Discussion Ryzen 3000 series benchmark thread Open

DrMrLordX · Jul 15, 2019

Det0x said:
Testing with a AMD graphic card could prove or disprove the rumors i mentioned above.

If that's the case, you want the fastest possible AMD card, which would be Radeon VII on custom water (to isolate the CPU).

daperl · Jul 15, 2019

maddie · Jul 15, 2019

daperl said:

As a resident of the tropics, this is very valuable info for me.

It seems that 7nm was potentially so good that AMD could increase yield a lot by raising voltages and still shame Intel. What happens in 6 months as the process further improves?

rbk123 · Jul 15, 2019

Bodes well for future laptops.

lopri · Jul 15, 2019

My 3700X is massively underperforming out of the box. It is scoring 10~20% lower than the results produced by AT. I am on AGESA 1.0.0.3 AB, and running 2 x 16GB sticks @3200MHz/14. Temps are fine according to Ryzen Master (30C~65C) and I have fiddled with vcore a bit to see if it makes a difference. Lowering vcore improved the situation somewhat (2~3% maybe), but it is nowhere near where it should be. I have a stinking suspicion that AMD sent cherry-picked specimens to reviewers.

daperl · Jul 15, 2019

Here's a reddit response to the youtube video I referenced above:

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

lopri · Jul 15, 2019

Sure that is a valid theory assuming motherboards are not overvolting the CPUs already.

amrnuke · Jul 15, 2019

lopri said:
My 3700X is massively underperforming out of the box. It is scoring 10~20% lower than the results produced by AT. I am on AGESA 1.0.0.3 AB, and running 2 x 16GB sticks @3200MHz/14. Temps are fine according to Ryzen Master (30C~65C) and I have fiddled with vcore a bit to see if it makes a difference. Lowering vcore improved the situation somewhat (2~3% maybe), but it is nowhere near where it should be. I have a stinking suspicion that AMD sent cherry-picked specimens to reviewers.

You aren't alone either. I think they're relying on "up to" xxx boost on the advertising. The thing is, you might be able to claim it's just a thermal regulation issue and the included cooler is just for basic use. But in that case, shouldn't we be able to be hitting advertised boost clocks consistently with AIO? Because a large number of people aren't. I wonder what's going on.

TheGiant · Jul 15, 2019

is it possible to change I_max ? I mean safely

Topweasel · Jul 15, 2019

amrnuke said:
You aren't alone either. I think they're relying on "up to" xxx boost on the advertising. The thing is, you might be able to claim it's just a thermal regulation issue and the included cooler is just for basic use. But in that case, shouldn't we be able to be hitting advertised boost clocks consistently with AIO? Because a large number of people aren't. I wonder what's going on.

The only advertised boost is the single core one which is nearly impossible to hit as it requires pretty much completely idle activity on other cores. To get what AT is posting you need to have PBO turned on. Even then it's a lottery. That rated clocks is based on PBO and that sets some limits on how it can achieve that, meaning some chips won't hit that upper limit.

moinmoin · Jul 15, 2019

daperl said:
Here's a reddit response to the youtube video I referenced above:

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

tl/dr: If the CPU is not running at a fixed frequency vCore doesn't actually affect voltage used by each core but only the overall power budget available, so too low vCore will start throttling the CPU and as such lower the performance.

But since PBO (currently) isn't working either too high vCore just burns power with no gain either. As such finding the lowest possible vCore before losing performance is still a valid optimization strategy.

Topweasel said:
That rated clocks is based on PBO

That's news to me. You sure that AMD picked a frequency that can only be reached by voiding the warranty?

scannall · Jul 15, 2019

The benchmarks that are out are interesting and all. But I think I will wait a few months while they get the bios sorted out to really start paying attention. And also wait a few months to build a 3800X machine. So thanks to all you early adopters! Your aggravation is greatly appreciated.

Hitman928 · Jul 16, 2019

Hardware Unboxed tested 9900K vs 3900x across 36 games. The 3900x was 6% slower at stock and 5% slower when both were overclocked:

Stock:

9900K OC vs 3900x PBO

lopri · Jul 16, 2019

amrnuke said:
You aren't alone either. I think they're relying on "up to" xxx boost on the advertising. The thing is, you might be able to claim it's just a thermal regulation issue and the included cooler is just for basic use. But in that case, shouldn't we be able to be hitting advertised boost clocks consistently with AIO? Because a large number of people aren't. I wonder what's going on.

Apparently these CPUs run on Goldilocks principle! You need to give it just a right amount of voltage under the right temperature. lol. AMD should be careful about this advertisement tactic because I see a potential class-action lawsuit brewing.

Topweasel said:
The only advertised boost is the single core one which is nearly impossible to hit as it requires pretty much completely idle activity on other cores. To get what AT is posting you need to have PBO turned on. Even then it's a lottery. That rated clocks is based on PBO and that sets some limits on how it can achieve that, meaning some chips won't hit that upper limit.

Did AT state that they had PBO turned on? I am still trying to understand what PBO does exactly. Does it raise the default power cap?

Atari2600 · Jul 16, 2019

All the review site data is of little use if there is a noticeable disparity in actual performance of purchased CPUs from reviewed CPUs.

If AMD have been supplying cherry-picked CPUs to review, shame on them.

If not, then they need to start getting a message out there about where the current state of AGESA is at and a roadmap (not necessarily with dates) of where will go in the future.

lopri · Jul 16, 2019

Even with Ryzen 2, Intel CPUs still rule StarCraft 2.

dlerious · Jul 16, 2019

lopri said:
Apparently these CPUs run on Goldilocks principle! You need to give it just a right amount of voltage under the right temperature. lol. AMD should be careful about this advertisement tactic because I see a potential class-action lawsuit brewing.

Did AT state that they had PBO turned on? I am still trying to understand what PBO does exactly. Does it raise the default power cap?

Explanation from Steve at Gamers Nexus.

daperl · Jul 16, 2019

Whoops:

Asterox · Jul 16, 2019

Even on very cheep A320 motherboard, Ryzen 5 3600 all core turbo is 4-4.1ghz.In my country this Asus motherboard is hm converted only 60$.

lopri · Jul 16, 2019

dlerious said:
Explanation from Steve at Gamers Nexus.

Thx. I have switched from stock HSF to Mugen 5 and a 120x38 mm fan. That helped significantly.

lopri · Jul 16, 2019

Tom's has a 3800X review out.

https://www.tomshardware.com/reviews/amd-ryzen-7-3800x-review,6226.html

TheGiant · Jul 17, 2019

lopri said:
Tom's has a 3800X review out.

https://www.tomshardware.com/reviews/amd-ryzen-7-3800x-review,6226.html

good review 3800X=9900K
3800X does make zero sense to me when 3700X exist

one interesting thing about power
in the 100+-15W range, the 7nm ryzen 8C doesnt bring any serious advantage (except cinepeen)
in the 9900K 5GHz 170W range, r3 is behind, but the power curve is really steep
I would like to see a review with 4/4,2,4,4,4,7,5GHz 9900K range and power corresponding to it

in short, what actually 7nm r3 line brought to us in perf/power

in GPU world the rx7 line just is performance comparable to Nvidia lines, but also power (nvidia 12nm, amd 7nm)
In the CPU world is a little better, but only a little when considering total system power, which is ofc dilluted by chipset and other components

r3600 (price) and r3900 (performance, price) is excellent, the 8C parts not so much

apple will soon laugh at all the x86 crap

B-Riz · Jul 17, 2019

lopri said:
Thx. I have switched from stock HSF to Mugen 5 and a 120x38 mm fan. That helped significantly.

I get good boost (3900X) on a Gammaxx 400, less than $20 on sale, with the B350 test system.

Cinebench R20 will hit 4.5 single thread, but only if pinned to Core 0 in task manager, if not, Windows bounces it from core to core and it only boosts ~4.3.

Atari2600 · Jul 17, 2019

B-Riz said:
Cinebench R20 will hit 4.5 single thread, but only if pinned to Core 0 in task manager, if not, Windows bounces it from core to core and it only boosts ~4.3.

Ah, windaz scheduler strikes again.

Abwx · Jul 17, 2019

Povray numbers a THG are complete BS with the 9900K dsiplayed as 35% faster than the 3700X while Computerbase put the difference at 0% :

https://www.computerbase.de/2019-07/amd-ryzen-3000-test/3/#diagramm-test-pov-ray-multi

aHR0cDovL21lZGlhLmJlc3RvZm1pY3JvLmNvbS85LzQvODQ1MzIwL29yaWdpbmFsL2ltYWdlMDE0LnBuZw==

The ST perf in the same test should had ringed a bell.

Discussion Ryzen 3000 series benchmark thread ** Open **

Lifer

Member

Diamond Member

Senior member

Elite Member

Member

Elite Member

Golden Member

Senior member

Diamond Member

Diamond Member

Golden Member

Diamond Member

Elite Member

Golden Member

Elite Member

Golden Member

Member

Golden Member

Elite Member

Elite Member

Senior member

Golden Member

Golden Member

Lifer

Discussion Ryzen 3000 series benchmark thread Open