*Official* NBR Desktop Overclocker's Lounge [laptop owners welcome, too]

Time to ask the resident AMD master.

Brother @ajc9988 What is a good voltage range for the 5950X? Soc and Vcore.

 

TL;DR
Vcore = 1.32V (edit: anywhere from 1.2 to this is good)
SOC = less than 1.2V.



So, one way to start is to use the Clock Tuner 1usmus put out which now supports the 5950X just to see what it comes up with for voltage.

https://www.techspot.com/downloads/7378-clocktuner-ryzen.html

Generally, the AMD Vcore is good for transient peaks of 1.4-1.5V. But, you usually want to have around low 1.3V like 1.32V for your all core. But a lot of this is based on transient things like the cooling you are using, which definitely helps with AMD chips on electron leakage, which then helps with the voltage set for the frequency.

For SOC, 1.2V is getting up there. For earlier Zen chips (Zen and Zen+), you wanted like 1.1V or less (even though people were saying 1.2V and degrading their memory controller, then wondering why they couldn't keep 3466 or 3600 stable after a year of cramming voltage into it; see also that the MB mfrs would have get voltages 100mV higher than set voltages for the 300 series: https://www.gamersnexus.net/guides/...-guidelines-how-to-kill-cpu-with-safe-voltage).

The later Zen 2 and Zen 3 memory controllers are more robust than earlier ones. At the same time, I found on my earlier chips with Zen, if I went too high on the voltage on the SOC, I actually couldn't go as far on my memory overclock as if I backed it off a little bit. Around 1.05V-1.08V was all I ever needed for that golden sample 1950X on SOC. So I'd start between 1.0-1.1V and try to play there first. If you need, you can creep toward 1.2V for air and water. Some MB mfrs said 1.3V, but I'd be worried about longevity still (from what I know, the I/O die is still on GF 12nm, after all, not TSMC 7nm).

Unfortunately I haven't had a chance to play with any Zen 3 chips, so I would start with the general principles of overclocking any chip: watch your temps, start around 1.25-1.32V on VCore and slowly walk up all core until it stops or is unstable. Then check out how much voltage to go up 50-100MHz. Or start at a stable all core level and lower voltage until unstable, then start marking how much to go up each level.

If going LN2, I've seen I think it was Steponz use up to like 1.4V on SOC. Not for water or air though, and short runs because LN2.

Hell, at stock, many of the boards will shove 1.4-1.5V for single core boosting on Vcore (not something I would keep).

 

Vcore=1.35V @ 4875 Ghz
SOC=1.1V
After doing a quick check, I can do 4850 @ 1.32V depending on bench.

https://hwbot.org/submission/4676121_johnksss_3dmark11___performance_geforce_rtx_3090_53780_marks
https://hwbot.org/submission/4676114_johnksss_3dmark_vantage___extreme_geforce_rtx_3090_102124_marks
https://hwbot.org/submission/467611...e___performance_geforce_rtx_3090_135521_marks
https://hwbot.org/submission/4676616_johnksss_hwbot_x265_benchmark___4k_ryzen_9_5950x_38.139_fps
https://hwbot.org/submission/4676110_johnksss_3dmark___cloud_gate_geforce_rtx_3090_87498_marks
https://hwbot.org/submission/467659...i_core_with_benchmate_ryzen_9_5950x_31926_pts

When you say 1.32V. What clock speed are we talking about here?

I just watch the gamersnexus video last night and that did help me to understand a bit more about it. As one does not want to kill a cpu with too much.

That last 25 mhz from 4875 to 4900 on all cores is a killer. Sure, you can set it, but attempt to run anything and it's instant crash. I have seen the 124 error code quite a few times, but not sure what it means with AMD. On intel that usually meant raising or lowering cpuvtt or uncore voltage or more or less core voltage.

No LN2, but will eventually use the phase change which is geared to about -50C right now.

At stock the board will boost to 5049 on 1 or 2 cores, but usage and performance is all over the place.

Thanks for the info!

 

There is also a voltage, VDDG, that you need to be aware of. Up to or around 1.1V is good on it and can help to stabilize certain errors.

As to 1.32V, it really is dependent on cooling to a degree. Kind of what makes AMD a pain in the butt, due to how it behaves, it is a bit of trying to keep temps in control, because it is the voltage plus the heat that can degrade AMD CPUs.

 

On the VDDG i'm around 1.05V to 1.075V max.

What kind of clocks are you running on your AMD setup?

 

I don't have Zen 3 at the moment, unfortunately. They were out of stock when my MB voltage sense went out, so I picked up Z490 instead.

So what I am working off of is what people used with Zen 2 and Zen 3 from what I've seen in different forums, videos, etc.

VDDG and VDDP can help to stabilize mem overclocks on AMD and can go from 0.9V to 1.1V, depending on your need. But, they too can make things go unstable if too high.

It's kind of like finding the perfect ring to throw it through, plus having to deal with variance on silicon quality on top of it. Kind of why I like it (brings me back to those early 2000 days before Intel got so reliable and gave all the things your want or need for settings).

 

Was going off any of your prior AMD investments. I almost forgot you now have a Z490 setup as well.

I just reset board back to stock as My board became available at my Microcenter and do you know what stock voltage is? 1.474V

 

2240MHz and an another driver worked out for 8th. I think Antoine06 will be pretty pissed I expect he have to try above 2265MHz. Finally up in water cooling territory.

https://hwbot.org/submission/4676921_papusan_gpupi___1b_geforce_gtx_1070_18sec_394ms


Edit. Here is a new one to make it a bit harder for him, LOOL
https://hwbot.org/submission/4676947_papusan_gpupi___1b_geforce_gtx_1070_18sec_378ms

 

Within top 5 for 1070. 4th and in between the water cooled cards I'm quite sure I could push more but not enough to take 3rd. At least top 1 in Norway, LOOL


https://hwbot.org/submission/467782...e_gtx_1070_23min_35sec_708ms?recalculate=true


Edit.
Also 3rd place for GPUI v3.3 - 32B
https://hwbot.org/submission/4677853_papusan_gpupi_v3.3___32b_geforce_gtx_1070_15min_41sec_678ms/

 

Here go a few shots of CP2077 Maxed out at 1440P with Ryzen 5950X / GODLIKE and K|ngP|n 3090.

 

Can you see or feel any difference in the gameplay on CP2077 with 5950X vs 10900K? I was not able to tell any difference in the gaming experience moving between 10900KF and 7980XE. In some things one CPU would have a few more FPS than the other, but overall it was unremarkable and never anything I could "feel" while playing.

Edit: My question relates to CP2007, but my comment about my own experience is in general terms relating to games I have played with both CPUs, as I do not own CP2077.

 

Thanks for posting up the screenshots.

The modern games are developed to maximize 6-8 cores and any improvements in fps with more count depends on the title. Racing games, flight sims and some open world games can benefit slightly from the extra cores, but the difference is negligible.

Even previous gen CPU's would do just fine as most gaming bottle neck occurs from the GPU end. I think the sweet spot right now is with the 8-core 9900K series of CPU's.

The 10-core 10900K family with the improvements in IPC, high all core clock frequency and other minor improvements make it the new practical gaming CPU for 4K and VR gaming.

One thing that's interesting is how a lot of YouTube reviewers when comparing the latest AMD and Intel chips only show a stock untuned Intel chip paired with a slow ram that's also used for the AMD test (much more potential than amd chips). Most reviews fail to show the extra potential that exists with intel chips, which misleads people into thinking that the intel chip is capped at what is being shown, when there's much more to be unleashed. The counter to balance a tuned Intel chip when comparing is the AMD having more core counts.

Ex. My 9900KS @ 5.3GHz all core surpasses the in game fps that most YT'ers show to the point where it'd top the chart in their comparisons that they're showing.

The 5950x is one heck of a value of a chip for sure. Extremely efficient too for those who like good MPG, but I'd say give it one more refresh and I think we'll have a winner.

On the Blue side, slow to the party of smaller nm processes, but they'll have some pretty good chips in the coming future to balance things out again.

What Intel has over AMD right now is maturity. Despite the 14nm+++++++++++++++++++++++, look how polished the nm process is. Pretty incredible actually, despite being on the older process for so long.

What AMD has is the advantage of a newer 7nm process that "naturally," allows them to have better efficiency with power draw etc...ie... The efficiency isn't something credited to AMD, rather it just came by default due to working with a newer tech. They still need to iron out all the latency issues and up their core frequencies. Maybe next gen?

There's the view point of having 8-10 working at 100% @ high speeds vs 12-16 working at 30-40%. With most modern games and because it only utilizes 6-8-10 cores effectively, I still would rather have the 8-10 workers @ 100%, to maximize the cores. + faster RAM speeds and timing with (Intel).

With gaming today Horsepower still outweighs having a ton of torque.

 

You do have a great point. Despite the tedious nature of tuning AMD in general, it does give you that element to explore old school tuning. Being able to tune and to get your chip to its peak operating potential is very rewarding.

I get that for most consumers they desire a chip that's already running close to its peak potential out of the box, but for tuners like us although we welcome that, we'd also want to know that there is more headroom to explore.

@Johnksss when adjusting the VDDG and VDDP, load up your desired CPU profile and set your vcore soc voltage in the bios to 1.2v and work your way down in 0.1 increments. This should also help stabilize your FCLK.

 

Hummm, it's more of what I don't see... That stuttering.
I see this with my 10900K, but it's smooth with my 5950X. I'll need to recheck it after this board situation is resolved.

Also, it's only doing it in the video due to me recording it.

And I noticed it was less of that while playing CP2077 and BOCW. Could be my imagination though. I'll need to recheck when my mb situation is resolved.

@iunlock

No problem. I'll need to redo it as the FPS is not showing in the picture.

 

Oh interesting... I've seen that before.. Can you run it again with hwinfo64 open in the background with the CPU graphs (Core 0-4) open? Checking for spikes etc.. It could also perhaps be a voltage issue? Depending on where the current ddr voltage is set at, try upping it by a tad and see if the stuttering still happens. Also, what's the tRFC set to? If it's too low it could cause issues in game (benches) as well. Does it stutter even with stock RAM and stock CPU settings?

 

Oh... No, the 5950X does not do that when benching, but the 10900K does. I was just lucky that recording it was able to produce the same effect. I was going to highlight the problem, but instead it showed in the video.
And at the end of the day, not really a problem. And no way i'm giving up my 10900K. Not for an 11900K. (Way higher clocks than AMD)

 

So an update. Seems this time around took a whopping 36 minutes to come to a conclusion that they are replacing my motherboard.

"We have received your product at the repair facility, and it is currently undergoing diagnostic testing." 2/08/2021 - 11:54 AM
"We are unable to complete the repair of your RMA and will be reaching out to you shortly to confirm the replacement of your unit." 2/08/2021 - 12:30PM
"Thanks for your inquiry! Service center has arranged to exchange your device"

Now that has to be the fastest diagnostic I have ever had! 36 minutes.

 

It seems K|ngp|in is a bargain Even vs a good AIB 3090 card. Who is the first one to try get one? @Rage Set ?

GALAX RTX 3090 HOF preorder starting at 3900 AUD videocardz.de

All three global models have been listed by the Australian store: HOF, HOF Premium, and HOF Limited Edition. As we explained in the previous article on the series, only the latter two will feature a massive 4.3-inch color display. The pricing for the cards starts at 3899 AUD (3013 USD) and goes up to 4499 AUD (3477 USD). This is more than double the RTX 3090 MSRP. The series is targeted at ultra-premium builds. In fact, the cards were designed with extreme overclocking in mind so they are definitely not for casual gamers.

Yeah, only for the Elite who can get them via own channels. Do they need to pay to play? Probably not.




Btw. Its Wccftech but still

ASUS Z490 Motherboards Tested With Intel Rocket Lake Desktop CPU, Lack PCIe Gen 4.0 Support on M.2 Slots

Back in July, we reported how ASUS was severely lacking PCIe Gen 4.0 support on its Z490 lineup of motherboards, a highlight feature of the Intel Rocket Lake Desktop CPUs. Fast forward to 2021 and we can finally confirm this as our sources have got the chance to test out several ASUS Z490 boards with a Rocket Lake CPU.

But users who bought ASUS's Z490 motherboards to couple them with Intel Rocket Lake CPUs for Gen 4.0 compatibility would be left disappointed as they can't fully utilize their brand new CPUs & will have to purchase an entirely new platform to get proper support.

 

Man, i'm gonna get 2 of those 4500 dollar ones!!!!! Oh snap! I had to slap myself straight!

 

A few trial runs under the Single Stage.... Man, I can't just run my bomb profiles. Have to start all over from scratch again...smh!

 

LOL. While I would love to have one of those for an all white build, I do not believe they are worth an additional 1500 over the KP card. Plus, hardware is half of the battle. They won't give me access to the special "sauce" software that is required to fully open up that card. None of these AIBs can ship their cards with unlocked vBIOS.

I'll stick with my lowly KP 3090 card, hahaha.

 

https://valid.x86.fr/097rtd

 

I have finally tried a Ryzen CPU.

https://imgur.com/a/bK0Bbx4

 

I finished putting my brother's 5600x together. I've been having some fun playing with it a bit while I tweak it. I switched him from a SFF to a MicroATX (26L total) so he'll be happy.

 

Has anyone seen this case?
https://www.sfftime.com/product-page/p-atx-v2-0

This would make for some super badass sff builds. You could fit a 240mm AIO in this case if you got a dual fan graphics card instead of a triple fan one, and if you used a pico-PSU. On top of that, you can fit any form factor motherboard in it except for EATX boards.

Slim and portable ATX gaming build? Sign me up!

 

And they're also ugly as sin! Check out those turquoise crowns! Although I think the chintzy look is more popular in Far Eastern countries from what I've heard on popularity of this kind of look.

 

Yeah, you can say that again. Really crazy ugly. Galax GPUs are almost always tacky-looking, which is a pretty major problem when the PCB is custom and you can't easily swap their ugly stuff for a normal-looking waterblock. It's not like you can use it 24/7 with an LN2 pot strapped to it and it really would suck having to look at it the rest of the time in its stock state.

 

I'm currently putting these timings through the paces to qualify for stability (a 2 week process; pass HCI memtest to 1000% per 2GBx16 windows last night; next being TM5 with 1usmus v3 testing profile for about 8 hours (or 56-64 rounds, approximately); then the Extreme anta777 cfg file for TM5; then some P95 only on the range to test memory; then possibly some OCCT for mem testing both SSE and AVX; etc.; When I qualify ram, I qualify ram) while using 1.44V. I've ran 1.42 and 1.43V on these DIMMs for most of their lives without degradation and they are still going strong. But, if I ever let the heat go above 38C on their built in temp sensor, they will kick errors. But they have done that since I bought them (also why I have always used the corsair dual fan ram fans on them, to help keep temps under where they throw errors). Right now, about 35.8C is how hot the hottest DIMM gets, with the coolest one one degree less and the others about 35.5C top temp.


That's in an in-optimized OS for benching.

Does anyone know if MemTweakIt is around and working on a Z490 platform? Or something that can show more timings and sub-timings than just Asrock Timing Configurator ( link to version 4.0.3 that works on Z490 here: https://www.asrock.com/mb/Intel/Z170 OC Formula/#Download ).

 

Yes, I have it and use it, but it will not work on the latest versions of Windows 10 due to some kind of compatibility crap. It throws error messages on 2004 and 20H2. It works fine on 1909 and older versions of Windoze OS X. I can change timings on the fly with MemTweakIt on LTSC. It refuses to launch on 2004 and 20H2.

Nice job on the memory tuning.

 

Worked great, but then I found the mb was auto setting vccio and vccsa to 1.375V to do it.

I take it safe is still around 1.25V, right?

And looks like I'll need it once I setup the bench drives

 

I doubt even 1.375v will damage anything. 1.25v is perfectly safe. I keep mine at 1.2v each since I'm only using 4000Mhz CL16 on 4x8gb dimms for daily.

 

That should be fine. VCCIO is safe up to at least 1.450V. That is within the "normal' range according to the legend in the Z490 Dark BIOS. I am using 1.450V for VCCIO and SA on both desktops.

I run my laptop (Tongfang turdbook) with the memory at 1.350V and both desktops with the memory at 1.550V or 1.600V depending on what kind of overclock. I did the same all the time on the X299 Dark. The BIOS in the Z490 Dark shows the "normal" range for RAM voltage goes up to 1.600V. I never use less than 1.550V memory voltage on both desktops The memory never get warm and never feels warm to touch. Touching the heat sinks it always feels the same as ambient, with no discernible difference than touching the Wetbench chassis.

 

Whenever you decide you're ready for it, the W7 Z40 ISO and other links are in this post: https://www.win-raid.com/t7149f52-Comet-Lake-Windows-Z-Intel-Driver-1.html#msg126664

Just remember on the first reboot during W7 setup to press F8 and disable Driver Signature Enforcement to avoid a BSOD due to the modded drivers I added that are not signed.

 

As a backup plan, you guys could keep one of these around as well....

https://www.amazon.com/JXSZ-Keyboard-Expansion-Profile-Bracket/dp/B07VB9TD6D

 

That will definitely work. I was actually going to buy one and couldn't find a super cheap PS2 keyboard and mouse at the time. I found a couple that were fairly expensive, but nothing cheap enough to consider buying it to use an hour or less. I wasn't expecting that. I had no idea that PS2 keyboards and mice had become so rare. Seems everything is USB now. I only looked on Amazon, but didn't check on eBay because I didn't want to have to wait a week to get it.

 

I think I got it from Ebay for like 30 bucks as the beginning of last year.

Your local mom and pop computer store did not have a ps2 keyboard or mouse?
But also you can use a hard wired usb keyboard and mouse with that card as well. My wireless Logitech does not work. (You need the 10 pin to usb port connector though. That can come from an old computer case.)

 

Probably so, but I don't even know where the closest mom and pop computer store is, LOL. I don't think there is one close to me. Places like Fry's and Best Buy have killed so many of them, and I'm not very close to anything (as you probably remember) other than your typical suburban big box chain stores and fast food outhouses.

 

Pretty sure i can ship you one, but it may not be in the best shape as they have been hanging around for a long time.

I just have to look first.

 

Yeah, tried dropping it that low. It was a no go. But at 1.3V for both resulted in 1.27V on VCCIO and 1.33V on VCCSA.

Also, these are about the oldest premier B-die you could get. I bought the two g.skill trident z 4133 19-21-21-41 kits a month or two apart from Newegg in late 2016 (so very early b-die) at $210 a pop or around there (right before the 2017 ram crazy pricing). Turns out, back then so few were buying them that I wound up with sequential ram. All 4 DIMMs are sequential off the factory floor. LOL.

It cannot do as much as newer B- die and seems to also stress the IMC more than newer b- die. I'd need to get a newer kit to confirm, but we are so close to DDR5 I kind of want to limit buying more DDR4 high end to play with. Alder and zen4 should both have DDR5 of rumors are correct, and that is double the bandwidth and possibly 6200+MT. I hate buying at the end of a generation of tech. At least if you buy at the beginning, you can always use it even as the platform ages.

Yeah, I got them down a little. And then tightened tFAW to 16 and am testing tightening tCKE to 1. I had 3 errors in 4 hours with it tightened and using 1usmus configuration v3 on TM5, so bumped up the voltage by 10mV on the ram.

I tried tightening the twtr l and s (write to read delay), but it was having none of it.

Also, it didn't seem to change the rtl I set in the bios, so have to figure out why.

And thanks for the win 7 image

Edit:



Where it is at now. I'd still like to get the read over 60000, but don't know I'll get there.

 

https://hwbot.org/submission/4682349_papusan_y_cruncher___pi_25m_core_i5_2430m_8sec_164ms


https://hwbot.org/submission/468240...re_i5_2430m_11min_6sec_793ms?recalculate=true


https://hwbot.org/submission/4682353_papusan_y_cruncher___pi_2.5b_core_i5_2430m_31min_56sec_975ms


https://hwbot.org/submission/468266...gle_core_with_benchmate_core_i5_2430m_576_pts


https://hwbot.org/submission/468274...hmate_core_i5_2430m_1366_pts?recalculate=true


https://hwbot.org/submission/468238...03_core_i5_2430m_1404_points?recalculate=true


https://hwbot.org/submission/4681849_papusan_superpi___32m_core_i5_2430m_11min_14sec_996ms

 

So, the system started randomly rebooting. I tried to troubleshoot, no dice. Still booted, but would not stay running. This is what I found on the mosfet heat sink...

 

woops

 

Yeah, came out by the part with the worm clip. I switched it from the ek to koolance qd3 with a compression fitting on it.

Turns out, spillage on the old mb in the same place. Slow drip that started after the 24 hour testing

 

Its why I went air this time around, decided to be lazy. Love the temps it affords me but despise the heat dump as a result of pushing things harder lol

 

Oh man, so sorry to see that happened. Did it actually damage anything, or only rebooting due to a temporary short?

That does not look like a QD3 fitting.

 

Meh... Linux still isn't performing with the CPU as good as Windows, but not way off. Just a little bit lower than Windows. This is with Intel vulnerability mitigations disabled using the Grub command. Maybe because this is running under Wine (no Linux version available) and the emulation carries some overhead.

 

That was the EK fitting with the worm clip. I switched it for the QD3. Do to the bend in that run to make it fit in the area, it put extra stress on the hose at the worm clip. So over time, it eventually leaked from there. The other part of the run with the EK fitting is dry as a bone but has way less stress on those clips.

Either way, I used the QD3 extra I had on hand and ordered two more QD3 compression style. Then, since Valentine's day sale, I just bit the bullet and bought the EK manifold and 5 in and out QD3s with g1/4 fittings so they screw directly into the manifold. That way, all systems will run in parallel to the manifold, then on to the 4x480mm rads.

I'll take some pics once setup.

I noticed also that there was a pile of glycerin (the base for the clear fluid I use; Mayhem's additive) on the old z170 board. So I cleaned that off. Hoping that just lights up today as well.

When I turned it off for an hour yesterday, before finding the spill, it then booted to windows, so likely a temp short. I've had it out drying all night and will remove the heatsink here in a little and clean with alcohol as needed. Pretty sure nothing dead. But that will be verified in a couple hours.

 

Oh, OK now I gotcha. I thought you were saying that the EK part leaked on the old system and the QD3 was leaking in the same spot in the loop on the newer system. I misunderstood.

Sounds like it will be a nice setup. Looking forward to the pics. It would be great to find out the Z170 board was still good and only had a temporary short.

 

I don't know if it is the BIOS or Z490, the 3090 GPU or recent changes to Linux, but I had a helluva time installing Linux on the Z490 Dark. It gave me more fits than installing Windows 7 did. I run the BIOS in Legacy mode (I hate UEFI) and it wouldn't even boot. It would hang. And, there was no way I was going to reinstall Windows 10 Pro, Windows 10 LTSC and Windows 7 in UEFI mode only so I could run Linux.

What I ended up doing is setting UEFI mode and once I got POP_OS! to boot fully, I used a manual installation process, set an MSDOS partition table to the MBR disk and manually created the partitions rather than letting Linux decide how to install itself. After that was finished I switched the BIOS back to Legacy and everything worked. I immediately ditched the uber-ugly stock POP_OS! desktop environment and installed Cinnamon. Other than the tacky UI, I find everything seems to work better for me using the POP_OS! distro. It is like Linux Mint on steroids. System76 does a great job of producing a reasonably functional Windows OS substitute. Other than being kind of worthless if you're into benching or gaming, Linux is so much better than Windows. It has come a LONG WAY for gaming. Not totally worthless for gaming, but definitely not wonderful at this point. But, it's still as worthless as the tits on a bull for anyone that enjoys benching. The tiny sample of benching applications that exist are mostly garbage.

 

OK, I was mistaken. This was actually much closer than I had thought. I don't normally bench anything except at my maximum functional overclock because I don't really see any point in benching that way. Like, why even bother wasting the electricity for that, LOL. I think this is close enough to call it as being within a normal margin of error. Same CPU and RAM settings in the BIOS, just booted W7 to compare to POP_OS! out of curiosity.