The RX 6000 series Owners' Club

[turbogear]

Hmm, interesting, maybe some mem chips just doesn't like fast timings. Mine does do fast 2150 but there artifacts and stuff, not to mention that it losses performance at that clock. Also whats your temps like with that oc? My 6700xt when oced and running furmark can reach 100 junction and 280 watts power draw

Many RDNA2 cards don't seem to like VRAM OCed much higher than 2100MHz.
The indication for the point where memory OC is not good any higher you go is that you will see performance drop for any frequency above that point.

My 6800XT could OC VRAM only to 2110MHz at Fast Timing and 6900XTU could not memory OC at all but as I mentioned in previous post it seems that some setting choice made by AMD is limiting this.

The proof of that seems to be the new AMD XTXH-LC card.

AMD launches Radeon RX 6900 XT Liquid Edition with 330W TBP and 18Gbps memory - VideoCardz.com

AMD launches Radeon RX 6900 Liquid Cooled with even more performance AMD has unveiled its flagship GPU from Radeon RX 6000 series, the RX 6900 XT LC. A hybrid cooling solution attached to an external 120mm radiator with a single fan are not the only changes in comparison to the vanilla RX 6900...

videocardz.com

People found out that it is basically having same memory chips but different settings.
This card shares reference design with other XTXH cards and people found out that flashing bios from this reference 6900XT XTXH card to other custom XTXH cards works and unlocks higher memory clocks.

Regarding hotspot temperatures, my Liquid Devil also approached 95°C on water before.
I changed standard Thermal past with Liquid Metal to get temperatures below 80°C on hotspot.

 

[HD64G]

Spent a bit of time exploring undervolting and underclocking my 6900 XT Liquid Devil Ultimate today. There's some interesting behaviour to be found here, especially around what seems to be a pretty hard voltage/clock scaling step, where the chip requires a significant boost in voltage to be stable at all.

At stock, my GPU is a bit limited by cooling (single slim 280mm rad for the whole system), but runs ~2550MHz when not thermally limited (i.e. before the liquid reaches steady state) at stock, i.e. 330W and I assume 1200mV (that's what Wattman defaults to, at least), with the default range of 500-2614MHz. (It steps down to ~2.4GHz when it heats up, maintaining 330W GPU power.) Undervolting at stock is essentially a no-go - even 1175mV isn't Time Spy stable.* And that applies when stepping down to quite a lot lower clocks - even 2100/2300 isn't stable at 1175mV.
My stock Time Spy scores are ~18900 total/21500 graphics. (That's with a 5800X.)

So I decided to start at the other end: rather than starting at stock and stepping downwards as I go, I chose an arbitrary low clock and voltage setting to see if it was stable - I picked 1600/1800 1000mV. And it was. So was 975mV, and 950mV. For now I haven't tested any lower. I didn't see much power scaling between 1V and 950mV - both reported ~150-160W GPU power draw during Time Spy. Performance was pretty crappy though, at 15500/16700.

So I decided to start stepping clocks back upwards, seeing how far I could go at 950mV. 1700/1900? Check. 1800/2000? Check. 1900/2100? Nope. Not at 975 or 1000mV either. But 1850/2050 at 950mV is TS stable, and scores ~17000/18600, all while consuming about 180W, +/- 10W.

One sort of interesting thing (at least to me): whatever I set my max clock to, it seems to sustain ~60MHz less than that, rock solid. 1800=1740, 1900=1840, 2000=1940, 2050=1990.

To summarize: for a ~11%/14% performance loss (total/GPU scores), I've cut power draw by ~42% (assuming sustained 330W/190W draws). That's a win in my book. At the very least I've got a very attractive-looking low power draw profile for games that don't need all the power available from this GPU.

I'll make some tests to see if anything lower than 950mV is feasible, as well as exploring if there are any intermediate stable settings between that range and 1.2V. But for now, I'm pretty happy with the results of about an hour's work.


*note (and this applies to all my testing): I run dual monitors, and as I'm looking for something suitable for daily use I'm testing with that, which does seem to hurt stability somewhat).

Edit: I've been using Aquasuite to monitor my power and thermals, and it seems their "GPU Core" power reading is what AMD calls "GPU Power", which seems to be what HWinfo calls either GPU ASIC power or TGP Power. Either way, it should be full-board power, not just the GPU itself.

Edit2: UV/UC is completely stable from playing a couple of hours of Metro Exodus, and performance is not noticeably changed. Power draw in-game fluctuates between 110-130W in outdoor areas to 160-170W in indoor areas. And my GPU is happily chugging along at ~45°C rather than 60-ish, and the CPU is much cooler too as the water in the loop is much cooler. Very happy with these results.

In your shoes I would load the UV profile for all games apart from the ones that need max power and the benchmarks. Much less life-consuming workload for your mighty GPU that could last at least 4 years on.

 

[Butanding1987]

Many RDNA2 cards don't seem to like VRAM OCed much higher than 2100MHz.
The indication for the point where memory OC is not good any higher you go is that you will see performance drop for any frequency above that point.

My 6800XT could OC VRAM only to 2110MHz at Fast Timing and 6900XTU could not memory OC at all but as I mentioned in previous post it seems that some setting choice made by AMD is limiting this.

The proof of that seems to be the new AMD XTXH-LC card.

AMD launches Radeon RX 6900 XT Liquid Edition with 330W TBP and 18Gbps memory - VideoCardz.com

AMD launches Radeon RX 6900 Liquid Cooled with even more performance AMD has unveiled its flagship GPU from Radeon RX 6000 series, the RX 6900 XT LC. A hybrid cooling solution attached to an external 120mm radiator with a single fan are not the only changes in comparison to the vanilla RX 6900...

videocardz.com

People found out that it is basically having same memory chips but different settings.
This card shares reference design with other XTXH cards and people found out that flashing bios from this reference 6900XT XTXH card to other custom XTXH cards works and unlocks higher memory clocks.

Regarding hotspot temperatures, my Liquid Devil also approached 95°C on water before.
I changed standard Thermal past with Liquid Metal to get temperatures below 80°C on hotspot.

I want to flash that AMD XTXH BIOS on my 6800 XT..

 

[GamerGuy]

Eh, wasn't there a post from someone about me starting this thread as a bragging thread (or something to that effect)? I could have sworn I'd saw it on my phone while I was out an about. I was gonna be cool about it, and say that he/she is entitled to his/her opinion....got on my desktop when I got back and 'poof!' that post vanished (did I imagine it?), regardless, I'd lost an opportunity to play the card, danggit!

 

[the54thvoid]

Post was deleted. No drama required.

 

[GamerGuy]

Post was deleted. No drama required.

Good to see mods are doing a standout job here , been to many forums where mods don't seem to bother, and let threads devolved into chaos with flames hurled all over before they'd act.

 

[turbogear]

Good to see mods are doing a standout job here , been to many forums where mods don't seem to bother, and let threads devolved into chaos with flames hurled all over before they'd act.

Yes, that is nice thing about TPU that admins are very vigilant.
I saw that post and was afraid that a flame war will start here messing up our thread and I reported it.

I want to flash that AMD XTXH BIOS on my 6800 XT..

You can try it but I think it will brick your card.
It is not compatible with 6800XT.

The nice thing about ch341a programmer is if your card is bricked you can flash it again with correct bios.
Other thing to mention, when flashing the bios with programmer the card should be disconnected from computer and lying on your desk.
The power to the bios chip will come from the programmer.

 

[Valantar]

In your shoes I would load the UV profile for all games apart from the ones that need max power and the benchmarks. Much less life-consuming workload for your mighty GPU that could last at least 4 years on.

I've been using it full time since I set it, and I'm planning to keep it on At least until I upgrade my main monitor - this profile is plenty for 1440p60. Metro Exodus played great with the profile enabled and all settings maxed out (including Ultra RT + RT reflections) - it did dip below 60fps at times, Radeon Software logged an overall average fps of 59.7, with 95th percentile frametimes of 17ms, so it's been lower than 60 for a bit, but in a game like that it's nothing I can say I actually noticed (even with Vsync on!). I did see some minor instability (one crash just after starting the game, and some in-game 2-3-second freezes), so I stepped up to 1V, and they seemed to go away with no increase in power draw.

Finished ME yesterday, tried my hand at some Jedi - Fallen Order (working my way into the "The Fury X couldn't handle this" part of my backlog), which was locked at 60 (again, all settings maxed) with the GPU even clocking down a bit further, around 1800MHz, so it was clearly not 100% busy. And I'm not one of those "ultra settings or death!" placebo addicts - even with a GPU like this I'd be perfectly fine with tuning some settings for better performance (which will likely be everyday reality if I end up moving to an UHD monitor). Ultra settings are generally not perceptibly different from High anyhow.

My plan going forward is to keep the profile enabled for anything where I don't really need more performance. In those (likely to be few) cases I'll happily let my system get toasty - it can handle it.

Oh, and with this UC/UV profile I've been seeing ~45°C GPU core temps and +~10° hotspot temps, compared to ~60°C core and +~25° hotspot. Not all that happy with the latter, makes me wonder if the card could do with a refit, but ... meh. It's fine, especially with the UV. Even with hotspot temps close to 90 in worst case scenarios that's still 20° below anything dangerous. And that's in the middle of summer with 30°C ambient temps, so most of the year should be much better. As you say, this will at best extend the lifespan of the card, at worst do nothing at all except allow me a cooler and more pleasant system, and I can always dial things back to stock should I want to. Heck, with my QDCs I could even hook up my spare 240mm rad externally if I really needed that extra performance (or wanted to try my hand at OCing).

 

[GerKNG]

Hmm, interesting, maybe some mem chips just doesn't like fast timings. Mine does do fast 2150 but there artifacts and stuff, not to mention that it losses performance at that clock. Also whats your temps like with that oc? My 6700xt when oced and running furmark can reach 100 junction and 280 watts power draw

around 60°C edge, 78-85°C Hotspot (full load, max powerlimit of 242W) and the memory sits in the low 60s at peak.

 

[Garlic]

around 60°C edge, 78-85°C Hotspot (full load, max powerlimit of 242W) and the memory sits in the low 60s at peak.

I'm starting to think my card has bad thermal paste application, under full load at 280w the edge is in the 50s and the hotspot being 100+. Perhaps I'll repaste it sometime

 

[turbogear]

I want to flash that AMD XTXH BIOS on my 6800 XT..

I see you buying 6900XT XTXH card and flashing it with XTXH-LC bios and beating my score by 2000 points.

I'm starting to think my card has bad thermal paste application, under full load at 280w the edge is in the 50s and the hotspot being 100+. Perhaps I'll repaste it sometime

If the difference between edge and hotspot is 50°C then there seems to be like an issue with cooler having proper contact with the GPU.
If you have warranty, maybe you can ask the shop where you bought it to RMA it and maybe you get a replacement.

 

[Garlic]

I see you buying 6900XT XTXH card and flashing it with XTXH-LC bios and beating my score by 2000 points.


If the difference between edge and hotspot is 50°C then there seems to be like an issue with cooler having proper contact with the GPU.
If you have warranty, maybe you can ask the shop where you bought it to RMA it and maybe you get a replacement.

I ran a set of tests again, when gaming the dege is around 55 and hotspot around 85, and when running furmark edge around 50 and hotspots 50 degrees higher, so the edge temp is a bit..strange?
I asked the store I got it from, they said its running in spec "Below 110C" Guess I'll have to find another way.

 

[Valantar]

I ran a set of tests again, when gaming the dege is around 55 and hotspot around 85, and when running furmark edge around 50 and hotspots 50 degrees higher, so the edge temp is a bit..strange?
I asked the store I got it from, they said its running in spec "Below 110C" Guess I'll have to find another way.

That could be a coldplate flatness issue, a mounting pressure issue, a thermal paste issue, a bad contact due to too thick thermal pads issue, and a few other explanations. Different workloads will also create different deltas between edge and hotspot temps, particularly ultra-hot workloads like Furmark. In general, Furmark has the potential to kill GPUs (exactly because of how it generates some insane hotspots), so ... don't use it. It's extremely unrealistic in so many ways. Pretty much any highly demanding benchmark is a better solution for stress testing.

Still, a repaste and remount sounds like it would be in order, just to rule out those issues. As long as it won't void your warranty, that is.

 

[Butanding1987]

I had issues before with my 6900XTU Liquid Devil Ultimate. It could not run memory OC at all independent of Fast Timing or not.
To my surprise this is somehow limitations depending on the bios settings that we cannot influence.

I have moved since some time to new XTXH-LC bios and now my card memory runs at 2310MHz at Fast Time level 2.
I had to flash this with a special programmer.
I got some help from Hardwareluxx members.

Here is my latest Time Spy Score:
View attachment 211377

It is at place 5 in overall score for 6900XT and 5900X combination.
This was impossible with original bios where my memory would not OC at all.

I am still trying to find some more points.
The guys above me in this list mostly have even better cooling solutions.
For example I know Holzmann from Hardwareluxx.
He put some air conditioning next to his Mo-Ra (https://shop.watercool.de/MO-RA3-420-LT-black_1) cooling radiator to reach that score.
I don't own anything like that and don't plan to do so.

That is a very good score. Good job. And that VRAM overclock is something else.
That 5900x is holding you back. You need a 5950x. LOL.
You should be able to raise your CPU score by just doing a manual straight overclock — simple and easy. No need for PBO. Try 4700 all cores @~1.338v (VID, not TEL). You should get at least 16,000 CPU points in Time Spy. This is just for 3dmark so it should be safe.

I tried the Wild Life and Wild Life Extreme benchmarks just for fun.

I ran Time Spy with a moderate GPU overclock using the 5950x and breached 22k for overall score. But my score is invalid because of some “time measurement inconsistencies”. I ran it twice and both my scores were invalid.

 

[turbogear]

That is a very good score. Good job. And that VRAM overclock is something else.
That 5900x is holding you back. You need a 5950x. LOL.
You should be able to raise your CPU score by just doing a manual straight overclock — simple and easy. No need for PBO. Try 4700 all cores @~1.338v (VID, not TEL). You should get at least 16,000 CPU points in Time Spy. This is just for 3dmark so it should be safe.

I am still not at end of tuning my 6900XT.
The OC community at Hardwareluxx is providing some tips what one can do in MPT.
I have not explored all options that are offered in the latest Beta 6 version of MPT.
When I get more time I will do so.

I was myself playing with the thoughts of going to 5950X.
But on the other hand I think AMD will release at end of the year a new processor with this 3D cache that they say improves 15% gaming performance.

I tried Dynamic OC switcher feature from ASUS Crosshair VIII Dark Hero which does Manuel OC on all cores.
Unfortunately, my 5900X does not like OC up to 4.7GHz. I tried Core VID up to 1.35v.
You can select different OC frequency on CCD0 and CCD1 on Dark Hero in Dynamic OC Switcher.
My CCD1 is weaker than CCD0.
I tried CCD0 of 4.6GHz and CCD1 of 4.5GHz with Core VID up to 1.35v.
I got nice blue screens with WHEA errors even before I can start Prime95 test.
If I put both CCD's at 4.6GHz then PC does not boot at all on similar VID.
With PBO and Curve optimization the CPU runs up to 4.575GHz-4.6GHz at all cores in Cinebench with a score of ~8960.

How high can I go on VID?

How I improved my CPU score from 12500 in Time Spy to ~13500 was by just changing the RAM from G.Skill F4-3600C14-8GTZNB to G.Skill F4-3800C14-16GTZN.
This is Dual Rank memory with higher speed. The old one was Single Rank.
Dual Rank provides speed advantages.
I want to try to OC it to higher performance with tightening the timing like Igor shows in the link below but I need to put some cooling on it.

Does the new high-end RAM for Ryzen 5000 live up to its promise? - G.SKILL DDR4-3800 CL14 2x 16GB kit put through its paces | Page 3 | igor´sLAB

The Trident Z RGB and Neo series from G.SKILL have been around almost as long as DDR4. However, for the new Ryzen 5000 CPUs, G.SKILL has refreshed their SKUs and added new XMP variants specifically…

www.igorslab.de

Temps are going higher than 54°C already at XMP when I run TestMem5.
There is not much airflow around the RAM due to water cooling setup for CPU.
I will try to mount some 120mm fan in the area to cool both RAM and mainboard VRM.

 

[Garlic]

That could be a coldplate flatness issue, a mounting pressure issue, a thermal paste issue, a bad contact due to too thick thermal pads issue, and a few other explanations. Different workloads will also create different deltas between edge and hotspot temps, particularly ultra-hot workloads like Furmark. In general, Furmark has the potential to kill GPUs (exactly because of how it generates some insane hotspots), so ... don't use it. It's extremely unrealistic in so many ways. Pretty much any highly demanding benchmark is a better solution for stress testing.

Still, a repaste and remount sounds like it would be in order, just to rule out those issues. As long as it won't void your warranty, that is.

Alright, guess I'll repaste it sometime. However there's two stickers on the screws. Perhaps I could try remove it without damaging it? Anyways thanks for the help.

 

[Valantar]

Alright, guess I'll repaste it sometime. However there's two stickers on the screws. Perhaps I could try remove it without damaging it? Anyways thanks for the help.

Where are you located? If you're in the US, those stickers are meaningless. Elsewhere it's a bit of a crapshoot whether they affect your warranty or not.

 

[turbogear]

Alright, guess I'll repaste it sometime. However there's two stickers on the screws. Perhaps I could try remove it without damaging it? Anyways thanks for the help.

It's not easy to remove those stickers without damaging them.

In Germany the warranty is void if one removes the stickers, but there is solution to buy replacement stickers from Aliexpress.
You can find almost every brand warranty stickers there.
I bought some by myself.
Shipping time from China could be long. I think I received mine almost 3 weeks after order.

https://m.de.aliexpress.com/item/1005002063221667.html?spm=a2g0n.productlist.0.0.6ad361f2VD21d2&browser_id=63d275f6859048d7b9107db2770720e0&aff_trace_key=e9dbc6fb9c2945fe99cd0edfd867d407-1621197532413-08282-UneMJZVf&aff_platform=msite&m_page_id=nmnhgfmutmccasxj17b23bcf67a1bfdf797214a1ec&gclid=&_imgsrc_=ae01.alicdn.com%2Fkf%2FHd8059a16da38490d97dfcd0e654dceccs.jpg_640x640Q90.jpg

The thing that is also possible to break when opening the GPU are the pads thermal.
You should try to save these. The original ones are needed if you want to save warranty.

Other question:
Is yours a reference cards?
If yes then it s not easy to reply thermal past.
The reference cards don't use past but a thermal pad.
If you apply thermal past instead then it might not work at all, because there could remain a gap.
The reference cooler is not designed for having thermal past and therefore the gap is larger and one need to tighten the screws very hard to get then a contact if one uses thermal past.
By the way that is a special high thermal conductive pad that is not easy to find a replacement for.

 

[Garlic]

It's not easy to remove those stickers without damaging them.

In Germany the warranty is void if one removes the stickers, but there is solution to buy replacement stickers from Aliexpress.
You can find almost every brand warranty stickers there.
I bought some by myself.
Shipping time from China could be long. I think I received mine almost 3 weeks after order.

https://m.de.aliexpress.com/item/1005002063221667.html?spm=a2g0n.productlist.0.0.6ad361f2VD21d2&browser_id=63d275f6859048d7b9107db2770720e0&aff_trace_key=e9dbc6fb9c2945fe99cd0edfd867d407-1621197532413-08282-UneMJZVf&aff_platform=msite&m_page_id=nmnhgfmutmccasxj17b23bcf67a1bfdf797214a1ec&gclid=&_imgsrc_=ae01.alicdn.com%2Fkf%2FHd8059a16da38490d97dfcd0e654dceccs.jpg_640x640Q90.jpg

The thing that is also possible to break when opening the GPU are the pads thermal.
You should try to save these. The original ones are needed if you want to save warranty.

Other question:
Is yours a reference cards?
If yes then it s not easy to reply thermal past.
The reference cards don't use past but a thermal pad.
If you apply thermal past instead then it might not work at all, because there could remain a gap.
The reference cooler is not designed for having thermal past and therefore the gap is larger and one need to tighten the screws very hard to get then a contact if one uses thermal past.
By the way that is a special high thermal conductive pad that is not easy to find a replacement for.

Guess I’ll get some of those stickers, I don’t have the amd reference design (Thank god) instead I have the sapphire nitro so I guess paste should work fine.

Where are you located? If you're in the US, those stickers are meaningless. Elsewhere it's a bit of a crapshoot whether they affect your warranty or not.

I’m in New Zealand, no idea if it’s legal here, these sticker are probably one of the worst features on a gpu

 

[Butanding1987]

I am still not at end of tuning my 6900XT.
The OC community at Hardwareluxx is providing some tips what one can do in MPT.
I have not explored all options that are offered in the latest Beta 6 version of MPT.
When I get more time I will do so.

I was myself playing with the thoughts of going to 5950X.
But on the other hand I think AMD will release at end of the year a new processor with this 3D cache that they say improves 15% gaming performance.

I tried Dynamic OC switcher feature from ASUS Crosshair VIII Dark Hero which does Manuel OC on all cores.
Unfortunately, my 5900X does not like OC up to 4.7GHz. I tried Core VID up to 1.35v.
You can select different OC frequency on CCD0 and CCD1 on Dark Hero in Dynamic OC Switcher.
My CCD1 is weaker than CCD0.
I tried CCD0 of 4.6GHz and CCD1 of 4.5GHz with Core VID up to 1.35v.
I got nice blue screens with WHEA errors even before I can start Prime95 test.
If I put both CCD's at 4.6GHz then PC does not boot at all on similar VID.
With PBO and Curve optimization the CPU runs up to 4.575GHz-4.6GHz at all cores in Cinebench with a score of ~8960.

How high can I go on VID?

How I improved my CPU score from 12500 in Time Spy to ~13500 was by just changing the RAM from G.Skill F4-3600C14-8GTZNB to G.Skill F4-3800C14-16GTZN.
This is Dual Rank memory with higher speed. The old one was Single Rank.
Dual Rank provides speed advantages.
I want to try to OC it to higher performance with tightening the timing like Igor shows in the link below but I need to put some cooling on it.

Does the new high-end RAM for Ryzen 5000 live up to its promise? - G.SKILL DDR4-3800 CL14 2x 16GB kit put through its paces | Page 3 | igor´sLAB

The Trident Z RGB and Neo series from G.SKILL have been around almost as long as DDR4. However, for the new Ryzen 5000 CPUs, G.SKILL has refreshed their SKUs and added new XMP variants specifically…

www.igorslab.de

Temps are going higher than 54°C already at XMP when I run TestMem5.
There is not much airflow around the RAM due to water cooling setup for CPU.
I will try to mount some 120mm fan in the area to cool both RAM and mainboard VRM.

That sucks. Have you tried overclocking per CCD? My 5900x can run at 4.725GHz (CCD1) and 4.525GGHz (CCD2) at 1.25v (TEL) or 1.271v (VID).

I think we have the same type of RAM. Igor's settings are actually a bit loose. I can run mine at CL14-8-14-14-28 at 3800MHz, rock stable. I can run it at 3933MHz but with WHEA errors though I have yet to encounter BSODs so that's how I run it.

Samsung b-die is sensitive to heat. I encountered errors when temps hit 46 degrees Celsius, so I decided to watercool my two sticks (2x16Gb). I also tried putting a fan, which limited temps to 44-45 degrees but it still wasn't enough. So I had to put the sticks under water.

Take a look at my CTR 2.1 settings, maybe you can compare your settings with mine. I'm using Hydra now, which is a lot more stable. I love my 5900x but the 5950x is an entirely different type of beast.

 

[Valantar]

That sucks. Have you tried overclocking per CCD? My 5900x can run at 4.725GHz (CCD1) and 4.525GGHz (CCD2) at 1.25v (TEL) or 1.271v (VID).

I think we have the same type of RAM. Igor's settings are actually a bit loose. I can run mine at CL14-8-14-14-28 at 3800MHz, rock stable. I can run it at 3933MHz but with WHEA errors though I have yet to encounter BSODs so that's how I run it.

Samsung b-die is sensitive to heat. I encountered errors when temps hit 46 degrees Celsius, so I decided to watercool my two sticks (2x16Gb). I also tried putting a fan, which limited temps to 44-45 degrees but it still wasn't enough. So I had to put the sticks under water.

Take a look at my CTR 2.1 settings, maybe you can compare your settings with mine. I'm using Hydra now, which is a lot more stable. I love my 5900x but the 5950x is an entirely different type of beast.View attachment 211683

A bit OT, but where is Hydra available? Is it only through 1usmus' patreon?

 

[Butanding1987]

A bit OT, but where is Hydra available? Is it only through 1usmus' patreon?

Yes, it's supposed to be available only on Patreon. But if you look around, you'll see it on a Chinese website (leaked).

 

[turbogear]

That sucks. Have you tried overclocking per CCD? My 5900x can run at 4.725GHz (CCD1) and 4.525GGHz (CCD2) at 1.25v (TEL) or 1.271v (VID).

I think we have the same type of RAM. Igor's settings are actually a bit loose. I can run mine at CL14-8-14-14-28 at 3800MHz, rock stable. I can run it at 3933MHz but with WHEA errors though I have yet to encounter BSODs so that's how I run it.

Samsung b-die is sensitive to heat. I encountered errors when temps hit 46 degrees Celsius, so I decided to watercool my two sticks (2x16Gb). I also tried putting a fan, which limited temps to 44-45 degrees but it still wasn't enough. So I had to put the sticks under water.

Take a look at my CTR 2.1 settings, maybe you can compare your settings with mine. I'm using Hydra now, which is a lot more stable. I love my 5900x but the 5950x is an entirely different type of beast.View attachment 211683

With regards to RAM it seem that we have same thoughts.
I have this new RAM since a week now and was already looking for water cooling solution before you mentioned it.
Which waterblock do you use for you DDR?

I tried per CCD ocing using my motherboard Dark Hero built in Dynamic OC Switcher with CCD0 of 4.6GHz and CCD1 of 4.5GHz with Core VID from 1.3 up to 1.35v.
It was crashing.

The CTR2.1 version that I find publicly seem to miss some of the features that you show on your screen shot.
Is that version that you get if you have Patreon?

https://www.patreon.com/posts/ctr-2-1-rc6-ver-53600047

I would like to try Hydra as well if I can get it but seems like only available to Patreon supporters.

My problem with CTR2.1 version that I have is that it never completes Diagnose fully.
When running Diagnose, PC crashes and after restart it continues but does not find meaningful stable settings.
I tried manually but without good base from CTR and did not had too much success as I missed a good base settings from auto Diagnostic.

I tried your setting only doing change to PX profile as 5025MHz was unstable.
I notice that I was using lower voltage at P1 and P2.
Your setting seem to work one processor except for PX profile.
I did not spend too much time to check stability though.

Similar settings on OC Switcher results in immediate crash.
It seems CTR here does a better job.
I still don't understand why CTR is not able to run successful Diagnostic on my system.
The values it suggested were too high for the voltages it applied. I tried to stay around that suggested values and it did not work.
Now using your setting as base, I will try to see what mine can reach.




The results that I get on Time Spy CPU score though are similar to what I get with PBO + Curve Optimizer only without manual OC.
With your settings as well as with PBO + Curve Optimizer, the score is in the same range of 14300-14550 varying between the runs.

I think you get higher than 15000 because you have tuned your memory further.
Memory seems to bring a good boost. If you remember my CPU score was in range of 12500 with older 3600 CL14 Single Rank memory.
Going to 3800 CL14 Dual Rank brought about 2000 points more.

My new memory is still running at XMP profile.
I did not get the time to tune it further since I bought it and I need to get some better cooling solution for the RAM; maybe water cooling.

CPU score with your settings without PBO or Curve Optimizer.


CPU score in one of the runs with PBO + Curve Optimizer without CTR and without manual OCing the CPU:


As I said, it varies between runs. The above was one of the best CPU scores but it was not the best GPU run.
Nice would be if both will get top performance in same run.
Maybe I need to go to Safe Mode so that some other background processes does not steal my CPU performance.

I noticed that Aquasuite X software running in the background was causing me about 200 points in Time Spy CPU score.
When running it collect water loop, CPU, GPU sensor data and send it to Aquacomputer Vision and thus steal some processor resources.
The both above runs are with Aquasuite X service stopped.

 

[Butanding1987]

Make sure your BIOS settings are correct for CTR (global C-states and CCPC enabled, CO disabled, etc.). Your LLC is also important. For Asus the recommendation is 3. CPU voltage should be default.

Run Cinebench to check for vdroop. Safe voltage for P1 is 1250mv and 1325mv for P2. Your TEL will be lower because of vdroop, that’s why I added 21mv and 13mv for P1 and P2, respectively. Your Cinebench run will tell you how much more you need (check CTR window for the reading).

Disable PX profile in the mean time because it’s the hardest to tweak. Or you can try 4.95GHz at 1.425v for PX high, 4.875GHz at 1.4v for mid and 4.775GHz at 1.375v for low.

I’m using a RAM block from Barrow. About RAM, yes, 3dmark favors fast RAM.

Here are my RAM settings. You can also use TRFC=244 if it boots.

P.S.
I sent you a PM.

 

[turbogear]

Make sure your BIOS settings are correct for CTR (global C-states and CCPC enabled, CO disabled, etc.). Your LLC is also important. For Asus the recommendation is 3. CPU voltage should be default.

Run Cinebench to check for vdroop. Safe voltage for P1 is 1250mv and 1325mv for P2. Your TEL will be lower because of vdroop, that’s why I added 21mv and 13mv for P1 and P2, respectively. Your Cinebench run will tell you how much more you need (check CTR window for the reading).

Disable PX profile in the mean time because it’s the hardest to tweak. Or you can try 4.95GHz at 1.425v for PX high, 4.875GHz at 1.4v for mid and 4.775GHz at 1.375v for low.

I’m using a RAM block from Barrow. About RAM, yes, 3dmark favors fast RAM.

Here are my RAM settings. You can also use TRFC=244 if it boots.

View attachment 211929

P.S.
I sent you a PM.

Thanks a lot for info.
For bios settings I used suggestions from Yuri Bubli in the guide video.

https://www.youtube.com/36990d23-e2a9-4e73-98ee-0483c7e3aadd

For CTR2.1, he was only changing VRM (LLC) settings in the bios. Therefore I left CCPC, global C-states, etc on defualt.
I used to change these too on previous versions of CTR.
I will change these and try.

Thanks for the hint regarding vdroop.
I am still in learning phase when it comes to CPU and VRAM OCing.
In the past my focus was mostly to OC GPU to get best gaming performance but did not care much about Time Spy global ranking.