Zen3 and 4 news from techpowerup

[maxfly]

Things are continuing to look better and better for amds cpu releases!

https://www.techpowerup.com/258099/...2021-zen-3-completes-design-phase-out-in-2020

 

[EarthDog]

I hope more frequently = better! Because that's all we know at this time. Wondering what they can squeeze out of these things as on the overclocking side, it's already like getting blood from a stone.

 

[Alaric]

EPYC is going to be the killer. It's nice to see AMD in the desktop market again, but the data centers are what will be the big news, IMO.

 

[EarthDog]

EPYC is going to be the killer. It's nice to see AMD in the desktop market again, but the data centers are what will be the big news, IMO.

Agreed. Hopefully they can get some market share quickly there. I'd love to see a 2080ti type card for less.

 

[maxfly]

Epyc is friggin incredible. Stompin xeons hard from what ive read. And for so much less!

That would be nice for sure. Id settle for 2080 speeds in the $500 range at this point.

 

[bigtallanddopey]

I hope more frequently = better! Because that's all we know at this time. Wondering what they can squeeze out of these things as on the overclocking side, it's already like getting blood from a stone.

I think from AMD’s side of things overclocking is just about dead. If you build a system with a 3600 and 5700xt you get 98% of the performance out of the box.

It’s one of the reasons that good old Linus did a paid by intel overclocking video the other day. Just for intel to show that there chips overclock to much higher speeds than AMD. The question is, will intel run into the same experience when they move to 10/7nm where the clock speeds are just not as high.

 

[Alaric]

I'll take "2070 Super Speeds for $350", Alex.

 

[bignazpwns]

Well I'm excited for more zen goodness. Looks like so far we only got 2 3900x's that can pull a stable 4.5+GHz oc. The guy from Italy. I really wanna see Ryzen 3 hit all core clocks at 4.6ghz+ because that 3900x gets pretty mean at those clocks. But overall very happy with Ryzen 2 and can't wait for Ryzen 3

 

[EarthDog]

Well I'm excited for more zen goodness. Looks like so far we only got 2 3900x's that can pull a stable 4.5+GHz oc. The guy from Italy. I really wanna see Ryzen 3 hit all core clocks at 4.6ghz+ because that 3900x gets pretty mean at those clocks. But overall very happy with Ryzen 2 and can't wait for Ryzen 3

I'm curious if you would be as happy at 4.2 GHz as you are at 4.5...

In other words, if you had a normal sample versus your truly golden one.

 

[bignazpwns]

I'm curious if you would be as happy at 4.2 GHz as you are at 4.5...

In other words, if you had a normal sample versus your truly golden one.

I definitely have one of the top 3900x's on water no doubt so far only one guy has been able to run me down. But 4.2 seems really low since I seen quite a few hit 4.3-4.4. 4.4 looks like the wall. With very hit able to hit over 4.5 all core and so far only 2 that pulled 4.6+ all core outside of Ln2.

I know I'm way above average I'm terms of clock but I don't thing 4.2 is the average I think it's 4.3.


Siliacon lottery's number seem extremely low and the fact they are using an aio and only going to what 1.25v looks like there leaving some on the table since 1.35v looks like the normal for the best chips to hit 4.4 to 4.5. if they had a real cooler on that chip they could go higher but these 3900x's kinda just beat on AIO's.

But I would be fine running it at 4.2. it's not like it's a slow CPU at 4.2. it's packing tons of power. I'm terms of overclocking I would call it a dud like my 6700k I had but from what I seen those number they posted are absolutely horrible and I seen quite a few hit 4.4ghz and if all they can get is 4.2 out of there best then something is wrong.

 

[EarthDog]

SL just uses what most use. Temperature differences between 2x120 AIO and 3x120 AIO aren't really much. Getting the heat out of the CPU is the tough part it seems.

But, just saying, I'd be floored too if I had your sample, but an average one at 4.2 GHz give a bit, not as much. People are also having issues with single core hitting the boost value as well. My 3700x, across 6 boards, has never even blipped to its rated boost.

These CPUs are good performers and well priced, but there are still some teething issues on the platform as a whole.

 

[Zerileous]

Everyone keeps saying "getting the heat out of the CPU is the tough part," so when are people going to de-lid a few?

bignazpwns: I can't say I would if I were in your shoes, but ever thought of sending your CPU to someone to see what it can do under LN2?

I'm not sure why SL is using such low voltage / clocks. I guess one could assume a 4.2 at 1.25v will do 4.3 or 4.4 at 1.35v or 1.45v, but I would think they could safely bin at 1.35v.

 

[EarthDog]

delidding/sub IHS TIM isn't the issue... it's the tiny 7nm dies (and the IO controller) and all the billions of transistors built-in that causes the problem IMO.

 

[Zerileous]

Turns out it's been done

, looks like he dropped 4c on a 3600. In other words you're right

 

[bignazpwns]

Everyone keeps saying "getting the heat out of the CPU is the tough part," so when are people going to de-lid a few?

bignazpwns: I can't say I would if I were in your shoes, but ever thought of sending your CPU to someone to see what it can do under LN2?

I'm not sure why SL is using such low voltage / clocks. I guess one could assume a 4.2 at 1.25v will do 4.3 or 4.4 at 1.35v or 1.45v, but I would think they could safely bin at 1.35v.

I got LN2 lol. I did a check before i got busy with work and all i got to was 5.4 on it. Higher then i thought i knew hitting that high on water i was not going to even get close to the top guys on LN2. I'll try more when i get time but 5.4 iits pretty much all shes got 1.68v and i had it pretty stable at -90c. But it did better then i thought i thought the most i could get out of it would of been 5.2.


And i dont get why SL is binning how they are. 4.2 at 1.25 4.1 at 1.22v. Those 4.2's could hit walls at 4.3 and need around 1.38+ to hit it. Something is just strange why they are binning how they are

Voltages

1.200
1.212
1.225
1.237
1.250


I mean .050 is not a ton of voltage. There not selling high clocking 3900x's. There selling ones that can do 4.2ghz with 1.250v and under. And for $839 they are on drugs. Unless they come out and tell you that $839 bin can hit the clocks that are now what we see are the golden 3900x's 4.5ghzx 1.38v then they got nothing and its pointless to buy from them since every single 3900x i seen can pull 4.2ghz with 1.32v and under. It was a huge disappointment. And i dont think people who have 4.5ghz 3900x's are the chosen ones. They are rare but im sticking with my first feeling that they are around 1%. I wanted to see what the top clocks were. Not what someone can pull on a $50 heatsink because they refuse to do 1.35v what is what most are taking to hit the higher clocks.


But i wanna find someone who bought on and see what it does on 1.35-1.38v.

 

[EarthDog]

If I had to guess, it's because much anything around 1.35V+ under stress test loads, will hit 95C on most any cooling (2/3x120mm rads etc). My H150i can't keep up with a 3700x. The few degrees C gain (at most) for a custom loop overradded for the wattage, wouldn't allow for 100 MHz more. My stress tests are 'only' AIDA64's default (CPU/FPU/Cache) and 1.33V is between 90-94C depending on room temp. No headroom for stress testing. Sure it's fine for games, but, I have no way to test that unless I use something which will draw less power and heat and then I wonder about stability...

 

[ShrimpBrime]

delidding/sub IHS TIM isn't the issue... it's the tiny 7nm dies (and the IO controller) and all the billions of transistors built-in that causes the problem IMO.

I agree with this statement. De-lidded a couple of 2000 series chips and that tremendous heat spike still exists.
Direct die chilling seems to lower the spike somewhat along with overall max load temps (stressed) but would never be enough to make any real difference in max clock speed on ambient or even sub ambient cooling.
Even testing from -30c onto a load such as Cinebench still showed a nasty 20c spike instantly with gradual increase to above freezing temps. Did this testing with a cold plate and Peltier.
Only real gains where typical, able to run a lower voltage at X overclock opposed to ambient cooling either air or liquid produced very similar overclocks.
The above testing was done with several Ryzen Processors 1000 series and 2000 series chips and all showed the same characteristics.

De-lidding these soldered chips on ambient actually has the ability to make temps slightly worse, but the temp spike is still about the same and the gradual increase on a load is slightly quickened.

In short, Delidding any Ryzen processor is not worth the risk and time of average overclockers. I think Der8auer demonstrated all this in youtubing filmatics as if de-lidding soldered processors was cool and new. I've been doing it since AM2 days and have quite a bit of experience with all AMD platforms in regards to de-lidding soldered high wattage processors all and all.