Ryzen Owners Zen Garden

[Samiam66]

Amd's have been cool

Was a intel Xeon user before... Hp Z workstations (5 of them)

Memory compatability has been a ongoing issue for most ..and Ryzen 7 seem to get a little warm on air
Aio are a lot quieter .....I find myself leaning over too make sure the pump light is on
Radiator has 2x120mm pushing & 2x120mm pulling & I still have my DVD-RW & a 3.5 IDontix hot swap
drive bay up front in the normal position..




Sam

 

[Kanan]

I stay the hell away from liquid metal. It's horrendous stuff and there's so much potential for it to destroy and it's barely better than paste. Just about the only valid reason to buy some is to experiment with it out of sheer curiousity.

Sure, if you need those last two degrees then it can give you that, but if your overclock is that close to instability you have bigger problems than what TIM you're using.

IMO, liquid metal is for experienced users who wanna take a risk and especially designed for hardcore overclockers. Everybody else should use the best TIM available and stay away from liquid metal, as it poses too many unnecessary risks for lowish gains. This has nothing to do with delidding though, I'm talking normal usage. If you delid, you basically have to use liquid metal, else it is not even worth delidding in the first place, it would be like, buying a fast racing car and driving it slow around the track.

PS. I had great gains in gaming loading times (actual missions in-game) through the improvements I talked about on the other page. Tested MechWarrior 5 so far. It seems, the clocks are clearly higher now and more stable too.

Result

www.3dmark.com

new 3DM11 record with the new PBO settings. TimeSpy & Firestrike barely changed. All in all, 3DM11 Physics is 6% faster with PBO than without, as shown here: https://www.3dmark.com/compare/3dm11/13828043/3dm11/13952611

 

[tabascosauz]

Was able to hit 3800 just relaxing tRFC slightly and upping to 1.41V, and IMC didn't seem to have any problems at 1.1V SoC, but the memory ran into an error 50% through HCI. By that point, the sticks were up to only 44C but with close airflow from an A14 iPPC-2000, so I'm not sure if I'm going to explore the rest of the headroom for DJR up to 1.45V. B-die performance only gets further and further away with frequency, and it was only a 1.5-second improvement 3733 to 3800; 3600 to 3733 was a 5-second difference.

3800 did make it through Membench's short built-in HCI test, but after my 3600 16-18-18 escapades I'm not risking data corruption again with "almost-stable" RAM.

3733 is 200% stable at 1.38V/1.1V, overall rather impressed with both what was just a 3600C17 bin of DJR and a rather mediocre (on the cores side) 3700X.

One thing that I have noticed is that my tRFC doesn't scale to what Reous thinks is DJR territory (https://www.hardwareluxx.de/community/attachments/trfc-v21-png.508862/), only the typical range for CJR (right at the tip of CJR @ 260ns with 471 at 3600, according to integral), even with extra voltage. Interesting, since otherwise the DJR seems to be performing rather well.

Any CJR/DJR users can chime in with the minimum tRFC they can do at 3600 or 3733? I can do 450 @ 3600, but it offered no benefit over 471 so I didn't test any further, any lower than 450 is a no-go.

 

[TheLostSwede]

With some inspiration here I changed some settings: lowered PPT to 170 W from 230 which works as well, 85 limits the CPU too much on all core clocks. Lowered TDC to 170 from 200, this is also to increase all core boost, which benefited greatly, 60 limits it a lot. And finally using the 1 A EDC trick @Mussels suggested.

Before: CPU all core boosted in Cinebench to about 4050-4150. Not very good. With mussels settings, it was 4000~ all core, but his 1 EDC trick fixed my single boost which was lowered from 4391 to just 4325 over time and now is better than ever at 4450 MHz on multiple cores, which I'm happy with. With the new settings as in the pic, and the 1 EDC bug or trick, the all core boost in CB20 is 4200 MHz constantly and 4150 MHz in Prime95 maximum stress test, which is 150 MHz more than with my old settings which were 230/200/200, as suggested by @buildzoid. As per stock, the CPU only boosts to 3600 MHz in Prime95 maximum stress. For reference: I'm using a 3700X with a Noctua NH-D14. CPU was bought a few months ago.

PS. CB20 score went up to 4824 from about 4600~.

I prefer using PBO over oldschool overclocks, since it leaves the CPU running normally otherwise.

Try using Alt+Print Scr for screenshots, they're so much more legible than taking a picture of a screen...
You can also just Ctrl+V them into posts here.

 

[Kanan]

Awesome clocks with a 3700X after the new tweaks from last page.

 

[Mussels]

Try using Alt+Print Scr for screenshots, they're so much more legible than taking a picture of a screen...
You can also just Ctrl+V them into posts here.

or use TPU capture

 

[Waldorf]

@Chrispy
sorry, but ppl that improperly apply/use LM should either RTFM or leave it to someone who can.
the biggest problem with chip heat is to transfer it away fast.
most LMs have up 10x the transfer ability of "normal" paste (even with silver/copper maybe 5-15W/mK),
so especially with smaller HS/lots of heat you will see much higher gains than on big/low heat chips.

but i like it as it means i have less material between block/HS and have yet to see anyone that can apply TP as thin as it can be with LM.
and as long as you replace it for 1 or 2 more times over a couple of month (so it can saturate block and HS), it wont dry out like all paste will eventually do.

running my 3700 with (auto) pbo on a silent AIO and i dont see temps go past 70C under full load (more depended on room temp),
and is with the added 2080S (full block).

 

[Kanan]

Liquid metal vs usual TIM.

Low gains as I said. And like I said, only worth for delidding, and basically a must then too, as normal TIM will not do the trick.

 

[Mussels]

Back on topic

My lapped, liquid metal TIM 2700x has its final numbers in
4.1GHz all core OC, 1.337v (proud of that), 16GB 2667MHz 1.2V - 62.6C max load with looped cinebench

Also managed to get my 3700x system to 3200MHz C14, which has never worked before this latest AGESA.

 

[Kanan]

1.337v (proud of that)

Great. Job done.

 

[tabascosauz]

running my 3700 with (auto) pbo on a silent AIO and i dont see temps go past 70C under full load (more depended on room temp),
and is with the added 2080S (full block).

I would sure as well hope it's under 70C and silent with a big 280mm, bro. I top out at 73C in stress testing on a DRP4 that also makes very little noise at max speed, and I don't use LM. That's my hot-ass 3700X pulling 1.35V @ all core 4.125 with 83A PBO.

Liquid metal vs usual TIM.

Low gains as I said. And like I said, only worth for delidding, and basically a must then too, as normal TIM will not do the trick.

Optimum's 8700K is already delidded. And it's an 8700K, so it doesn't have any of Ryzen 3000's thermal density problems. Being a delidded 6-core Coffee Lake, the only issue at that point is how well you can push the air that the HS has soaked up out of the system; the Pallas 120 from that video is a kinda suck-ass air cooler that is only relevant to the Ghost S1 because the S1 can't fit any good air coolers. Pallas' six heatpipes are pointless because it has barely any fin surface area, so LM isn't going to do it any favours because it can only handle so much heat.

Add to that the fact that the S1's airflow for air cooling is laughable at best, and I can easily see why LM in that test barely makes any difference. I'd be more interested in seeing how the C14S, U9S and D9L (all of which I have and fit/have fitted in the M1) do with and without LM, on Ryzen 3000, in the M1 or Streacom with intake fans.

On that note, Cooler Master's NR200 is a really interesting case and that $80 price tag is really turning heads as it seems to be a very solid budget direct alternative to the M1.

 

[Kanan]

I would sure as well hope it's under 70C and silent with a big 280mm, bro. I top out at 73C in stress testing on a DRP4 that also makes very little noise at max speed, and I don't use LM. That's my hot-ass 3700X pulling 1.35V @ all core 4.125 with 83A PBO.



Optimum's 8700K is already delidded. And it's an 8700K, so it doesn't have any of Ryzen 3000's thermal density problems. Being a delidded 6-core Coffee Lake, the only issue at that point is how well you can push the air that the HS has soaked up out of the system; the Pallas 120 from that video is a kinda suck-ass air cooler that is only relevant to the Ghost S1 because the S1 can't fit any good air coolers. Pallas' six heatpipes are pointless because it has barely any fin surface area, so LM isn't going to do it any favours because it can only handle so much heat.

Add to that the fact that the S1's airflow for air cooling is laughable at best, and I can easily see why LM in that test barely makes any difference. I'd be more interested in seeing how the C14S, U9S and D9L (all of which I have and fit/have fitted in the M1) do with and without LM, on Ryzen 3000, in the M1 or Streacom with intake fans.

Maybe this review in particular isn't great, but I can provide endless more sources:

Thermal Paste Round-up: 85 Products Tested

This much-needed update to our thermal paste round-up now includes 85 contenders. It's one of the most comprehensive comparisons available, so we hope you find it useful!

www.tomshardware.com

Here another one, and don't only discard it because it's Toms. This one is with a "good old" FX 8350 CPU. My point still stands, liquid metal is more or less only good for delidding purposes, for normal usage, a good TIM is sufficient, and is way more convenient. Hey, I use really old Arctic MX-2, it was way beyond the expiration date. Doesn't change the fact, that even with OC the CPU is still easily cooled, even with Prime 95 maximum stress test. Heat problems of Ryzen were greatly exaggerated.

 

[lexluthermiester]

if you're happy with DX9 low settings and okay with 20FP

No. Just no. 4k is pointless unless at least 60fps can be maintained consistently. If not, it's time to drop the resolution and a few settings. The Ryzen is not the bottleneck in that setup, the 750ti is.

 

[ne6togadno]

No. Just no. 4k is pointless unless at least 60fps can be maintained consistently. If not, it's time to drop the resolution and a few settings. The Ryzen is not the bottleneck in that setup, the 750ti is.

you are not funny

 

[HD64G]

Back on topic

My lapped, liquid metal TIM 2700x has its final numbers in
4.1GHz all core OC, 1.337v (proud of that), 16GB 2667MHz 1.2V - 62.6C max load with looped cinebench

Also managed to get my 3700x system to 3200MHz C14, which has never worked before this latest AGESA.

Very nice info in this post. I didn't expect that the latest UEFI that was supposed to support the 3X00XT CPUs would help RAM compatibility on Ryzen 2X00 series. Might test it myself when ready.

 

[lexluthermiester]

Liquid metal vs usual TIM.

Low gains as I said. And like I said, only worth for delidding, and basically a must then too, as normal TIM will not do the trick.

The use of LM must be combined with the use of proper cooling. The heatsink in that video is not one that can take advantage of the benefits LM has to offer. So of course the benefit was minimum. LM is only one variable in an equation of providing optimal cooling for an extreme OC. Normal, casual usage will never benefit greatly from LM.

you are not funny

Wasn't trying to be...

 

[Kanan]

Back on topic

My lapped, liquid metal TIM 2700x has its final numbers in
4.1GHz all core OC, 1.337v (proud of that), 16GB 2667MHz 1.2V - 62.6C max load with looped cinebench

Also managed to get my 3700x system to 3200MHz C14, which has never worked before this latest AGESA.

Some serious effort you did there, any pics for the 2700X system?

 

[HD64G]

 

[droopyRO]

How many ? ALL of them !

 

[Kanan]

Some news.

 

[Mussels]

Some serious effort you did there, any pics for the 2700X system?

havent taken any recently, its mostly used for my kid to game on (7 days/ark)
1070Ti also has liquid metal on the die and stock heatsink, lost around 10C - helps a lot with fan noise

 

[Kanan]

havent taken any recently, its mostly used for my kid to game on (7 days/ark)
1070Ti also has liquid metal on the die and stock heatsink, lost around 10C - helps a lot with fan noiseView attachment 162287

Looks great m8.

 

[droopyRO]

1070Ti also has liquid metal on the die and stock heatsink, lost around 10C

I have the same card. It's inaudible by default.
I got it second hand, never replaced the original TIM. But undervolting helped by lowering the temps by about 3-4 ºC. There is my "curve".

 

[Waldorf]

@tabascosauz
With the room at 74 (24C), fans running silent at 500-800rpm (74cfm at 1350rpm) and a pump only doing 70L/H is what surprised me.
and the 3700 doesnt even match my 3600 when it comes to (low) voltage needs so..

@lexluthermiester
Even that I agree the 750 is the bottleneck, there is no need for 60fps for every game.
Ive played Rivals thru about 4 times without once seeing the need for 60fps, no matter what res i played at (1440p and now 2160p).

 

[lexluthermiester]

Even that I agree the 750 is the bottleneck, there is no need for 60fps for every game.
Ive played Rivals thru about 4 times without once seeing the need for 60fps, no matter what res i played at (1440p and now 2160p).

That's a matter of personal perspective. 30fps was considered by many devs to be a standard in years past. There were even game engines than locked to 30fps. However, those days are behind us. 60fps is considered the golden standard now. As a gamer and a PC seller, I do not consider a system properly build/configured unless it can get a minimum 60fps consistently at the native resolution on offer by the display being matched to the system. The 20fps mentioned previously by @Mussels is completely unacceptable. 4k is nice, but not at 20fps and there is no way that 750ti will deliver that 20fps consistently.