Ryzen Owners Zen Garden

[Psychoholic]

What kind of improvements will be seen if I were to go from a DDR5-6000 CL36 to a DDR5-6000 CL32 kit?

Very minor, if any at all, in my opinion. Zen 4 doesn't seem to be as sensitive to memory speed and latency as previous generations.

Agreed, Probably would not notice.
Also, Subtimings can be just as important as CL.
A CL32 Kit with Crappy subtimings might be the same or slower than a good CL36 kit with tighter subtimings.

 

[trparky]

There is another video here.
Like ZEN 3 tuning RAM can increase performance on ZEN 4.

Yeah, I just saw that. I'm watching it now.

I looked up on PCPartsPicker and it says that the memory kit that I have has a 12 ns time to the first word. Whatever the hell that means.

From what I understand, first word latency counts. Anything above 14 ns is shit, 10 ns is best; but you're going to pay a premium for that or you're going to have to endure a lot of testing to fine tune your kit.

 

[tabascosauz]

Yeah, I just saw that. I'm watching it now.

I looked up on PCPartsPicker and it says that the memory kit that I have has a 12 ns time to the first word. Whatever the hell that means.

From what I understand, first word latency counts. Anything above 14 ns is shit, 10 ns is best; but you're going to pay a premium for that or you're going to have to endure a lot of testing to fine tune your kit.

The Microcenter kit is bundled because it makes perfect sense for Zen 4 - where you're always UCLK (CPU) limited. On a platform where you're not strictly limited to 6000-6400 (ie. not AM5), it probably has garbage OC potential as not-Hynix and no one who's into memory would want it.

Kinda like why paying out the ass for one of those 5333 DJR kits makes no sense if you have a non-APU AM4 CPU; all the OC headroom is wasted on the hardware.

You can always have some healthy gains if you're willing to work on your timings as opposed to XMP. The test intentionally magnifies the obvious CPU bottleneck so that it can show memory performance, with a 4090 @ 1080p which is not something you actually want to game at.

 

[trparky]

4090 @ 1080p which is not something you actually want to game at.

Why do so many test at that? It's nowhere near real world setups.

 

[tabascosauz]

Why do so many test at that? It's nowhere near real world setups.

Because the point here is to illustrate differences in memory performance. You tailor the test to what you want to convey.

If you review CPUs, you test at lower resolutions to let CPUs stretch their legs. If you review high end GPUs, you do higher resolutions to let GPUs do their thing.

 

[trparky]

The Microcenter kit is bundled because it makes perfect sense for Zen 4 - where you're always UCLK (CPU) limited. On a platform where you're not strictly limited to 6000-6400 (ie. not AM5), it probably has garbage OC potential as not-Hynix and no one who's into memory would want it.

Yeah, it seems like Ryzen 7000 doesn't like anything above DDR5-6400. Even with the latest AGESA code, as Linus stated in one of his videos, getting memory faster than DDR5-6400 to even boot is nearly impossible due to, as he puts it, a weak memory controller.

 

[The King]

Yeah, it seems like Ryzen 7000 doesn't like anything above DDR5-6400. Even with the latest AGESA code, as Linus stated in one of his videos, getting memory faster than DDR5-6400 to even boot is nearly impossible due to, as he puts it, a weak memory controller.

Same was the case with ZEN 1. Many struggled to even run 3200MT in some cases but my 1700X happily did FCLK 1800 / 1833 stable.

I'm sure I could get above 6400 to work on ZEN 4.

 

[tabascosauz]

Same was the case with ZEN 1. Many struggled to even run 3200MT in some cases but with my 1700X happily did FCLK 1800 / 1833 stable.

I'm sure I could get above 6400 to work on ZEN 4.

6400 is not the sweetspot, 6000 is. I've only seen a select few lucky enough to hit 3200MHz.

I'm sure you could, but same story as AM4, if you break 1:1 UCLK to do so you might as well just not do something pointless. Gear 2 is a fact of life for Intel DDR5, but at least it just keeps on scaling like that into the 8000MT/s region.

Whether it's a 1:1 problem or the UMC literally stops there, who knows. Everyone scurried off to 13th gen for DDR5, I still haven't found any good non-anecdotal testing beyond those speeds for Raphael.

 

[The King]

6400 is not the sweetspot, 6000 is. I've only seen a select few lucky enough to hit 3200MHz.

I'm sure you could, but same story as AM4, if you break 1:1 UCLK to do so you might as well just not do something pointless. Gear 2 is a fact of life for Intel DDR5, but at least it just keeps on scaling like that into the 8000MT/s region.

Whether it's a 1:1 problem or the UMC literally stops there, who knows. Everyone scurried off to 13th gen for DDR5, I still haven't found any good non-anecdotal testing beyond those speeds for Raphael.

In the Hardware unboxed Video posted above about ZEN 4 with tuned RAM, I was impressed with the 5600 CL28 DR tuned setup. I think if you can get that on ZEN 4 then you would not need more more than that. Seems close to the sweetspot. The 5600 CL28 SR tuned setup was also very good overall.

 

[AusWolf]

6400 is not the sweetspot, 6000 is. I've only seen a select few lucky enough to hit 3200MHz.

I'm sure you could, but same story as AM4, if you break 1:1 UCLK to do so you might as well just not do something pointless. Gear 2 is a fact of life for Intel DDR5, but at least it just keeps on scaling like that into the 8000MT/s region.

Whether it's a 1:1 problem or the UMC literally stops there, who knows. Everyone scurried off to 13th gen for DDR5, I still haven't found any good non-anecdotal testing beyond those speeds for Raphael.

Word is, dropping UCLK to 1:2 has more negative impact on system performance than an increased memory clock has positive, that's why 6000 is the sweet spot - that's the max RAM speed most CPUs can keep UCLK at 1:1 with. Without first-hand experience, I can't verify it, though.

 

[Mussels]

Not just useless, but potentially counter-productive when you're impacted by the increased latency between the two CCDs.

What i've been trying to point out in the intel threads/13900K review is the current single CCD core counts are the primary bottleneck - we can see this with the older ryzens with a 3+3 setup where games that need/use 4 cores are suffering pretty badly - and it scales up

Right now the most cores we can get in a single CPU is 8 from both camps, but we can still buy 4 and 6 core CPU's (with secondaries, be they identical or E-cores)
We're seeing this in things like the 7700x vs the 7900x, because it's a 6 core CPU - as soon as it needs more threads, you get the latency penalty and wattage limits lowering clocks




Got my new Aorus B550-I PRO AX 1.2 in, bought some new RAM and threw the 3700x in there for now
The rev 1.2 changes to wifi 6E with an AMD chip, weirdly it cant see one of my 5GHz networks but sees another one just fine (Was actually router issue)


So far, i really like the board. Great design, BIOS is great.
CMOS clears are really slow - seems to require PSU plugged in and switched on to work? Never seen that before.

Ram is 2x16GB single rank - i wanted dual rank, but for the future that's probably best anyway.
3600 C18 that rans 3800 C18 and 3200C14, but the CPU has WHEA errors above 3333 unless SoC is at 1.2v so 3200 C14 and tuning it is

Currently using a 100W PBO limit with undervolting, seeing 4.1GHz in R23 at 79C, and 4.6GHz max single thread boost - unsure if it will ever REACH it, but HWinfo reports that as the maximum

Ram doesnt want to run 14-16-16, but runs 14-18-18 happily
I *love* this BIOS feature

Edit: Those bastards, i thought they UPGRADED the wifi, but nooo

Apparently the intel bluetooth had issues, so they moved to the AMD/mediatek wifi to fix that, but halved the wifi speed to do so

XMP asks for more power from the RAM and the CPU's memory controller, but only marginally. You have nothing to lose by enabling it.

Even with extreme voltage increases for memory overclocking here, my SoC voltages have only gone up around 10W from stock - even on a 65W CPU (88W in reality) that performance difference is incredibly small since you get diminishing returns as the clocks raise up (10W could be 10% more power, but 2% performance)

 

[tabascosauz]

So far, i really like the board. Great design, BIOS is great.
CMOS clears are really slow - seems to require PSU plugged in and switched on to work? Never seen that before.

Not sure how clear CMOS can be slow. You mean the first train after a CMOS clear is slow?

The Aorus AX has a quirk where sometimes actually clearing CMOS requires unplugging/turning off PSU and draining residual power. Probably Gigabyte's safeboot functionality getting in the way. Or else it'll appear to have set your settings but is actually running JEDEC, and will continue to run JEDEC until power is drained.

 

[Mussels]

Not sure how clear CMOS can be slow. You mean the first train after a CMOS clear is slow?

You have to short it for 10+ seconds or it just doesnt clear
And yeah it's weird, you either need the PSU connected and the switch on, or have it disconnected for a long time to drain any caps

Never seen another board behave that way, had to google it and find other peoples ways of working with it


Oh and yeah failed boots sometimes load after a few attempts at DDR4 2400, which i can imagine throws off a few beginners relying on XMP
This build i have plans for a 5700g in the future

 

[tabascosauz]

You have to short it for 10+ seconds or it just doesnt clear
And yeah it's weird, you either need the PSU connected and the switch on, or have it disconnected for a long time to drain any caps

Never seen another board behave that way, had to google it and find other peoples ways of working with it

Why wait forever, just unplug and press the power button and wait 2 seconds. See if that does the trick after clearing CMOS.

I had a habit of hooking up a spare vandal power button to the clear CMOS header, saved a ton of time.

Expect one reboot to properly apply some RAM profiles - any more reboots on end and the timings are no bueno

 

[Mussels]

Why wait forever, just unplug and press the power button and wait 2 seconds. See if that does the trick after clearing CMOS.

I had a habit of hooking up a spare vandal power button to the clear CMOS header, saved a ton of time.

Expect one reboot to properly apply some RAM profiles - any more reboots on end and the timings are no bueno

It doesnt clear - it's a weird board hiccup, i've never seen anything like it. This board just takes a long time to clear if the PSU doesnt have a ground for some reason


Logged into TPU on test rig for screenies

4.25GHz single core max so far, with 4.1GHz all core

Global frequency limit is whats up high and i'm waiting to see if it ever actually uses it, or if the +200 tis but a dream with these limits

*pokes with stick*

bad cores. go higher.

It's not like effective clocks are tanking or anything, it all matches up

just doesnt want to ST boost very far (wrong amperage PBO defaults, perhaps?)

Why do i feel like manually setting this for an all core OC wont run anywhere near as well

Good old clock stretching - seems effective clocks doesnt see it on this setup

4.2Ghz all core, meanwhile

 

[tabascosauz]

I looked at the BIOS list, not very promising if AGESA 1207 and 1208 are your only two choices......4.1 all core seems okay slightly low for 3700X but 13k+ score looks normal. Mine was 13k and 1310 basically stock

Weren't you running 4.4 on this CPU? If you are above 1300 ST then things look fine to me...PBO didn't really scale too well on 3700X, you can give it as much power as you want but it doesn't mean it'll change anything. It's not Zen 3. Or just run all core oc.

 

[trparky]

What i've been trying to point out in the intel threads/13900K review is the current single CCD core counts are the primary bottleneck

Wait. What? What do you mean by that?

 

[dont whant to set it"']

Exploring BCLK, where there are holes. 107 has been stable for months with either 4 single rank dimms(current setup) or 4 dual rank dimms(maxing only to around 3200MT) SMT off to avoid thermal throttle.

Score could be higher by a bit as it was a benchmark run with HWiNFO64 minimized.

 

[johnspack]

Well I need some advice on how to keep this monster below 68c under heavy load... Who in here has experience with 1st gen tr, or even ryzen? I added a 2nd 140mm
exhaust, replaced a failing 120mm intake, there are 3. Removed the drive bays as this is a ssd only system anyways. Installed the u14 cooler with 2 140mm fans.
Have applied mx4 tim twice now, using 2 different methods, one that I saw online, and one that came with the instructions for my cooler. Under heavy load I'm still
hitting 70c! What am I doing wrong?

 

[trparky]

Under heavy load I'm still hitting 70c! What am I doing wrong?

Then you’re damned lucky. Under a full load, like if I’m compiling C++ code on all eight “cores” or running Cinebench test and they’re all pegged at 100% usage, my temps hit 95c.

It’s just the nature of these chips. They’re designed to safely hit those temps.

 

[johnspack]

I pulled the cooler and redid the tim.... now at 65c under heavy load. I'll take that! And this chips tdp is 68c.... I don't need it throttling!

 

[tabascosauz]

I pulled the cooler and redid the tim.... now at 65c under heavy load. I'll take that! And this chips tdp is 68c.... I don't need it throttling!

That's not what the 68C spec is referring to, your temps are just fine

Only starting with Ryzen 3000 did AMD do away entirely with the offsets and just let Tctl=Tdie

AMD Ryzen Threadripper 1920X review

Next to the sixteen core 1950X AMD also has released its twelve core Ryzen Threadripper 1920X processor which we review in this article. So what's it gonna be for you, what does make more sense: 12 o... Power Consumption & Temperatures

www.guru3d.com

 

[Mussels]

I looked at the BIOS list, not very promising if AGESA 1207 and 1208 are your only two choices......4.1 all core seems okay slightly low for 3700X but 13k+ score looks normal. Mine was 13k and 1310 basically stock

Weren't you running 4.4 on this CPU? If you are above 1300 ST then things look fine to me...PBO didn't really scale too well on 3700X, you can give it as much power as you want but it doesn't mean it'll change anything. It's not Zen 3. Or just run all core oc.

4.2 all core is happy at 90W, same as stock wattages at around 1350 in R23
When it was at 4.4 it was on a different board and different cooler - it's moved from ATX x370 to ITX B550

Ram is at 3200C14 which is helpful as well, I've learned that this CPU just cant do above 3333 without WHEA errors regardless of the RAM - I've always had ram stability issues with it, but prior to a stick dying it had a 2x8 2667 corsair kit so I assumed the RAM itself was the problem

I also suspect there was clock stretching or lying BS from that board, as this board clearly is too with PBO involved

That's not what the 68C spec is referring to, your temps are just fine

Hwinfo gives two temp readings for those CPU's, one 20C higher than the other - makes it clear whats the real temp

 

[JrRacinFan]

Updated to the latest AGESA bios earlier today (7000x3d microcode update), setting a manual pbo configuration in bios still causes my chip to be locked down to 5.5x cpu multiplier. I even tried in advanced CPU settings. ALTHOUGH i found where eco mode is in my bios under "SMU common options".

 

[Mussels]

compared to the <60ns i managed on the 5800x, this is a tad slower
buuuut its getting there for zen2