Latest AMD AGESA that Nerfs Ryzen 7000X3D Voltage Control Also Limits Memory Overclocking

[somebodys_kid]

Do you mean 1410? 1401 is still 1.0.0.6 afaik. Sounds like your memory overclock itself is unstable if you’re tuning timings. It may not be the case with all CPUs but 1.2-1.25 VSOC should typically be enough for 6000-6200.

ASUS’s X670 boards do seem to be a mess though.

Oh good call. I'll try the 1410 later today. 1401 just capped the voltage without the AGESA update.
The spec listed by GSKILL wants 1.35 volts for the EXPO settings.

 

[Metroid]

I said this would happen and some people here said I was wrong. Now you trolls who said I was wrong can see it clear as a bright sky.

 

[rv8000]

Oh good call. I'll try the 1410 later today. 1401 just capped the voltage without the AGESA update.
The spec listed by GSKILL wants 1.35 volts for the EXPO settings.

G.skill lists VDD, not VSOC. 1.35v VDD (essentially vdimm/vmem). The naming scheme for voltage rails has gotten out of hand with newer platforms.

1.35 VDD for your memory is absolutely fine and 100% safe.

 

[unwind-protect]

You predicted 24 GB and 48 GB memory configs would be clock limited after this update? Wow that’s a pretty specific prediction. Kudos to you! I would have never predicted that this specific memory configuration would be only effected.

You misunderstood. I predicted that when the overvoltage issue gets fixed that they throw out the baby with the bathwater and disable a random bunch of legitimate overclocking options.

 

[phanbuey]

To be fair it didn't scale past 6400 anyways, so it's not like you're losing a whole lot here?

Not sure what SOC needs to hit 6000 cl30 but that seems to be the sweet spot.

 

[Slizzo]

To be fair it didn't scale past 6400 anyways, so it's not like you're losing a whole lot here?

Not sure what SOC needs to hit 6000 cl30 but that seems to be the sweet spot.

I'm running 1.2v SoC currently at 6000MT/s

 

[phanbuey]

I'm running 1.2v SoC currently at 6000MT/s

In my case on the Intel platform the difference between 6000 cl30 to 6200 cl32 to 6400 cl34 was basically nothing... doesn't seem like nerfing that last 400mhz will do anything -- especially since x3d chips already have that cache cushion.

 

[rv8000]

To be fair it didn't scale past 6400 anyways, so it's not like you're losing a whole lot here?

Not sure what SOC needs to hit 6000 cl30 but that seems to be the sweet spot.

There’s plenty of “scaling” to be had, it’s just very rare for a chip to be able to do 6600, or maintain mclk/uclk/fclk ratios to not tank performance.

I’ve seen a few posts with 7000-7200 if I recall, but then you can’t maintain mclk:uclk @ 1:1, so performance takes a nose dive.

With 6600 c28-c30 you can shave another 2-3ns off latency and get some additional bandwidth over 6400. Requires a lot of VDD(vdimm) and good 7000 chip.

 

[phanbuey]

There’s plenty of “scaling” to be had, it’s just very rare for a chip to be able to do 6600, or maintain mclk/uclk/fclk ratios to not tank performance.

I’ve seen a few posts with 7000-7200 if I recall, but then you can’t maintain mclk:uclk @ 1:1, so performance takes a nose dive.

With 6600 c28-c30 you can shave another 2-3ns off latency and get some additional bandwidth over 6400. Requires a lot of VDD(vdimm) and good 7000 chip.

This is the max difference I could find without tuned subs between 6200CL34 and higher ram speeds (13900K with 4090) -- the jump of 1000mhz on the ram netted 3fps at the lows and 7.6% at the averages - so with less headroom than this... scaling is what - 5% max? I wouldn't say that's plenty -- that's basically you couldn't tell the difference unless you were staring at a framerate graph.

If you're saying "Im nerfing memory from 8000 to 6000" then yes- that's a huge nerf... outliers that can do 6600 aside (and even then) at most this is a 5% difference given the same timings? Probably less once you tighten timings on the faster kits as low as they will go.

 

[b1k3rdude]

"It is possible to max out 192 GB, but while the older AGESA 1.0.0.6 allowed memory frequencies of up to DDR5-6000 with SoC voltage of 1.3 V, the newer AGESA is only stable up to DDR5-4400 at this densit"

As Long as I can install 32GB of DDR5-6000 CL30 or faster, then like a lot of users, I would never run into the issue above. But its seems the memory compatibility pages for mobo makers still need updating -

MAG X670E TOMAHAWK WIFI

Powered by AMD Ryzen 7000 series processors, the MSI MAG X670E TOMAHAWK WIFI is hardened with performance essential specifications to outlast enemies. Tuned for better performance by Core boost, memory Boost, Premium Thermal Solution, M.2 Shield Frozr.

www.msi.com

 

[rv8000]

View attachment 295011

This is the max difference I could find without tuned subs between 6200CL34 and higher ram speeds (13900K with 4090) -- the jump of 1000mhz on the ram netted 3fps at the lows and 7.6% at the averages - so with less headroom than this... scaling is what - 5% max? I wouldn't say that's plenty -- that's basically you couldn't tell the difference unless you were staring at a framerate graph.

If you're saying "Im nerfing memory from 8000 to 6000" then yes- that's a huge nerf... outliers that can do 6600 aside (and even then) at most this is a 5% difference given the same timings? Probably less once you tighten timings on the faster kits as low as they will go.

Like I mentioned earlier, reviews that show memory scaling are pretty pointless. They don’t spend with secondary and tertiary timings, so auto rules start loosening those timings and simply upping the ddr5 frequency and controlling primaries isn’t really “tuning”.

It will vary from game to game, but from personal testing going from c36 6000 expo to 6200 c30 tightening primary/secondary/tertiary was about a 6% gain in CP2077 at 1080p low preset. I can probably test 6400 now that I have a good kit of A-die.

 

[phanbuey]

Like I mentioned earlier, reviews that show memory scaling are pretty pointless. They don’t spend with secondary and tertiary timings, so auto rules start loosening those timings and simply upping the ddr5 frequency and controlling primaries isn’t really “tuning”.

It will vary from game to game, but from personal testing going from c36 6000 expo to 6200 c30 tightening primary/secondary/tertiary was about a 6% gain in CP2077 at 1080p low preset. I can probably test 6400 now that I have a good kit of A-die.

That's right -- I think my argument was more that the SOC limit is really at the frequency, not as much for the timings -- you're absolutely spot on about the subs loosening at the higher speeds which makes those comparisons pretty misleading.

That being said -- most people don't really tune subs, and to them 6000 to 6600 or 6400 EXPO is going to be very little difference for the reason above. For the people that DO tune subs the focus on OCing will fall to the timings -- and a C26/C28 6200 with aggressive primary/secondary/tertiary will still net a huge performance boost, -- maybe you will lose 5% not being able to go to 6400/6600 if that's the sweet spot of the kit... but the difference isn't going to massive in either case.

 

[rv8000]

That's right -- I think my argument was more that the SOC limit is really at the frequency, not as much for the timings -- you're absolutely spot on about the subs loosening at the higher speeds which makes those comparisons pretty misleading.

That being said -- most people don't really tune subs, and to them 6000 to 6600 or 6400 EXPO is going to be very little difference for the reason above. For the people that DO tune subs the focus on OCing will fall to the timings -- and a C26/C28 6200 with aggressive primary/secondary/tertiary will still net a huge performance boost, -- maybe you will lose 5% not being able to go to 6400/6600 if that's the sweet spot of the kit... but the difference isn't going to massive in either case.

This is the other half of the story, and sadly reviewers only show this.

They could have a c32 6000 kit and a c32 6400 kit with a 1% difference or less when set to expo. The 6000 kit could have very tight sub timings like trfc, terfi, for some hefty gains. The 6400 kit could have the most abysmal trfc and trefi at almost double/half the other kit. This will vary from kit to kit, brand to brand, and even worse mobo vendor to mobo vendor as most of them will define their own timing rules for secondary/tertiary as dram vendors don’t always seem to define those values in xmp/expo profiles (or autocorrection takes over).

So technically to say there’s no scaling just is not true; ticking expo and dialing up the frequency while having no idea what the motherboard is setting sure.

Id say overclocking has changed, and requires a lot more finesse and understanding than previous brute forcing methods in the past. And again, sadly, reviewers don’t have the time or luxury to spend actually testing or showing this.

 

[Airisom]

I haven't updated, but I'm running 1.25v SOC with 2x32GB 6000CL30 with tweaked timings on PBO 7800X3D with 104mhz eclk. Stable as a rock on X670E Hero

 

[evernessince]

You misunderstood. I predicted that when the overvoltage issue gets fixed that they throw out the baby with the bathwater and disable a random bunch of legitimate overclocking options.

It's a little early to say which changes will stick. The recent BIOS updates were put together in a day or two and were not intended to be the long term fix. They are only to prevent any further burnt CPUs / motherboards while AMD fully investigates the issue.

 

[somebodys_kid]

G.skill lists VDD, not VSOC. 1.35v VDD (essentially vdimm/vmem). The naming scheme for voltage rails has gotten out of hand with newer platforms.

1.35 VDD for your memory is absolutely fine and 100% safe.

Tried again with the 1410 BIOS and still crashes after 20 minutes with RE4 remake. Will try again with the next BIOS until they get it right I suppose.

 

[ir_cow]

Looks like the DDR5-4400 is just for 4x Dual-rank (128/196). I am able still able to run 2x single-rank 6200.

Every IMC will be different but this is what I'm needing for stability. Basically at least on this GB Elite AX motherboard you need to manually set the VDDIO_Mem because it just stays at 1.1V otherwise. I assume the DDR5-6000 is still a thing, but some for some that may be 5600 now. Dependent on the IMC.

Yes, Gigabyte still allows over 1.3V, you just get a warning now is all...


�

Memory Support BIOS F8d
Speed	Ranks / DIMMs	Boots	Stable(?)	Auto Voltages	MCLK Gear	FCLK	SoC	VDDIO_Mem
DDR5-6000	Single-Rank / 2x	Yes	Yes	Yes	1:1	2000 MHz	1.25 V	1.10 V
DDR5-6200	Single-Rank / 2x	Yes	Yes	No	1:1	2000 MHz	1.30 V	1.20 V
DDR5-6400	Single-Rank / 2x	Yes	No	No	1:1	2000 MHz	1.40 V	1.25 V
DDR5-6600	Single-Rank / 2x	No	N/A	N/A	1:1	1500 MHz	1.40 V	1.30 V
DDR5-6000	Single-Rank / 4x	Yes	Yes	No	1:1	2000 MHz	1.30 V	1.20 V
DDR5-6200	Single-Rank / 4x	Yes	No	No	1:1	2000 MHz	1.35 V	1.30 V
DDR5-6000	Dual-Rank / 2x	Yes	Yes	No	1:1	2000 MHz	1.25 V	1.20 V

 

[MarsM4N]

I said this would happen and some people here said I was wrong. Now you trolls who said I was wrong can see it clear as a bright sky.

I mean that's just some basic common sense. If AMD is setting new maximum limits & RAM vendors selling RAM kits that run outside of spec, there will be incompatibility.
Just looked at the listenings over at heise.de. 170 kits listed with AMD EXPO, and their voltage specs:

If you narrow it down to kits that are specified for 6000MHz (AM5 sweet spot) there are left 65 kits. Their specified voltages are:

Which means you have only 5 kits that run within AMD's specs at 6000MHz (from TeamGroup, Apacer & Lexar). 60 kits are certified by RAM vendors to run outside of AMD's specs! Which btw. does not mean they won't run with lower voltages. It's just not guaranteed to run at lower voltages. Each kit is of different qualitiy and RAM vendors just set higher voltages to guarantee compatibility across the board.

I guess the solution would be to use/develop RAM kits that run with lower voltages. It kinda reminds me of the DDR3 times when all vendors had kits running on 1.5+V and G.Skill came around the corner with their "G.Skill ECO Series" which did run on only 1.35V. It also marked the rise of the 1.35V kits.

G.Skill ECO Kit 4 GB PC3-12800U CL7-8-7 Review

G.Skill is offering a full range of P55 compatible memory kits, with the ECO being one of them. As the name implies, the memory aims to deliver the most performance while only requiring a mere 1.35V to run at 1600 MHz CL7-8-7.

www.techpowerup.com

 

[AusWolf]

It's not an issue with the normal 2 dimm 32gb config. I have no issues running 6000cl30 with 1.0.0.7

This would have been my question to the article, thanks.

 

[btarunr]

AMD’s biggest weakness is the memory controller. It needs improvements across the board (pun intended).

Intel designs its own memory controllers that it extensively iterates and tests on FPGAs. AMD uses third-party IP blocks such as from Synopsis.

 

[R-T-B]

Hmmm

AMD's 'Ryzen Burnout' Fixes Reportedly Plagued With Bugs

The bugs keep getting worse for AMD's AM5 platform.

www.tomshardware.com

My man you are literally posting in the TPU thread on the exact same news piece.

 

[MarsM4N]

My man you are literally posting in the TPU thread on the exact same news piece.

He's just setting his flags for "Team Blue".

 

[R-T-B]

He's just setting his flags for "Team Blue".

View attachment 295052

Anyone who literally thinks either of these companies give two shits about you is extra special. To be a "team" requires them to well... notice you, and they do not care about you as an individual, I assure you.

 

[Why_Me]

Anyone who literally thinks either of these companies give two shits about you is extra special. To be a "team" requires them to well... notice you, and they do not care about you as an individual, I assure you.

My bad .. I deleted it.

 

[Minus Infinity]

Zen 4 is the gift that just keeps on giving to Intel, So glad I didnt rush out and update my 3700X for 7800X(3D). Definitely hoping they get their crap together for Zen 5, but openSIL is years away so no guarantee AGESA won't continue to be a crap shoot for the next release. Arrow Lake is looking more and more appealing if those clowns can release on time.