Ryzen Owners Zen Garden

[Belfaborac]

Strange. 1.07c works like a charm here and was the first to provide proper compatibility and a reasonable boot time.

 

[Mussels]

does running IF faster than 1:1 have any performance benefits

There is no such thing, if it's not 1:1 it's slower

 

[tabascosauz]

Damn, OCCT has evolved since I last tried it.... probably in the Core2 days though. Lowering the temperature (= fan noise) is the main reason why I want this to run with lower voltages.

Does it work that I put everything to auto in bios and then use Ryzen Master, THEN put those working values to bios?

Also a question for you guys, does running IF faster than 1:1 have any performance benefits?

If you aren't at least DDR4-5133+ and running a tight profile, then not running 1:1 is just unequivocally slower.

 

[Mussels]

If you aren't at least DDR4-5133+ and running a tight profile, then not running 1:1 is just unequivocally slower.

They're on AM4, so it's 1:1 or bust


AM5 is far more confusing as its 2:3 or 1:2
2000IF:3000MHz/6000MT/s or the new gear with 2000/4000 (8000MT/s)

people and their love of numbers always misunderstand fractions and ratios


AM5 is more tolerant of being async, i've got a 7900x coming in and already got some DDR5 8000 and a decent board here, so i'll be testing that out soon
2000 IF and 6000 vs 8000, seeing what happens as they go out of sync.

In theory dropping to 7800 should be worse than 6000 for latency, even if timings are the same

 

[AusWolf]

They're on AM4, so it's 1:1 or bust


AM5 is far more confusing as its 2:3 or 1:2
2000IF:3000MHz/6000MT/s or the new gear with 2000/4000 (8000MT/s)

people and their love of numbers always misunderstand fractions and ratios


AM5 is more tolerant of being async, i've got a 7900x coming in and already got some DDR5 8000 and a decent board here, so i'll be testing that out soon
2000 IF and 6000 vs 8000, seeing what happens as they go out of sync.

In theory dropping to 7800 should be worse than 6000 for latency, even if timings are the same

Just to add: the infinity fabric (FCLK) is not in sync with anything on AM5. It is a standalone number. The only clocks running in sync are the memory clock (MCLK) and the memory controller (UCLK). The recommendation is 1:1 up to 6000 MHz RAM (faster UCLK) and 1:2 above it (slower UCLK), but it varies by system.

Therefore, there is no such thing as 1:1:1 or 1:2:1 or any other ratio with three numbers in it. The FCLK is 2000 MHz. That's it.

 

[Mussels]

Just to add: the infinity fabric (FCLK) is not in sync with anything on AM5. It is a standalone number. The only clocks running in sync are the memory clock (MCLK) and the memory controller (UCLK). The recommendation is 1:1 up to 6000 MHz RAM (faster UCLK) and 1:2 above it (slower UCLK), but it varies by system.

Therefore, there is no such thing as 1:1:1 or 1:2:1 or any other ratio with three numbers in it. The FCLK is 2000 MHz. That's it.

I'm aware it matters a lot less - but i've seen some reports of lower latency when they are synced up and i'm going to test it myself in the coming days (Got an AM5 system coming this week, already got some intel parts)
Those dividers do exist, they're just not exposed to the user. It's how memory dividers have worked since SD-RAM, before that i was learning not to shit myself so i'm not sure about EDO and earlier stuff.

From what i've seen so far from the people testing it fully, you get the best results with 2000MHz/6000MT/s and 2000/8000MTS, the in between values arent as good. It's one of those things that you can get higher throughput but worse latency, so 'faster' depends on what you're measuring.

Because early AM5 stuff was chaotic AF with BIOS settings (especially MSI) a lot of early conclusions changed, automatic settings can't be fully relied upon.

I primarily care about gaming results myself, so i'm focused on the latency values. Already got some AM5 boards, just need my 7900x to arrive


Edit: Oh and this is likely tied into the longer boot times, people with lower values like 5600MT/s don't have these issues. Testing is the hard part, as people with the hardware have no motivation or only one one specific value tested (because they're overclocking, not testing a theory)



Late mailman update:
7900x arrived
Waterblock arrived
motherboard arrived


Waterblock is missing the AM5 standoffs


Sadness time.

 

[dont whant to set it"']

Waterblock is missing the AM5 standoffs

I made some out of glue-stick for glue gun, same length and drilled.These are compressible a bit , but maybe I got lucky this time as it worked first time: the whole cooler gets warm.

 

[Mussels]

Can confirm AM5 memory is weird
I can boot 7800 fine, with 7600 being stable.

7600 benches really well, high speeds and low latencies.


6000? No matter what i set for U and Fclk, latency goes to absolute shite (75ns)

This goes against all logic, and what people are saying to do with AM5

Yes theres a RAM speed advantage, but having F and U out of sync shouldnt make latency lower.

After i get sleep, i'll test if having a gap between them is what matters

M
F
U
with say 200Mhz each step of the way?

(Yes theres CL34 to 36, but the same issue occured with the same timings - i tried multiple RAM kits)

 

[TheLostSwede]

Can confirm AM5 memory is weird
I can boot 7800 fine, with 7600 being stable.

7600 benches really well, high speeds and low latencies.


6000? No matter what i set for U and Fclk, latency goes to absolute shite (75ns)

This goes against all logic, and what people are saying to do with AM5

View attachment 317710

View attachment 317714


Yes theres a RAM speed advantage, but having F and U out of sync shouldnt make latency lower.

After i get sleep, i'll test if having a gap between them is what matters

M
F
U
with say 200Mhz each step of the way?

(Yes theres CL34 to 36, but the same issue occured with the same timings - i tried multiple RAM kits)

Your settings are all over the place. Your tRFC is insanely high if you have Hynix chips.
Ok, so my kit is rated at CL30 so your other timings might not be able to run as tight, but your 6000 and 7200 settings obviously differ, with the tRFC being even higher at 6000 which is causing some additional latency.
See below for my timings. The one on the right is default XMP settings, which gave me a latency at over 70 ns and I'm now at around 65 ns.
I was down to 63 ns something at CL 28, but it wasn't stable.

See this list for tRFC timing suggestions.

Ryzen Owners Zen Garden

Yeah, I get it. Overthinking has nothing to do with this, if that was the case then about 90% of comments on computer enthusiasts forums would be classed as that. If AMD claim the 'sweet spot' for Zen 4 is 6000MT/S & yet if the end user does not touch the FCLK setting & leave it on auto, it...

www.techpowerup.com

 

[freeagent]

Gotta watch that trfc, it will literally get you every time

 

[Ruru]

Not gonna upgrade to AM5 in while, but is the E chipset worth it for a normal consumer/gamer?

 

[Belfaborac]

Can confirm AM5 memory is weird

As mentioned already your timings are all over the place, so the high latency is expected. These are my current (and final) timings, which land me at ~59ns (immediately after boot. After a while of doing...anything at all, I'm in the mid-60s and when running AIDA while doing other stuff I'm up in the low 70s).

No doubt they can be tightened even more, but I can't be bothered spending any more time on it. Anyway, this has been 100% stable through stress testing, benches, gaming and idling for a few months, ever since I picked up the set.

Not gonna upgrade to AM5 in while, but is the E chipset worth it for a normal consumer/gamer?

I'd say no. If all you want is a gaming/general purpose rig with a couple of M.2s, there's no point whatsoever in spending extra on an E-type. You won't get any extra performance, stability or whatever else, just more slots and maybe some more tweakability.

 

[Count von Schwalbe]

I'd say no. If all you want is a gaming/general purpose rig with a couple of M.2s, there's no point whatsoever in spending extra on an E-type. You won't get any extra performance, stability or whatever else, just more slots and maybe some more tweakability.

I believe the change is to upgrade the PCIe spec on some slots. So there is nothing really that needs PCIe 5.0 at the moment, but it is a way to possibly continue using the same MB in the future, as new M.2 drives and GPUs come out.

 

[Belfaborac]

Well, you'll find Gen 5 PCIe on non-E boards, both slots and M.2, so that's not really an issue. You just won't get as many. As far as Gen 5 M.2 is concerned I see it as pointless anyway, especially for someone looking to build a PC for general use. Hardly any regular user will benefit from the increased speed and in any case it's not worth the hassle caused by the added heat. Maybe that'll change in the future, but if it was me I would find a board which maximises Gen 4 M.2 slots instead.

 

[TheLostSwede]

Not gonna upgrade to AM5 in while, but is the E chipset worth it for a normal consumer/gamer?

I guess that's up to you. Most non E boards lack PCIe 5.0 for the GPU, not because the chipset lacks support, but because segmentation/cost.
You'll have to figure out how important that is to you.
Assuming AM5 lives as long as AM4, it might matter, but this also depends on what chipsets and CPU features AMD has planned for future generations.

Here's what Wikipedia has to say on the matter.

E or "Extreme" branding is a guarantee that PCIe 5.0 is supported on both the motherboard's graphics slot and NVMe slots. E models provide access to all 24 PCIe 5.0 lanes from the processor. Non-E models only support 8 PCIe 5.0 lanes on NVMe slots while the rest of the lanes on graphics slots are dropped to PCIe 4.0.

List of AMD chipsets - Wikipedia

en.wikipedia.org

 

[Ruru]

I guess that's up to you. Most non E boards lack PCIe 5.0 for the GPU, not because the chipset lacks support, but because segmentation/cost.
You'll have to figure out how important that is to you.
Assuming AM5 lives as long as AM4, it might matter, but this also depends on what chipsets and CPU features AMD has planned for future generations.

Here's what Wikipedia has to say on the matter.

List of AMD chipsets - Wikipedia

en.wikipedia.org

Hell, I run my card at 3.0 x8 so at least PCIe 5.0 isn't in a need for a while.

 

[TheLostSwede]

Hell, I run my card at 3.0 x8 so at least PCIe 5.0 isn't in a need for a while.

Well, we don't know that, but it seems like it won't have a major impact at least.
Not like anything consumery supports it as yet. I guess the MTT S80 is missing from the database, but you can't really buy that outside of xina so...

TechPowerUp

Graphics card and GPU database with specifications for products launched in recent years. Includes clocks, photos, and technical details.

www.techpowerup.com

 

[AusWolf]

Can confirm AM5 memory is weird
I can boot 7800 fine, with 7600 being stable.

7600 benches really well, high speeds and low latencies.


6000? No matter what i set for U and Fclk, latency goes to absolute shite (75ns)

This goes against all logic, and what people are saying to do with AM5

View attachment 317710

View attachment 317714


Yes theres a RAM speed advantage, but having F and U out of sync shouldnt make latency lower.

After i get sleep, i'll test if having a gap between them is what matters

M
F
U
with say 200Mhz each step of the way?

(Yes theres CL34 to 36, but the same issue occured with the same timings - i tried multiple RAM kits)

I guess that confirms it... F is in no sync to anything, and only M and U matter?

Not gonna upgrade to AM5 in while, but is the E chipset worth it for a normal consumer/gamer?

That's up to you, but I'd say no. PCI-e 5.0 is an absolute waste (of money and heat on SSDs), and by the time it's needed, the new platform will probably already be around.

 

[freeagent]

Slapped that Thermalright AIO in my system.. its ok with this 8 core. Kinda curious how it does with my 5900X. I have a feeling it will be about what I am used to.

 

[AusWolf]

Slapped that Thermalright AIO in my system.. its ok with this 8 core. Kinda curious how it does with my 5900X. I have a feeling it will be about what I am used to.

View attachment 317839

It's a lot better than what my AIO did with the 7800X3D and just a little bit better than with the 7700X.

 

[freeagent]

It's a lot better than what my AIO did with the 7800X3D and just a little bit better than with the 7700X.

Not bad for 75 beaver bucks then.. it was pretty fun to get it all mounted.. about 3 hours for the entire operation start to finish lol..

Well maybe not that long.. I tend to do other things like look for youtube videos lol..

 

[AusWolf]

Not bad for 75 beaver bucks then.. it was pretty fun to get it all mounted.. about 3 hours for the entire operation start to finish lol..

75 bucks is a nice price for an AIO - mine was £120, and it's absolutely useless with an X3D. Works pretty well on Intel, though. But 3 hours to mount? Holy moly!

 

[freeagent]

75 bucks is a nice price for an AIO - mine was £120, and it's absolutely useless with an X3D. Works pretty well on Intel, though. But 3 hours to mount? Holy moly!

You have not worked in a Torrent Compact before

And I have a short attention span

 

[AusWolf]

You have not worked in a Torrent Compact before

True... but I did fit a single-fan AIO with a 50 mm thick radiator inside a CM Elite 110 once. That was fun.

 

[TheLostSwede]

You have not worked in a Torrent Compact before

What did you find so hard working on it?
I thought it was pretty straightforward to work with.