Matisse (Ryzen 3000) overclocking/undervolting

[heky]

Anyone want to guess what Silicon Lottery will post for golden chips? Suppose to show up at 10PM CST.

AMD Matisse

Binned Ryzen Matisse CPUs available from Silicon Lottery!

siliconlottery.com

4.5Ghz allcore oc on 3800x @ 1.325v is my guess...

 

[Wavetrex]

Suggest installing the cooler 1st and run some stuff you deem necessary to compare performance VS stock cooler and then swap the memory and repeat the tests: a bit more hassle but it gives us better idea of how much changing from stock cooler to a very good air cooler with everything else the same can do.

Okay, will do.
Cooler will arrive first it seems ( tomorrow ). Memory later in the week.

 

[biffzinker]

Interesting:

There is absolutely a difference in silicon quality between the 3700X and 3800X, to get that out of the way. On average, the 3800X will hit about 100MHz or more higher all core.

https://www.reddit.com/r/Amd/comments/cg0f0z

 

[Vario]

Actually the coolers between a 939 and an FX are different even though they may look/appear the same. Changes happened along the way and I'll say not to the good, the coolers got "cheaper" as time passed with the last relying on the heatpipes mostly.

The old 939's had a base with a bit of substance to them compared to the later ones, out of all from 939 to FX with this design they are still the best.

I would dare say they are at least just as good as the Wraith coolers too.

Yeah they had a big copper plate on them. Had one of them back in the day on top of my Opteron.

 

[Bones]

The copper plate is what makes the difference, more mass to absorb heat and move it along. Those that came later with the direct heatpipe contact are..... OK but not as good, even though they too have a plate but it does little in comparison to the 939 type.

I do not like direct heatpipe contact coolers, they are just a cheaper form of cooler.

 

[TheLostSwede]

Anyone want to guess what Silicon Lottery will post for golden chips? Suppose to show up at 10PM CST.

AMD Matisse

Binned Ryzen Matisse CPUs available from Silicon Lottery!

siliconlottery.com

That must be a joke no? That looks very mediocre at best and not worth paying extra for.
I guess some people snagged some cheap CPU's though...

 

[Wavetrex]

That website/company is a scam and anyone with half a functioning brain knows it.

Unfortunately there are plenty of people with less than half of a brain functioning... which is why they are still in business.

 

[Bones]

That website/company is a scam and anyone with half a functioning brain knows it.

Unfortunately there are plenty of people with less than half of a brain functioning... which is why they are still in business.

Truth!
No way I'd pay the extra they are charging for these.

 

[EarthDog]

You can thank amd for the results. SL did what they do...there's just zero headroom.

 

[Wavetrex]

You can thank amd for the results.

Actually, I think you can thank TSMC for not being able to hit AMD's expectations.

I really believe AMD designed this core to boost up to 5 Ghz on 7nm, but failed to do so for reasons we will probably never know.
It's possible that "A0" silicon targeted those clocks, but proved to be unstable and causing errors, no matter what, and they altered some circuits to make the cores run well and error free and had to lower the clocks to make the changes to the current "B0" silicon work.

This is obviously a wild guess, but it might explain why there's no headroom whatsoever on these chips.

Also, it is possible there will be a refresh down the line that DOES boost higher ( by 200-300 Mhz, potentially hitting that magic 5Ghz on single thread ). All core... no way, but single thread definitely possible if they find a fix to what's preventing the chips to go there. The process itself should definitely be capable !

 

[EarthDog]

I think this has more to do with the uarch than it does with the process. Zen, and Zen+ behave similarly. No headroom. It's AMD getting the most out of these CPUs out of the box. The process has little to do with it.

So, SL is just doing what they are doing... hardly a scam, but clearly not worth it to this subset of ENTHUSIASTS... whereas those who may not be as well versed thinks it is worth the time and effort. But this is hardly a scam and a very myopic look at the whole thing IMO.

 

[Vario]

That website/company is a scam and anyone with half a functioning brain knows it.

Unfortunately there are plenty of people with less than half of a brain functioning... which is why they are still in business.

Yeah, the CoffeeLake binning didn't seem legit at all based on what my sample does and what people were reporting on Overclock.net. Also the pricing for 5GHz 5.1, 5.2 etc CPUs is ridiculous given most CoffeeLake can achieve that easily with enough voltage and a solid board.

Furthermore, the binning they do is very simplified and not very exact. The voltages they use for 5.1+ are way too high and that tells me their top bins are nothing special. A high bin 8700K for example would be 5.1 @ 1.25, not 5.1 @ 1.4v. 5.1 @ 1.4V is average and quite mundane for an 8700K especially with a -2 offset. Your typical retail CPU can do that with a midrange Z370.

A lot of their bins rely on AVX offset which, when subtracted from the "binned" frequency shows they are nothing special. I am also very skeptical of the binning statistics.

 

[EarthDog]

What were people reporting?

What do you mean by the binning they do is simplified and not 'exact'? They needed the voltage they for stability in how they test.

Many users run an AVX offset in the first place as with typical CPUs at 5GHz+ with AVX, most coolers can't keep up anyway.

I strongly disagree with the assertion that SL is a scam. It may not be the best deal, but for those who do not want to gamble, it can be worth the premium. It may not be worth it to you, but you're also more of an enthusiast than most and take the chance and time to manually overclock your own.

 

[Wavetrex]

Bring out the big guns ! (Well, average size guns, but with good muzzle velocity )

Will redo testing after this is installed on the 3700X

 

[Vario]

Nice, perfect. Now you will probably be in the 60s.

 

[Wavetrex]

Nice, perfect. Now you will probably be in the 60s.

Extremely unlikely, there's over 30 degrees in this room...

They weren't kidding in the reviews that installation is a breeze.

It literally took me only 5 minutes to mount it, and another 10 to admire the craftsmanship...

But now my computer is all over the place in colors ... oh well

First observation:
I can't hear it !
The noise outside in the street is stronger than the fan at auto-speed.
Even at full speed (~2000 rpm it's barely audible inside the case)
WOW !

---- First quick round of benchmarks (Single thread) ---
Note: It's VERY hot here, I don't have AC and it's afternoon. Room thermometer shows around 30 degrees Celsius ambient ( 86 F )

The percentages are related to this post:
https://www.techpowerup.com/forums/...overclocking-undervolting.257505/post-4084338 (The "Default" 100% reference)

wPrime 32M - 29.0 (101.7%)
SuperPI 4M - 45.8 (101.5%)
Cine R10 - 10156 (100.7%)
CpuZ - 517 (100.2%)
3DMark2003 CPU T1 - 58.8 (102.8%)
CPU-M 1.6 - 20310 (101.2%)

Extra: 7zip single threaded (32MB dic): 7014 MIPS, 5227 KB/s Comp, 90546 KB/s Decomp
Also 7zip but 16 threads: 73860 MIPS, 51410 KB/s Comp, 1000594 KB/s Decomp

One observation about clocks, during all the single threaded tests it was consistently boosting between 4.3 and 4.4, with tiny spikes to 4.45 ( I have PBO+100 enabled )
Max temp during 7zip ran for 5 mins ... ~ 81 degrees.
Cinebench R20 went to 85 ( no longer hitting 90+, so that's progress). Will retry some testing when it's colder (night).

Faster? Yes. But 1-2% is nothing.
Did it worth the 95 euro cost for the cooler ? If you look at performance... no.
But... I spend 8-10 hours per day working on my computer ( job ), so for NOISE... it's a definite YES. Even at 100% it has a pleasant "wind like" sound, not a harsh "turbine" sound. I LOVE IT.

 

[kapone32]

Extremely unlikely, there's over 30 degrees in this room...

They weren't kidding in the reviews that installation is a breeze.

It literally took me only 5 minutes to mount it, and another 10 to admire the craftsmanship...

But now my computer is all over the place in colors ... oh well

First observation:
I can't hear it !
The noise outside in the street is stronger than the fan at auto-speed.
Even at full speed (~2000 rpm it's barely audible inside the case)
WOW !

---- First quick round of benchmarks (Single thread) ---
Note: It's VERY hot here, I don't have AC and it's afternoon. Room thermometer shows around 30 degrees Celsius ambient ( 86 F )

The percentages are related to this post:
https://www.techpowerup.com/forums/...overclocking-undervolting.257505/post-4084338 (The "Default" 100% reference)

wPrime 32M - 29.0 (101.7%)
SuperPI 4M - 45.8 (101.5%)
Cine R10 - 10156 (100.7%)
CpuZ - 517 (100.2%)
3DMark2003 CPU T1 - 58.8 (102.8%)
CPU-M 1.6 - 20310 (101.2%)

Extra: 7zip single threaded (32MB dic): 7014 MIPS, 5227 KB/s Comp, 90546 KB/s Decomp
Also 7zip but 16 threads: 73860 MIPS, 51410 KB/s Comp, 1000594 KB/s Decomp

One observation about clocks, during all the single threaded tests it was consistently boosting between 4.3 and 4.4, with tiny spikes to 4.45 ( I have PBO+100 enabled )
Max temp during 7zip ran for 5 mins ... ~ 81 degrees.
Cinebench R20 went to 85 ( no longer hitting 90+, so that's progress). Will retry some testing when it's colder (night).

Faster? Yes. But 1-2% is nothing.
Did it worth the 95 euro cost for the cooler ? If you look at performance... no.
But... I spend 8-10 hours per day working on my computer ( job ), so for NOISE... it's a definite YES. Even at 100% it has a pleasant "wind like" sound, not a harsh "turbine" sound. I LOVE IT.

Excellent, based on my own testing these Noctua coolers are indeed golden. I agree totally with the noise argument. I have one on my TR 4 build and it keeps the 1900x in the 30s while idle and a max of 67 C when gaming.

 

[HTC]

Okay, will do.
Cooler will arrive first it seems ( tomorrow ). Memory later in the week.

Don't forget to also test that "high tech extra wind source"

 

[Sybaris_Caesar]

@Wavetrex how hot is there in NL? It's 33c here during the day.

 

[Wavetrex]

@Wavetrex how hot is there in NL? It's 33c here during the day.

hot.
Bad time to test new computer hardware which changes speeds based on temperature

@ 15:00

Windy as forecasted

Wind map and weather forecast

www.windy.com

It seems that Spain, south-France, Italy, Greece and Romania had it worst today.

 

[EarthDog]

Bad time to test new computer hardware which changes speeds based on temperature

Which is why most people normalize temps and go from there. So, essentially, you pick a temp to start with, say 22C, and whatever your ambient is, add/subtract that from the results. So if my room was 26C at the time of testing, I would subtract 4C from all my results.

Martin/Skinee labs used this method years ago when they were around. Or...... you can use a delta which temps shouldn't matter anyway.

 

[Wavetrex]

Yes but the problem with Zen 2 core... it boosts depending on ABSOLUTE temperature, not the relative over ambient.

I'm not a rich youtuber that can control the temp with LN2 *cough*Steve , or run AC at 100% to keep ambient at 20 at all times *cough*Jay ...
And most of us aren't either.

So we deal with real-life situations, when we can't "normalize" anything... If the room has 33 degrees like @Khonjel 's... then that's his starting temperature and his potential Zen 2 cores will run slower than for someone that burns 1500W of electricity per hour to cool the room to 20.

 

[Ferrum Master]

Yes but the problem with Zen 2 core... it boosts depending on ABSOLUTE temperature, not the relative over ambient.

I'm not a rich youtuber that can control the temp with LN2 *cough*Steve , or run AC at 100% to keep ambient at 20 at all times *cough*Jay ...
And most of us aren't either.

So we deal with real-life situations, when we can't "normalize" anything... If the room has 33 degrees like @Khonjel 's... then that's his starting temperature and his potential Zen 2 cores will run slower than for someone that burns 1500W of electricity per hour to cool the room to 20.

Get a ticket to Alaska or Siberia? Case solved.

It is +18C outside for me now too thou lol

 

[EarthDog]

That is correct. I was talking testing/reviews, which cannot account for such variables.

 

[Wavetrex]

New GN video about this exact subject...

New toy to play with... (probably a waste of money, just like for those that bought 3600X and 3800X ... ^.^)