and asked everyone over 3800 to provide stability testing and screenshoots, if that number would be the same. Self reporting with no evidence is highly unreliable. It would be more interesting to have everyone post screenshots, tally the results and report percentages and speeds afterwards. To draw conclusions about anything would become much more valid. I could give a crap about dick swinging. I’d like to learn more about the science through evidence based reporting. So, how about any of us with 4000/2000 do a screenie of an hour of OCCT memory test with cpu z, zen timing and Thaiphoon?

Problem with that is that memory tests (HCI, TM5, Karhu, P95 Large) only test memory. On Ryzen there are three domains that need to be stable - the memory, the UMC, and the IF. You can kind of use P95 to test the UMC, but there is basically no proper test for 100% IF stability, aside from doing as diverse a workload as possible day after day. So while you can quite easily verify if your B-die itself is stable at any profile, all it takes is one unexpected reboot/failure to POST/Bus Interconnect WHEA to unravel days/weeks/months of supposed "stability" at 2000MHz IF. Knowing how IF works, it'll be a long time before I can take any owners' word for it that they can "stably" do 4000 1:1. Helps to be playing games that are simultaneously memory, CPU and GPU intensive, but gaming is far from consistent enough to be considered a "test".

 

[freeagent]

I am just running a paltry 1600/1600

Not missing out on much

What's the best bus for the best latency with Ryzen 5000?

3800 CL14 > Latency around 57 ns
4000 CL16 > Latency around 55 ns

Why 3800CL14 get more latency than 4000 CL16???

At 3200 cl 14 I am down to 58ns

 

[thesmokingman]

I am just running a paltry 1600/1600

Not missing out on much


At 3200 cl 14 I am down to 58ns

For all intents and purposes, that is all anyone needs but it seems ppl like big numbers thus so many have issues chasing that goal. A gskill 3200c14 b-die kit is not that expensive which begs the question why bother? It's kind of hilarious, when you do a 3200mhz c14 search on newegg, no other brands come up save for Team which uses b-die.

G.SKILL Flare X Series 16GB (2 x 8GB) 288-Pin DDR4 SDRAM DDR4 3200 (PC4 25600) AMD X370 / B350 Memory (Desktop Memory) Model F4-3200C14D-16GFX - Newegg.com

Buy G.SKILL Flare X Series 16GB (2 x 8GB) 288-Pin DDR4 SDRAM DDR4 3200 (PC4 25600) AMD X370 / B350 Memory (Desktop Memory) Model F4-3200C14D-16GFX with fast shipping and top-rated customer service. Once you know, you Newegg!

www.newegg.com

 

[freeagent]

For all intents and purposes, that is all anyone needs but it seems ppl like big numbers thus so many have issues chasing that goal. A gskill 3200c14 b-die kit is not that expensive which begs the question why bother? It's kind of hilarious, when you do a 3200mhz c14 search on newegg, no other brands come up save for Team which uses b-die.

G.SKILL Flare X Series 16GB (2 x 8GB) 288-Pin DDR4 SDRAM DDR4 3200 (PC4 25600) AMD X370 / B350 Memory (Desktop Memory) Model F4-3200C14D-16GFX - Newegg.com

Buy G.SKILL Flare X Series 16GB (2 x 8GB) 288-Pin DDR4 SDRAM DDR4 3200 (PC4 25600) AMD X370 / B350 Memory (Desktop Memory) Model F4-3200C14D-16GFX with fast shipping and top-rated customer service. Once you know, you Newegg!

www.newegg.com

Well.. I cant speak for the CPU clocking side of things, but for the memory side it is kind of fun to see the gains. Not that there is a lot to be gained really.. The IF can only move so much, so fast through the memory bus. One thing I did notice is the L3 cache performance, especially it's latency is improved with tight cl14 timings vs fast (2k+) cl16. I am not sure if it is internal straps for certain if/mclk ranges or what exactly is going on yet. Sometimes it would pay to be an engineer

Edit:

There is a strap system. I just looked in the aida thread at my 4200 14-14-14 shot and l3 latency is at 11.5ns. So it slows down as memory and/or IF speed up.

Edit again..

Not sure if it means anything tangible yet.. probably somewhere.

 

[Makaveli]

Most will need to retry this with the new bios out.

AMD AM4 AGESA V2 PI 1.2.0.1 I'm see more reports of better clocks at higher IF.

The L3 Aida fix is in also.

On Bios 3001

On Bios 3602

 

[Taraquin]

Makaveli: Looks good. Tried running higher IF to see where you get? Maybe this improved aswell in new agesa?

 

[PaulieG]

For all intents and purposes, that is all anyone needs but it seems ppl like big numbers thus so many have issues chasing that goal. A gskill 3200c14 b-die kit is not that expensive which begs the question why bother? It's kind of hilarious, when you do a 3200mhz c14 search on newegg, no other brands come up save for Team which uses b-die.

Exactly. Chasing the numbers is fun. For some us who don't game, chasing numbers is the game. It's the same reason I did powerlifting for so long. Tearing my rotator cuff and bicep tendon to get that bench PR? Hell yes.

 

[Taraquin]

With agesa 1.2.0.1 4066/2033 was easy, but I stick to 4000/2000 since voltage reqs are a bit lower and trash binned B-die dislikes the volts it needs above 4000.

 

[529th]

Can't run 1900 without WHEA errors, so 1866 it is, however I'm doing GDM disabled CMD 1T 1700/3400

 

[Taraquin]

Can't run 1900 without WHEA errors, so 1866 it is, however I'm doing GDM disabled CMD 1T 1700/3400

What CPY and agesa do you have? It seems like 5600X/5800X have much better chance at running abpve 3800 than 5900X/5950X, more complexity and 'longer' signals due to 1 extra chiplet might be the reason.

 

[529th]

What CPY and agesa do you have? It seems like 5600X/5800X have much better chance at running abpve 3800 than 5900X/5950X, more complexity and 'longer' signals due to 1 extra chiplet might be the reason.

I have a 5800X

 

[PaulieG]

For those who said they were able to do 4066/2033, can you post screen shots after a successful OCCT memory and a realbench run? I've tried like hell to get anything over 4000/2000 stable using both B-die sticks and high binned Micron Rev. B and I can get them both to do daily tasks at 4066/2033 but I cannot get them truly stable.

 

[woam]

For those who said they were able to do 4066/2033, can you post screen shots after a successful OCCT memory and a realbench run? I've tried like hell to get anything over 4000/2000 stable using both B-die sticks and high binned Micron Rev. B and I can get them both to do daily tasks at 4066/2033 but I cannot get them truly stable.

Hi, I signed up just to show you my configuration at 4133 / 2066 stable, with 5600x and 2 x 16 GB micron rev b 3600c16 @ 4133 c16, @1.45v, vsoc 1.15. 1H on occt, and 11 H on tm5 extreme profile. 0 WHEA error. All is in the screenshot

 

[GerKNG]

5800X 1900
5900X 1800
2nd 5900X 1800
5950X 2000

 

[dgianstefani]

The B die sensitivity to heat really irritates me as there is no reliable way to keep them under 40c under all loads when pushing 1.5v+.
I've looked for waterblocks for two stick kits but they don't exist, and even 2 x NF A4x10 fans blowing directly onto the ram from the side panel (cool air), they'll still hit 45c+ under prolonged stress test.

Getting very tempted to disassemble the stock Trident Z Neo heatsinks which apparently is a pain, and put Gelid GC Ultimate 13w/mk pads instead of whatever factory crap is on there, I suspect that would reduce temps by a few degrees.

It's not really an issue when gaming or doing other intensive tasks because it's never the strain of 100% synthetic load, and temps are usually in the high 30s.

 

[GerKNG]

The B die sensitivity to heat really irritates me as there is no reliable way to keep them under 40c under all loads when pushing 1.5v+.
I've looked for waterblocks for two stick kits but they don't exist, and even 2 x NF A4x10 fans blowing directly onto the ram from the side panel (cool air), they'll still hit 45c+ under prolonged stress test.

Getting very tempted to disassemble the stock Trident Z Neo heatsinks which apparently is a pain, and put Gelid GC Ultimate 13w/mk pads instead of whatever factory crap is on there, I suspect that would reduce temps by a few degrees.

It's not really an issue when gaming or doing other intensive tasks because it's never the strain of 100% synthetic load, and temps are usually in the high 30s.

the Trident z Heatspreader (wouldn't even call that a heatsink) is awful.

i have 16GB Dimms and they run (now in the summer) at up to 50°C at 3600 at 1.35V and they are in the airflow of 140mm noctua industrials and a NH D15 (all at over 1400RPM)

 

[freeagent]

Stable? 2K on my 5600X and 1900 on my 5900X. Unstable? 2100 on 5600X and 2K on 5900X. My B-Die has never seen 40c.. came close though

 

[PaulieG]

Hi, I signed up just to show you my configuration at 4133 / 2066 stable, with 5600x and 2 x 16 GB micron rev b 3600c16 @ 4133 c16, @1.45v, vsoc 1.15. 1H on occt, and 11 H on tm5 extreme profile. 0 WHEA error. All is in the screenshot

Hell, this is going to drag me back into this mess.

 

[Makaveli]

Stable? 2K on my 5600X and 1900 on my 5900X. Unstable? 2100 on 5600X and 2K on 5900X. My B-Die has never seen 40c.. came close though

These are my average temps for my b-die as well @ 1.35v

 

[freeagent]

These are my average temps for my b-die as well @ 1.35v

View attachment 203982

I’m running OCCT right now to cycle my TIM, not the best of examples.. but this is generally where they run at, sometimes a little higher or lower.. it’s about 20c in my computer room..

 

[Makaveli]

I’m running OCCT right now to cycle my TIM, not the best of examples.. but this is generally where they run at, sometimes a little higher or lower.. it’s about 20c in my computer room..

View attachment 203984

Looks good ambient temp in my room is about 24c right now.

 

[msroadkill612]

For all intents and purposes, that is all anyone needs but it seems ppl like big numbers thus so many have issues chasing that goal. A gskill 3200c14 b-die kit is not that expensive which begs the question why bother? It's kind of hilarious, when you do a 3200mhz c14 search on newegg, no other brands come up save for Team which uses b-die.

G.SKILL Flare X Series 16GB (2 x 8GB) 288-Pin DDR4 SDRAM DDR4 3200 (PC4 25600) AMD X370 / B350 Memory (Desktop Memory) Model F4-3200C14D-16GFX - Newegg.com

Buy G.SKILL Flare X Series 16GB (2 x 8GB) 288-Pin DDR4 SDRAM DDR4 3200 (PC4 25600) AMD X370 / B350 Memory (Desktop Memory) Model F4-3200C14D-16GFX with fast shipping and top-rated customer service. Once you know, you Newegg!

www.newegg.com

Its not just ram that uses the Fabric bus.

 

[TheHoof]

It looks like 1933 for sure, and possibly 1966. I can go up to 2166, run out of ram at 2133 but can run every single stress test, benchmark whatever and pass.. no errors logged by windows but occt got me @ 2100. I have it at 1900/1900 with four sticks right now and its pretty tight.

Hard necrobump, do you have a screen of 2166?

 

[freeagent]

Hard necrobump, do you have a screen of 2166?

Negatory, I think I meant 2133, if I did hit 2166 it was with a beta bios. For memory speed I can do it by breaking the IF link.

I might be able to do 2166 1:1 with my current board, as it is a little better than the one I was using back then.

That little 5600X is a ripper, but it’s only running at 1600 1:1 right now

 

[INSTG8R]

Good to see this one again because with the latest AGESA I can now OC my RAM to 3800 on my 5800X3D. Does it show any improvements anywhere? Not really but it’s “because I can” thing.