What air cooler to get for 3950x? I don't want water cooling.

I am not sure where you have read the Noctua D15 hit 95C with 3950x or 3900x with stock frequency. I have a 3900x with Noctua D15 and I am getting 76C with 100% cpu utilization with ~165W package power. It is a fact: not myth! 80C@~200W package power. I had made it very clear in the past post. All this is for stock and undervoltage without overclocking/PBO. Wraith Prism works because it is measured against package power with default locked power limit. (~165W)

Let me be clear, I never said that it did. I said that air cooling wasn't recommended, nor sufficient to achieve the maximum potential performance of a 3950X. It seems that people running on air cooling get lower clocks than those who aren't. There may be something else in the boost clock algorithm that causes this when specific temperatures are reached. That is, if your temp hits X amount in X time at 100% utilization, it downclocks to a lower speed. I do not know. AMD doesn't make its boost clock algorithm details known. However, AMD is clear in its press kit for the 3900X that better cooling will produce better results. They even used the D15 as an example of good air cooling. Again, AMD doesn't recommend air cooling at all for the 3950X.

Was this test done while cpu was at 100% utilization? When my PBO is on auto so I see that all core can go 4250 before but drop out like 5 sec later to 4100/4050.

Of course. Mine starts off at 4.3GHz or higher and then drops down. It settles at 4.12GHz using AutoOC in Cinebench R20 and remains there until the test completes. Frankly, Cinebench R20 is a bad test for this as it's over too quickly.

All,
I am not sure you guys understand what I had been implying but I will make it explicit. Please post your package power along with your temperature. Under utilized cpu will run cooler than fully utilized cpu at same frequency or higher frequency. Prime95/OCCT does not fully max out cpu utilization. Cinebench R20 can peak fully at one point but does not constantly. PBO is consider to be overclocking up to +200MHz on top of PB2. PB2 is turbo boosting: it is not overclocking and is within AMD specification.

I did. 40% under PB2 and 41% under PBO or auto-overclocking. Under PB2, temperature is 68c. Under PBO/AutoOC its 79c or more at full load using Blender / Gooseberry. Under Cinebench R20, it only goes to about 76c because the test doesn't run long enough to allow the CPU to get warmer. And yes, I understand that PBO is overclocking. PB2 is stock operation, but again, even so your clocks are influenced by the same values. If you run weaker cooling, you may be limited by your CPU temps before you run into other limits. It looks like AMD targets 68c for these under PB2 and as a result, the use of different cooling solutions will produce different results. A better cooler will allow for higher clock speeds. AMD even states this is the case.

I stand by my statement. A D15 will probably work, but air cooling isn't recommended and it isn't enough in my opinion. If you use air coolers with the 3950X, you will lose performance. Whether or not that's acceptable to you is another matter entirely.

And I do not want to install that shitty Wraith Prism on the 3950X just to prove that it won't be worth a shit.
 
And I do not want to install that shitty Wraith Prism on the 3950X just to prove that it won't be worth a shit.
I agree with you there and thank your emphasizing it!!! But there are people out there looking for the cheap and just want good enough to get by and It can be done. For those few folks in the North does have the advantages! :) I disagree on reduced performance and that is my personal experience from testing.

I have run
It seems that people running on air cooling get lower clocks than those who aren't. There may be something else in the boost clock algorithm that causes this when specific temperatures are reached. That is, if your temp hits X amount in X time at 100% utilization, it downclocks to a lower speed. I do not know. AMD doesn't make its boost clock algorithm details known. However, AMD is clear in its press kit for the 3900X that better cooling will produce better results. They even used the D15 as an example of good air cooling. Again, AMD doesn't recommend air cooling at all for the 3950X.
My 3900x had ran at ~200W constantly for 24 hours with no reduced performance. This is part of my criteria for stability when I first buy cpu or motherboard. Stress it hard at stock frequency and while within return period! :) You had missed it again, I said 3950X with undervolt should yield 200W package power, the same as my 3900X at stock. Anyway if i had gone 3950X, I would have no hesitation to buy Noctua D15 again!

It looks like AMD targets 68c for these under PB2 and as a result, the use of different cooling solutions will produce different results.
I am not aware of this! I will do a little more testing this weekend to confirm this.

My current stock frequency with 3900x -.1V offset cinebench R20 is 7384. Any score within 40 is consider the same imo.
 
I agree with you there and thank your emphasizing it!!! But there are people out there looking for the cheap and just want good enough to get by and It can be done. For those few folks in the North does have the advantages! :) I disagree on reduced performance and that is my personal experience from testing.

I have run

My 3900x had ran at ~200W constantly for 24 hours with no reduced performance. This is part of my criteria for stability when I first buy cpu or motherboard. Stress it hard at stock frequency and while within return period! :) You had missed it again, I said 3950X with undervolt should yield 200W package power, the same as my 3900X at stock. Anyway if i had gone 3950X, I would have no hesitation to buy Noctua D15 again!


I am not aware of this! I will do a little more testing this weekend to confirm this.

My current stock frequency with 3900x -.1V offset cinebench R20 is 7384. Any score within 40 is consider the same imo.

That's a good score
 
My 3900x had ran at ~200W constantly for 24 hours with no reduced performance.

And you compared that to a watercooling setup? I mean I get what you are saying, but I guarantee you, TR's perform better under water. If you can achieve a decent score in Cinebench on air, I'd lay money under water it would be better with a decent setup.
I personally wouldn't run a TR under air. Even my crummy 1900x, I wouldn't run on air, and the 2920 I had, ran great due to the XSPC Raystorm stuck to it.

You make fair points, but until you make the actual comparison, I don't know that you can make a valid argument without all the data. Take the plunge man.....
 
That's a good score

With just amperage tuning that's pretty good.


I agree with Dan, I'd never use the wraith prism cooler on the 3950x. I can't even stand it on my 3900x. On a 3950x man it would spin full rpm on a web page....

I just got another d15 which should show up on Saturday.
 
Last edited:
Thanks again for your analysis. I guess I was thinking that my chip could pull higher clocks at lower power draw due to some higher quality silicon, but, I agree with you, I'm thermally limited and can't see it until I unlock it all and release the hounds. I'll find out tonight and post back.

Did some more testing tonight:

I set my limits to match what Dan_D shows on his for PB2 since my mobo doesn't go above default. I also upgraded to the newest Ryzen Master and it seems to push a little harder and run a little hotter.
PB2 01-09-20-2.jpg

So, somewhere around 4033 MHz and 80 degrees with auto overclocking enabled.

For kicks, I also did some manual OC'ing. I got a successful run of CB20 at 4.35 all core on 1.3v. Almost hit 10,000. I tried 4.375 at the same voltage but it failed. Obviously, not a stability test.

Manual OC 4.35 01-09-20 -1.jpg

Also, once again tonight, when I booted my computer the first time, I had to reset CMOS after it kept failing to post and eventually started in safe mode. I got a lot of 8d codes, so I manually set SOC to 1.11250v. That's just slightly higher than it goes auto. Hopefully, that will take care of it.
 
Keep in mind that it's not the water that cools the CPU. It's the air flowing through the radiator that does the cooling.

A good air cooler is like a compact AIO, except you have heat pipes instead of water to move the heat from the CPU to the radiator surfaces.

The real advantages of water/AIO are:

1) The radiator is mounted to the exterior of the case, so it's easier to get unimpeded flow of cool ambient air. With an air cooler you have to pay attention to case ventilation to ensure the hot air is pumped out of the case. Ducting air coolers provides a decent boost, which is why OEMs do it. Ducting would be more popular with the enthusiasts if everyone didn't have giant windows in their cases.

2) Water radiators can be bigger than air coolers. This is why I think water is only interesting with 280mm rads or larger. You just can't get the same surface area of something like a 360mm or 420mm radiator in an air cooler.

3) Water has thermal mass to soak up temperature spikes. If your workload is bursty with a lot of idle time, the thermal mass of the water and radiator setup can smooth out the system temperatures a bit.

I was a strong believer in water cooling until I had to temporarily swap an NH-D15 into my main system. To my surprise, the temps weren't that much worse than my custom loop. I left the NH-D15 in because it's lower maintenance than a custom loop, works well enough, and I don't have to worry about leaks.

If you haven't tried an NH-D15 and you don't already have a nice AIO, you really should give it a chance.
 
I agree with you there and thank your emphasizing it!!! But there are people out there looking for the cheap and just want good enough to get by and It can be done. For those few folks in the North does have the advantages! :) I disagree on reduced performance and that is my personal experience from testing.

I have run

My 3900x had ran at ~200W constantly for 24 hours with no reduced performance. This is part of my criteria for stability when I first buy cpu or motherboard. Stress it hard at stock frequency and while within return period! :) You had missed it again, I said 3950X with undervolt should yield 200W package power, the same as my 3900X at stock. Anyway if i had gone 3950X, I would have no hesitation to buy Noctua D15 again!


I am not aware of this! I will do a little more testing this weekend to confirm this.

My current stock frequency with 3900x -.1V offset cinebench R20 is 7384. Any score within 40 is consider the same imo.


Show me a screen of that score and some temps because I for one do not believe you on an air cooled 3900x stock except a voltage offset. Sure I can get above that if I clock mine all core to 4,3Ghz with my power setting to MB instead of Processor. But i have yet to see a 3900x in it stock form score much over 7000 every single run. I have to have my chip start off nice and cool and stay there to get 7000 consistently. When I ran the D15 on it and Prime 95 for more than 20 minutes it would peg at 95 degrees, that is with just the default clicking OK for the test and not changing any options. I tried many coolers and many cases and a few AIO's and Air coolers and I have never seen an air cooler not hit 95 degrees under full load for any amount of time. Hell overclocked to 4.3 it struggled to keep it below 80 without a load. Now it sits nice and pretty and barely goes above 70 with 100% load if it ever does. I want to see pretty screenshots that you got a 7384 on R20 with Ryzen Master screen cap showing me what it up because you must have got the best chip and I got the worst.
 
264507_r20.jpg

4 - 4.5ghz to 4.3ghz ratio
3 - 4.3ghz all core
2 - 4.2ghz all core
1 - completely stock 3900x
 
View attachment 214783
4 - 4.5ghz to 4.3ghz ratio
3 - 4.3ghz all core
2 - 4.2ghz all core
1 - completely stock 3900x


Yeah yeah yeah! Once I get a new job we shall see. I really do not need one but, why the hell not. My Facebook Posts would post before I even click Post with one of those!

And yes stock around 7000. Not 7300. You had to OC all core to get that, and I got the same results. He said stock he got 7384 with a slight offset.
 
I'm not sure if this will add anything of value to the conversation or not... but here goes.

I have a 3900X on an ASUS ROG STRIX B450-i motherboard with 32 Gb of 3000 MHz Corsair Vengeance DDR4. My use case is light gaming, video transcoding, CAD, and finite element analysis. This system is in an NCASE M1 small form factor case, so the best I can do is a 240mm radiator. I'm trying to decide if water cooling in my use case is worth it. I have been running my CPU stock without PBO.

I originally had this in my system with a Noctua NH-U9S 125mm tall air cooler with dual 90mm fans. During a handbrake video transcode (1:55 long movie, 2 pass, mp4), my CPU temperatures were up around 87C. I ordered a 240mm AIO cooler (DeepCool Castle 240EX) to help with the high temps. While I was waiting on the AIO to arrive, I discovered that I could restrict the CPU temperature in the BIOS to lower it from the default 95C limit.

Here are my results when transcoding the same video.

Air Cooler (95C limit) - Transcode time = 36 min, 17 sec, Average CPU frequency = 4,000 MHz, CPU Temperature = 87C
Air Cooler (78C limit) - Transcode time = 36 min, 48 sec, Average CPU frequency = 3,925 MHz, CPU Temperature = 78C
AIO (95C limit) - Transcode time = 36 min, 11 sec, Average CPU frequency = 4,030 MHz, CPU Temperature = 75C

I'll probably go back to the air cooler for a few reasons... (1) it's slightly quieter, (2) I had to remove one of my 9.5mm thick 2.5-in. HDD due to the radiator, (3) I won't have to worry about losing water from the air cooler over a prolonged period of time like I would an AIO, and (4) my transcode was only about 1.7% faster (although it was 3C cooler) than when I restricted the CPU temperature to 78C on air.

3900x handbrake temperatures.png
 
Last edited:
I'm not sure if this will add anything of value to the conversation or not... but here goes.

I have a 3900X on an ASUS ROG STRIX B450-i motherboard with 32 Gb of 3000 MHz Corsair Vengeance DDR4. My use case is light gaming, video transcoding, CAD, and finite element analysis. This system is in an NCASE M1 small form factor case, so the best I can do is a 240mm radiator. I'm trying to decide if water cooling in my use case is worth it. I have been running my CPU stock without PBO.

I originally had this in my system with a Noctua NH-U9S 125mm tall air cooler with dual 90mm fans. During a handbrake video transcode (1:55 long movie, 2 pass, mp4), my CPU temperatures were up around 87C. I ordered a 240mm AIO cooler (DeepCool Castle 240EX) to help with the high temps. While I was waiting on the AIO to arrive, I discovered that I could restrict the CPU temperature in the BIOS to lower it from the default 95C limit.

Here are my results when transcoding the same video.

Air Cooler (95C limit) - Transcode time = 36 min, 17 sec, Average CPU frequency = 4,000 MHz, CPU Temperature = 87C
Air Cooler (78C limit) - Transcode time = 36 min, 48 sec, Average CPU frequency = 3,925 MHz, CPU Temperature = 78C
AIO (95C limit) - Transcode time = 36 min, 11 sec, Average CPU frequency = 4,030 MHz, CPU Temperature = 75C

I'll probably go back to the air cooler for a few reasons... (1) it's slightly quieter, (2) I had to remove one of my 9.5mm HDD due to the radiator, (3) I won't have to worry about losing water from the air cooler over a prolonged period of time, and (4) my transcode was only about 1.7% faster (although it was 3C cooler) than when I restricted the CPU temperature to 78C on air.

View attachment 214812

Very nice results with comparisons, real world ones and not just benchmarks.

This is for everyone else and the OP.

I mean it all depends at the end of the day what you want. Slightly limit the CPU from hitting its potential but never seeing any real world difference or being able to get the most out of the money you spent and get the Benchmark win! But if going for the first one then why does one need a $750 top of the line CPU? Why not get a 3900X or a 3800X if you are going to settle for just below maximum performance? Why do you need 4 more cores if 12 will do? I mean if you are trying to build a top end, in your budget, system then why hobble the CPU from getting the Max out of it? It is like buying a track car and never taking it to a track! Why bother, just to say you have it? Do what you want, if you want air cooling and never to encounter the best the CPU can do then do it! If you want slightly better then AIO! If you want to get the most out of it then Custom Loop! Personally I went for the middle but some day in either the next gen or a ThreadRipper I will do Custom. Right now I see my boards VRM as limiting me from getting everything out of this CPU. My board is speced to 192A EDC and 144A TDC instead of the higher ones you see in this thread and if I OC to 4.3 all core I hit that EDC at 100%. So there is more there. I see it when I try 4.35 and the benches are lower in R20. Sure if I tweak it in the BIOS more I can squeeze a few more out but in Auto mode I see max single core speeds of as high as 4.5 regularly with maybe a blip to 4.55. Mine does not seem to be limited by heat so much as power delivery.
 
Yeah yeah yeah! Once I get a new job we shall see. I really do not need one but, why the hell not. My Facebook Posts would post before I even click Post with one of those!

And yes stock around 7000. Not 7300. You had to OC all core to get that, and I got the same results. He said stock he got 7384 with a slight offset.

Pretty sure he used auto-oc extending amperage. On the stock cooler and setting it to 25 MHz I start above 7200 (100 percent fan), this progressively goes down for obvious reason (as the wraith prism slowly but surely gets overwhelmed with heat). This is a 10 minute run. Ended up slightly above 7100.

*I also get my second d15 tomorrow, I'm not sure I'm going to buy the water cooling mount for my 3900x to test with, but I'll think about it*
 
Last edited:
Pretty sure he used auto-oc extending amperage. On the stock cooler and setting it to 25 MHz I start above 7200 (100 percent fan), this progressively goes down for obvious reason (as the wraith prism slowly but surely gets overwhelmed with heat). This is a 10 minute run. Ended up slightly above 7100.

*I also get my second d15 tomorrow, I'm not sure I'm going to buy the water cooling mount for my 3900x to test with, but I'll think about it*

That's not stock either way.
 
So, somewhere around 4033 MHz and 80 degrees with auto overclocking enabled.

For kicks, I also did some manual OC'ing. I got a successful run of CB20 at 4.35 all core on 1.3v. Almost hit 10,000. I tried 4.375 at the same voltage but it failed. Obviously, not a stability test.
Just a head up, 1.3V on my 3900X works just fine for multi-core but it drop performance on single core. Single core needs aroudn 1.38 to work properly. I think the internal chip voltage boost is limited to how much current it can boost. Try Stock multi-core and stock single-core and note this as your reference. After you do your voltage offset, gotta run both test again to ensure the performance for both doesn't drop. Single core is a pain because the bench is a bit longer wait. 1.38V is the sweet spot for my 3900X. See my third response for more details.

Show me a screen of that score and some temps because I for one do not believe you on an air cooled 3900x stock except a voltage offset. Sure I can get above that if I clock mine all core to 4,3Ghz with my power setting to MB instead of Processor. But i have yet to see a 3900x in it stock form score much over 7000 every single run. I have to have my chip start off nice and cool and stay there to get 7000 consistently. When I ran the D15 on it and Prime 95 for more than 20 minutes it would peg at 95 degrees, that is with just the default clicking OK for the test and not changing any options. I tried many coolers and many cases and a few AIO's and Air coolers and I have never seen an air cooler not hit 95 degrees under full load for any amount of time. Hell overclocked to 4.3 it struggled to keep it below 80 without a load. Now it sits nice and pretty and barely goes above 70 with 100% load if it ever does. I want to see pretty screenshots that you got a 7384 on R20 with Ryzen Master screen cap showing me what it up because you must have got the best chip and I got the worst.
See the attachment. Cinebench R20 does not run long enough to have the heat build up. Prime95 and OCCT is not stress enough. My current tackling around this is to run two instances of handbrake to final settle temperature. Also I forgot to highlight that the cpu fan only ramp up to ~1000rpm and I am only using 1 fan atm! Nice and quiet! This is with -.1V offset. I think you are doing something wrong!

With just amperage tuning that's pretty good

This is voltage tuning. What ever current the cpu need does not matter unless you are close to your limited. The voltage also limit the transistor switching so going to slow will hurt your performance. My surveying of Ryzen Master, it seems to have internal votlage boost (essentiall pulling more current to boost up voltage) when running higher frequency. The -.1V offset essential is manufacture processing tolerance and you are just tightening it down. (E.G. 1.48V@9A is require for 4.6ghz and it is locked in at that setting and you cannot change it. You can still give it 1.34 but it will pull 9.9A. 1.48V*9A [email protected]. Of course you will get a loss from pulling more current because of I^2. Essentially the processing is constant power.) This is the best that I know off at the moment. It is not really an undervoltage but a tightening of manufacturer tolerance: allow us to output less heat!
 

Attachments

  • Sample1.png
    Sample1.png
    1.4 MB · Views: 0
  • Sample 2_edit.png
    Sample 2_edit.png
    1.8 MB · Views: 0
Last edited:
Here are my data for single core bench with .1V offset!
 

Attachments

  • SingleCore.png
    SingleCore.png
    681.3 KB · Views: 0
  • SingleCore CPU Frequency.png
    SingleCore CPU Frequency.png
    1,012.6 KB · Views: 0
  • Singlecore Final Scores.png
    Singlecore Final Scores.png
    1.2 MB · Views: 0
Just a head up, 1.3V on my 3900X works just fine for multi-core but it drop performance on single core. Single core needs aroudn 1.38 to work properly. I think the internal chip voltage boost is limited to how much current it can boost. Try Stock multi-core and stock single-core and note this as your reference. After you do your voltage offset, gotta run both test again to ensure the performance for both doesn't drop. Single core is a pain because the bench is a bit longer wait. 1.38V is the sweet spot for my 3900X. See my third response for more details.


See the attachment. Cinebench R20 does not run long enough to have the heat build up. Prime95 and OCCT is not stress enough. My current tackling around this is to run two instances of handbrake to final settle temperature. Also I forgot to highlight that the cpu fan only ramp up to ~1000rpm and I am only using 1 fan atm! Nice and quiet! This is with -.1V offset. I think you are doing something wrong!



This is voltage tuning. What ever current the cpu need does not matter unless you are close to your limited. The voltage also limit the transistor switching so going to slow will hurt your performance. My surveying of Ryzen Master, it seems to have internal votlage boost (essentiall pulling more current to boost up voltage) when running higher frequency. The -.1V offset essential is manufacture processing tolerance and you are just tightening it down. (E.G. 1.48V@9A is require for 4.6ghz and it is locked in at that setting and you cannot change it. You can still give it 1.34 but it will pull 9.9A. 1.48V*9A [email protected]. Of course you will get a loss from pulling more current because of I^2. Essentially the processing is constant power.) This is the best that I know off at the moment. It is not really an undervoltage but a tightening of manufacturer tolerance: allow us to output less heat!

You're gonna degrade your chip. You're way over the maximum average for FIT voltage for high current loads of around 1.325v.
 
Here is my 200W 3900X with no voltage off set.
 

Attachments

  • 200W 3900X  no voltage offset.png
    200W 3900X no voltage offset.png
    1.2 MB · Views: 0
Just a head up, 1.3V on my 3900X works just fine for multi-core but it drop performance on single core. Single core needs aroudn 1.38 to work properly. I think the internal chip voltage boost is limited to how much current it can boost. Try Stock multi-core and stock single-core and note this as your reference. After you do your voltage offset, gotta run both test again to ensure the performance for both doesn't drop. Single core is a pain because the bench is a bit longer wait. 1.38V is the sweet spot for my 3900X. See my third response for more details.


See the attachment. Cinebench R20 does not run long enough to have the heat build up. Prime95 and OCCT is not stress enough. My current tackling around this is to run two instances of handbrake to final settle temperature. Also I forgot to highlight that the cpu fan only ramp up to ~1000rpm and I am only using 1 fan atm! Nice and quiet! This is with -.1V offset. I think you are doing something wrong!



This is voltage tuning. What ever current the cpu need does not matter unless you are close to your limited. The voltage also limit the transistor switching so going to slow will hurt your performance. My surveying of Ryzen Master, it seems to have internal votlage boost (essentiall pulling more current to boost up voltage) when running higher frequency. The -.1V offset essential is manufacture processing tolerance and you are just tightening it down. (E.G. 1.48V@9A is require for 4.6ghz and it is locked in at that setting and you cannot change it. You can still give it 1.34 but it will pull 9.9A. 1.48V*9A [email protected]. Of course you will get a loss from pulling more current because of I^2. Essentially the processing is constant power.) This is the best that I know off at the moment. It is not really an undervoltage but a tightening of manufacturer tolerance: allow us to output less heat!

Interesting. You board can push almost 100 Amps more on the TDC than mine will. I think maybe I am limited by more board. Guess I should have bumped up a model or 2 on that side of things.
 
Just a head up, 1.3V on my 3900X works just fine for multi-core but it drop performance on single core. Single core needs aroudn 1.38 to work properly. I think the internal chip voltage boost is limited to how much current it can boost. Try Stock multi-core and stock single-core and note this as your reference. After you do your voltage offset, gotta run both test again to ensure the performance for both doesn't drop. Single core is a pain because the bench is a bit longer wait. 1.38V is the sweet spot for my 3900X. See my third response for more details.


See the attachment. Cinebench R20 does not run long enough to have the heat build up. Prime95 and OCCT is not stress enough. My current tackling around this is to run two instances of handbrake to final settle temperature. Also I forgot to highlight that the cpu fan only ramp up to ~1000rpm and I am only using 1 fan atm! Nice and quiet! This is with -.1V offset. I think you are doing something wrong!



This is voltage tuning. What ever current the cpu need does not matter unless you are close to your limited. The voltage also limit the transistor switching so going to slow will hurt your performance. My surveying of Ryzen Master, it seems to have internal votlage boost (essentiall pulling more current to boost up voltage) when running higher frequency. The -.1V offset essential is manufacture processing tolerance and you are just tightening it down. (E.G. 1.48V@9A is require for 4.6ghz and it is locked in at that setting and you cannot change it. You can still give it 1.34 but it will pull 9.9A. 1.48V*9A [email protected]. Of course you will get a loss from pulling more current because of I^2. Essentially the processing is constant power.) This is the best that I know off at the moment. It is not really an undervoltage but a tightening of manufacturer tolerance: allow us to output less heat!

I ran CB20 multi with -.1 offset and did get a bump by about 138 points over stock up to I think 9038 from 8900 stock. However, single core CB suffered and the score was 498. There is definitely a tradeoff. The temperature savings in multicore give you higher boosts, but single core is already fine at full voltage on temps and suffers when you restrain it. It all really depends on your workload. Obviously, the best scenario is extreme cooling at normal voltage. That idea has been beaten like a dead horse in this thread.
 
Obviously, the best scenario is extreme cooling at normal voltage.
I strongly disagree with statement. I strongly believe the majority of folks can tightening their Vcore offset to approximate -.1V and reduce excess heat without losing any performance of PB2 or PBO on single or multi-core. As long as folks are informed and happy then that is all that matter. I am happy with my setup with 3900X with Noctua D15.

You're gonna degrade your chip. You're way over the maximum average for FIT voltage for high current loads of around 1.325v.
I am not sure what you are referring to. The chip stock voltage is 1.485V. Material degrade long the curve with temperature. If temperature are not there on the damage curve, then there is no damage. AMD rated these CPU with tolerance up 105C (not accounting buffered tolerance)? I was running at 80C max. You are talking abut 25C between package and hotspot on cpu. That is a very large difference there so I am not even worry about degrading. If it was 5-10C then I would worry. Not sure if you have the RX 5700 has package and hotspot difference of 15C. If that is the worse case, I still have an extra 10C buffer. After my voltage offset, I only get 76C max now.
 
Last edited:
I strongly disagree with statement. I strongly believe the majority of folks can tightening their Vcore offset to approximate -.1V and reduce excess heat without losing any performance of PB2 or PBO on single or multi-core. As long as folks are informed and happy then that is all that matter. I am happy with my setup with 3900X with Noctua D15.


I am not sure what you are referring to. The chip stock voltage is 1.485V. Material degrade long the curve with temperature. If temperature are not there on the damage curve, then there is no damage. AMD rated these CPU with tolerance up 105C (not accounting buffered tolerance)? I was running at 80C max. You are talking abut 25C between package and hotspot on cpu. That is a very large difference there so I am not even worry about degrading. If it was 5-10C then I would worry. Not sure if you have the RX 5700 has package and hotspot difference of 15C. If that is the worse case, I still have an extra 10C buffer. After my voltage offset, I only get 76C max now.


The stock voltage is not 1.485 - that's only for short single core loads. The voltage, at stock, changes based on the load. All core is way lower than that and I believe anything over 1.325V fixed all core on zen2 will start to cause degradation. More reading here, but if you are running a fixed voltage of 1.38 all core you're going to burn up your chip.

https://www.overclock.net/forum/10-amd-cpus/1728758-strictly-technical-matisse-not-really.html
 
I strongly disagree with statement. I strongly believe the majority of folks can tightening their Vcore offset to approximate -.1V and reduce excess heat without losing any performance of PB2 or PBO on single or multi-core. As long as folks are informed and happy then that is all that matter. I am happy with my setup with 3900X with Noctua D15.

I would love to see an LN2 test with voltage left at stock vs. -0.1 offset. Surely, someone out there has done this.
 
The stock voltage is not 1.485 - that's only for short single core loads. The voltage, at stock, changes based on the load. All core is way lower than that and I believe anything over 1.325V fixed all core on zen2 will start to cause degradation. More reading here, but if you are running a fixed voltage of 1.38 all core you're going to burn up your chip.

https://www.overclock.net/forum/10-amd-cpus/1728758-strictly-technical-matisse-not-really.html

He doesn't use a fixed voltage. He is using pbo. If he was loading up avx workload his voltages would have been lower (at least mine were when I tested it).
 
He doesn't use a fixed voltage. He is using pbo. If he was loading up avx workload his voltages would have been lower (at least mine were when I tested it).

He doesn't even know how the voltages work, nor you apparently. I warned him because he was clearly going over max fit voltage. Read the link bobzdar posted.
 
He doesn't even know how the voltages work, nor you apparently. I warned him because he was clearly going over max fit voltage. Read the link bobzdar posted.

I not only read the link but understand the link. Additionally I don't use PBO, or anything besides stock, in my short stint, I have decided it's not worth it.
 
Last edited:
I not only read the link but understand the link. Additionally I don't use PBO, or anything besides stock, in my short stint, I have decided it's not worth it.
Thank you for the input! I but share my knowledge through reading, understanding, and also "personally witnessing/experiencing it" and feedback to your own reading knowledge. Keep getting fed with old information or not to further own verification is more like brainwashing. Well I guess it is more socially acceptable nowaday because that is the majority of the crowd. "It is better to be collective wrong than individually right!" If I am wrong then I am happy to admit it! But this is not the case!

Here is an excerpt from the link:
"A word regarding the "Auto Overclocking" feature...

The new "auto overclocking" feature, which is advertised with up to 200MHz frequency increase, in reality does close to nothing, at least on higher-end SKUs.
The lower-end SKUs, such as Ryzen 5 3600 definitely get some advantage however, the higher-end SKUs such as the 3700X and 3900X can be completely maxed out simply by increasing or removing the power limit (through PBO).
These SKUs are already clocked so high that further frequency improvements theoretically made possible by the "Auto OC" feature are disallowed by the silicon fitness monitoring feature (FIT), due to the required voltage for higher frequencies being too high. For instance,
on the 3700X test sample the best core of the CPU raises its frequency by 25MHz when the highest 200MHz option is selected. The rest of the seven cores remain at their default frequency, which varies between 4.35GHz and 4.375GHz.
Meanwhile the 3900X, which has stock max boost of 4.6GHz, there are no gains what so ever. In fact, none of the cores within this CPU even reach the advertised 4.6GHz. The two best cores reach 4.575GHz, while the ten other cores reach 4.325 - 4.4GHz peak. The variation between the different cores even on the same piece of a silicon appears to be huge, which would indicate that the process isn't very mature at this point. Even AMD themselves state in their slides that the frequencies are limited by the voltage they can safely feed to the CPU."

Breakdown:
"Meanwhile the 3900X, which has stock max boost of 4.6GHz, there are no gains what so ever." Fallacy for me! I do get additional performance when PBO enable with all core loaded vs setting PBO offset 0MHz.

"The two best cores reach 4.575GHz, while the ten other cores reach 4.325 - 4.4GHz peak." Partially agreed but my two best cores reach 4.61 per AMD advertised.

"Even AMD themselves state in their slides that the frequencies are limited by the voltage they can safely feed to the CPU." - Yes this true but you have to deduct what they are saying. The law of physic does not break! Higher voltage = higher temperature. You cannot have degradation without temperature.


If I leave my VCORE Bios on Auto, I boot up with 1.485V. Tried on two of the same board, same result. When I bench single core and multicore on a fixed voltage 1.485V vs Auto, it makes no difference in temperature or power reading. When I am on fixed voltage and in Ryzen Master, VCore is still jumping around for various single/multicore, this conclude to me that Mobo voltage is not necessary to change the voltage because the CPU is regulating its own the proper voltage/frequency (though its has it own limit). 1.385 should be the sweet spot because it helps minimize internal CPU regulation thus lower power usage. Also when I am in offset I am also in auto, it is the board cpu voltage with minus .1V. Telling me leaving it on auto is best makes no sense because is always +.1V higher voltage!

Set various fixed voltages and look at Ryzen Master for your cpu VCORE when running at full load. Of course measure against ST and MT. What is your conclusion then? It is okay to figure stuff out by yourself instead letting other feed you information!
 
I've found auto-OC does net you something. Whether its worth doing is up to you. In my tests, it means a few points in different benchmarks here and there based on what I've seen.
 
I'm actually gonna take a flyer on this: https://www.amazon.com/CORSAIR-Radi...R+iCUE+H150i+RGB+PRO+XT&qid=1578339511&sr=8-1

New, looks to be the CoolIT version (as opposed to Asetek version) of the H150i Pro called the

iCUE H150i RGB PRO XT Liquid CPU Cooler $159.99

View attachment 213628View attachment 213634 View attachment 213636 View attachment 213638

The ML fans on this go up to 2400 rpm as opposed to 1600 rpm on the outgoing model. I only happened upon this on Amazon when I was looking at the current H150i Pro which is out of stock until February, but they have renewed versions for $100.00. I almost pulled the trigger on one of those, but glad I found this. It doesn't release until January 13th though, so my 3950x will either wait until then or limp along on the H100i Platinum SE which is also a CoolIt product. I don't really care for how you mount the CoolIt version, but I like the size of the coldplate better than the Asetek models. Here's the link to Corsair's product page: https://www.corsair.com/us/en/Categ...uid-CPU-Cooler/p/CW-9060045-WW#tab-tech-specs

So, this thing is still MIA in spite of me purchasing it on January 6, 2020. The 240mm and 280mm versions are out there, but apparently, they had some manufacturing issues on the 360 that they worked through, but the corona virus has shut everything down. The Lunar New Year was extended by the Government until 2/9/20 according to shockism, who appears to be a Corsair rep, based on his reddit post
Reddit post about H150i Pro XT availability . I'm wondering even if production started this week how long it will take for these to make it across the ocean and into our hands. I'm not doing a lot with the computer at this time, but I'm itching to get a bigger cooling solution in place. So, I may reconsider the Thermaltake Floe 360 or the EVGA CLC 360 with a fan swap.
 
As an Amazon Associate, HardForum may earn from qualifying purchases.
So, this thing is still MIA in spite of me purchasing it on January 6, 2020. The 240mm and 280mm versions are out there, but apparently, they had some manufacturing issues on the 360 that they worked through, but the corona virus has shut everything down. The Lunar New Year was extended by the Government until 2/9/20 according to shockism, who appears to be a Corsair rep, based on his reddit post
Reddit post about H150i Pro XT availability . I'm wondering even if production started this week how long it will take for these to make it across the ocean and into our hands. I'm not doing a lot with the computer at this time, but I'm itching to get a bigger cooling solution in place. So, I may reconsider the Thermaltake Floe 360 or the EVGA CLC 360 with a fan swap.

Yep... I ordered it as soon as you recommended it and my order was delayed till March. Cancelled it and ordered the Kraken :D
 
...
I mean it all depends at the end of the day what you want. Slightly limit the CPU from hitting its potential but never seeing any real world difference or being able to get the most out of the money you spent and get the Benchmark win! But if going for the first one then why does one need a $750 top of the line CPU? Why not get a 3900X or a 3800X if you are going to settle for just below maximum performance? Why do you need 4 more cores if 12 will do? ...

Why does one need a $750 top of the line CPU? because not all workloads are CPU-bound; some are quite parallelizable and want cores as much or more than a couple hundred MHz. Database work is a good example.

A couple months ago, I toyed with the notion of upgrading my 2700X DBMS development box, and had I gone ahead with it, I would have gone with a 3950X and an NH-D15 or LGM. I need the added 8 cores way more than I need a few percent more single thread speed, and quietness is very important to me. But even leaving that last point aside, I will claim that there are real workloads that need a 3900X or 3950X and yet don't really care about leaving 200 MHz on the table. (and besides, it's hard to max out the CPU with a real life sized database load, there's lots of I/O to be done as well.)
 
Why does one need a $750 top of the line CPU? because not all workloads are CPU-bound; some are quite parallelizable and want cores as much or more than a couple hundred MHz. Database work is a good example.

A couple months ago, I toyed with the notion of upgrading my 2700X DBMS development box, and had I gone ahead with it, I would have gone with a 3950X and an NH-D15 or LGM. I need the added 8 cores way more than I need a few percent more single thread speed, and quietness is very important to me. But even leaving that last point aside, I will claim that there are real workloads that need a 3900X or 3950X and yet don't really care about leaving 200 MHz on the table. (and besides, it's hard to max out the CPU with a real life sized database load, there's lots of I/O to be done as well.)

Yeah, I mean to the point of the thread, going 3900X with $250-$300 worth of exotic cooling will provide significantly lower performance than even a throttled 3950X with a wraith. So spend your money accordingly, if you have $800 to spend, a 3950X with a $50 Noctua cooler will crush a 3900X with a custom loop. Stretching another $50 for a 240mm AIO might provide 1-2%, where the 3950 and air cooler will provide 20% over a 3900X and custom water. Stretching another $150 for custom water might be better spent on the next tier video card. If you have no budget limit, well then go with whatever your case of choice will fit and your tolerance for maintenance/pita to build will handle. High end air will cost you less than 5% in performance if you push the system to it's limits, for some it's not worth the time, money or maintenance pita for that.

I could barely fit a 240mm AIO, so that was the limit for me - but it actually fit better than high end air due to the height of my case. I did not want a huge case and custom water would have been a major pita as there's nowhere to put the res. To swap cases, build custom water etc. would have added $400 to my build and gotten me pretty much zero performance. Even for 5% performance it wouldn't have been worth it, I'd have spent that on a new video card. Will you get the absolute most out of a 3950X with air cooling? No. Will it crush any other desktop processor, even on air? Yes, and it's not even close.
 
I fully understand about keeping the clocks stable and even the Ryzen 5 3600 likes water cooling as it's auto overclockimg to 4.4Ghz now .. but it's more the base clock as how high will it hold .



Now this is how it games

 
Last edited:
Yeah, I mean to the point of the thread, going 3900X with $250-$300 worth of exotic cooling will provide significantly lower performance than even a throttled 3950X with a wraith. So spend your money accordingly, if you have $800 to spend, a 3950X with a $50 Noctua cooler will crush a 3900X with a custom loop. Stretching another $50 for a 240mm AIO might provide 1-2%, where the 3950 and air cooler will provide 20% over a 3900X and custom water. Stretching another $150 for custom water might be better spent on the next tier video card. If you have no budget limit, well then go with whatever your case of choice will fit and your tolerance for maintenance/pita to build will handle. High end air will cost you less than 5% in performance if you push the system to it's limits, for some it's not worth the time, money or maintenance pita for that.

I could barely fit a 240mm AIO, so that was the limit for me - but it actually fit better than high end air due to the height of my case. I did not want a huge case and custom water would have been a major pita as there's nowhere to put the res. To swap cases, build custom water etc. would have added $400 to my build and gotten me pretty much zero performance. Even for 5% performance it wouldn't have been worth it, I'd have spent that on a new video card. Will you get the absolute most out of a 3950X with air cooling? No. Will it crush any other desktop processor, even on air? Yes, and it's not even close.
Very true, for what I spending on watercooling parts, I could sell my 1080ti and buy a 2080ti.
 
I am air cooled on my 3950x and don't see the real need for water. I have run it so far the longest using handbrake to convert 1080/730 HD tv shows to HEVC 265 for over 12 hours straight. For the setting I have it set is 4300 with voltage set to 1.624 and LLC at 3. As long as I don't see over 80c for the CPU I am satisfied. Right now I am doing 10 2hour 1080 TV movies to 1080 mp4 in HEVC 265 running two instances of HandBreak. Here is a current screen shot of that right now and a pic I took when I moved it to my nite stand last week. It is a new system in a case I first used back in 2009 with my I7 920.


.1583281256882.pngPuter.jpg
 
I am air cooled on my 3950x and don't see the real need for water. I have run it so far the longest using handbrake to convert 1080/730 HD tv shows to HEVC 265 for over 12 hours straight. For the setting I have it set is 4300 with voltage set to 1.624

1.6+ volts!!!???
 
Thanks for pointing that out. I transposed the numbers it should of been 1.264. Today I changed the orientation of my HSF and changed the TIM to a liquid metal and it did drop a few degrees. I ran a test to see how it did with a normal workload for me (2 instances of handbrake converting 1080 tv shows mpg2 to x265).
A shot of it at work and one of my system I took while changing the TIM and cooler orientation.My system
 

Attachments

  • Annotation 2020-03-10 124713.png
    Annotation 2020-03-10 124713.png
    959.9 KB · Views: 0
Back
Top