Discussion Ryzen 3000 series benchmark thread ** Open **

[maddie]

Ah, windaz scheduler strikes again.

Don't be too quick to blame Windows. It seems that we are seeing a very large variation in silicon characteristics, probably a function of 7nm at present development stage. It's not the switching cores causing the lower clocks but that cores are several hundred MHz apart in top speed.

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.

Can you pin it to the different cores and see what happens? Potentially could be quite informative.

 

[B-Riz]

Don't be too quick to blame Windows. It seems that we are seeing a very large variation in silicon characteristics, probably a function of 7nm at present development stage. It's not the switching cores causing the lower clocks but that cores are several hundred MHz apart in top speed.


Can you pin it to the different cores and see what happens? Potentially could be quite informative.

You can, but, I am at the limits of the B350 right now, being that, I don't think I will hit 4.6 on the board, *maybe* after a few BIOS updates on it.

Cinebench is also odd, it will "override" what you set in task manager depending on when you set it. I had to set to CPU0 *after* the single core benchmark started.

 

[PotatoWithEarsOnSide]

Per CCX overclocking appears to be the current best strategy with Zen 2. Some reports of higher clocking CCXs even within the same chiplet.

 

[Arkaign]

I can't remember a major release being simultaneously so impressive yet janky lol. I built an office system with 3600 (vanllla, X seems wasted) and a leftover B350 board (Asus Prime 350 Plus, updated to latest bios using my OG 1700). Worked ok for about a day, and while installing Office 365 it started freezing up. Turned it off via the PSU switch when nothing else would respond, and the CPU is just dead as a doornail now. Stock volts, stock cooler from my 2700X, EVGA 500W PSU, Evo 960 250GB SSD, 8GB single stick DDR3000, and an old 750ti 2GB.

Everything else works, fired right back up with the old 1700. Wth.

 

[Markfw]

I can't remember a major release being simultaneously so impressive yet janky lol. I built an office system with 3600 (vanllla, X seems wasted) and a leftover B350 board (Asus Prime 350 Plus, updated to latest bios using my OG 1700). Worked ok for about a day, and while installing Office 365 it started freezing up. Turned it off via the PSU switch when nothing else would respond, and the CPU is just dead as a doornail now. Stock volts, stock cooler from my 2700X, EVGA 500W PSU, Evo 960 250GB SSD, 8GB single stick DDR3000, and an old 750ti 2GB.

Everything else works, fired right back up with the old 1700. Wth.

Reflash I don't think the CPU is dead, I think its a bios issue. If the reflash does not work, wait for a newer bios. or get a really cheap x570 ?? or maybe even an older x370 or x470.

 

[thigobr]

That's not good... Either bad sample or maybe firmware set some of the voltages outside the safe zone?

 

[VirtualLarry]

I can't remember a major release being simultaneously so impressive yet janky lol. I built an office system with 3600 (vanllla, X seems wasted) and a leftover B350 board (Asus Prime 350 Plus, updated to latest bios using my OG 1700). Worked ok for about a day, and while installing Office 365 it started freezing up. Turned it off via the PSU switch when nothing else would respond, and the CPU is just dead as a doornail now. Stock volts, stock cooler from my 2700X, EVGA 500W PSU, Evo 960 250GB SSD, 8GB single stick DDR3000, and an old 750ti 2GB.

Everything else works, fired right back up with the old 1700. Wth.

That's interesting. There was a post on either the Gigabyte user forum, or Reddit (was reading both those few days), where someone had flashed an older board, I think a B350, and installed a R5 3600, and it never booted either one. Both were reported "dead" by vendor upon RMA.

Something's really not quite right, with some of these "early" 3rd-Gen BIOSes. Whether the issue is with the AGESA, or is board-specific, I don't know. Mine's (3600) has been cruising along, just flashed Gigabyte B450 AORUS PRO WIFI ATX BIOS F41b, AGESA 1.0.0.3AB, last night, still running good, got my SoC SATA ports back, haven't tried NVMe socket, but temps are still crazy, but my heatsink vapor chamber is likely broken, so I need to get a new cooler ASAP. HWMonitor shows my CPU temp went to 99C at max.

 

[Arkaign]

Reflash I don't think the CPU is dead, I think its a bios issue. If the reflash does not work, wait for a newer bios. or get a really cheap x570 ?? or maybe even an older x370 or x470.

Thanks! The only other board I have on hand is my 470 Taichi / 2700x build. They're dragging their feet on a Zen2 bios, so I'll hold out for a bit. I have a work pc with 470 Aorus Gaming 5, but don't want to necessarily bring it home and tear it down, I put one of the huge PITA hsfs on it that you have to take the board out to properly access.

 

[Markfw]

Thanks! The only other board I have on hand is my 470 Taichi / 2700x build. They're dragging their feet on a Zen2 bios, so I'll hold out for a bit. I have a work pc with 470 Aorus Gaming 5, but don't want to necessarily bring it home and tear it down, I put one of the huge PITA hsfs on it that you have to take the board out to properly access.

I have a good (well seems pretty good) bios for the x470 Taichi. Flash it as far as you can one step at a time (like the instructions say) and when you get to 3.43, PM me your email address, and I will send you 3.46, it is agesa 1.0.0.3ab and works well.

BTW, this should all be in the 3000 builders thread, not the benchmark thread, lets do future updates there please.

 

[positivedoppler]

Why is TH only benching on 1920 x 1080?? Who on earth buys a 8 core cpu and a 2080 TI to game on 1080p??

 

[amg377]

Hardcore competitive gamers and single PC live game streamers come to mind...easily chew through gpu/cpu usage without a dedicated encoder rig.

Why is TH only benching on 1920 x 1080?? Who on earth buys a 8 core cpu and a 2080 TI to game on 1080p??

 

[IEC]

While I wait for improved BIOSes... testing per CCX OC. Worst CCX's worst core does 4325 MHz at 1.35V. Best CCX's worst core does 4450. Result is 7634 CB for Cinebench R20:

 

[IEC]

I seem to recall this being a thing at some point:

Drawcall benchmark with instancing disabled, 29 FPS with 3900X:

 

[lopri]

TPU has their 3600 review up:

https://www.techpowerup.com/review/amd-ryzen-5-3600/

Interesting among the results:

3600 maintains its max frequency like a rock throughout different loads.

 

[moinmoin]

TPU has their 3600 review up:

https://www.techpowerup.com/review/amd-ryzen-5-3600/

Interesting among the results:

3600 maintains its max frequency like a rock throughout different loads.

What's surprising is that TPU didn't manage to surpass 4.125GHz in manual overclocking, so stock behavior is (currently) actually the best performance one can get from 3600. That it never goes above the defined 4.2GHz when not manually overclocked is fully according to the spec. Though somebody needs to test manually overclocking every CCX individually, that may improve on 4.125GHz at least for one CCX.

 

[DrMrLordX]

What's surprising is that TPU didn't manage to surpass 4.125GHz in manual overclocking, so stock behavior is (currently) actually the best performance one can get from 3600.

If you look at some of the actual benchmark results, you'll see that the 4.125GHz static OC outperforms the "4.2 GHz" PBO behavior. Example:

https://www.techpowerup.com/review/amd-ryzen-5-3600/6.html

Look at the MT score for CBR20 or Blender. The graph shows the 3600 holding 4.2 GHz up to 12T, but the reality is that the chip doesn't perform as well in MT scenarios despite all cores reporting clockspeeds 75MHz higher with PBO. Remember how undervolting would permit higher clocks but with lower performance? PBO (and stock) may be boosting into "false clock" territories where the voltage it's providing is low enough that the CPU operates with a lower current limit. Which is funny, since PBO is supposed to uncap current limits . . .

edit: the Corona Render results also don't fit the mold, which is weird.

 

[Zucker2k]

If you look at some of the actual benchmark results, you'll see that the 4.125GHz static OC outperforms the "4.2 GHz" PBO behavior. Example:

https://www.techpowerup.com/review/amd-ryzen-5-3600/6.html

Look at the MT score for CBR20 or Blender. The graph shows the 3600 holding 4.2 GHz up to 12T, but the reality is that the chip doesn't perform as well in MT scenarios despite all cores reporting clockspeeds 75MHz higher with PBO. Remember how undervolting would permit higher clocks but with lower performance? PBO (and stock) may be boosting into "false clock" territories where the voltage it's providing is low enough that the CPU operates with a lower current limit. Which is funny, since PBO is supposed to uncap current limits . . .

edit: the Corona Render results also don't fit the mold, which is weird.

This anomaly reminds me of the Stilt's post about how a certain mobo vendor is manipulating certain hidden voltage parameters to fool the cpu into thinking it's consuming less than it is. Wouldn't that make the cpu boost clocks a tad higher without the necessary energy to allow the boosted cores to perform optimally at those clocks, resulting in 'lower' numbers for those clocks?

 

[moinmoin]

If you look at some of the actual benchmark results, you'll see that the 4.125GHz static OC outperforms the "4.2 GHz" PBO behavior. Example:

https://www.techpowerup.com/review/amd-ryzen-5-3600/6.html

Look at the MT score for CBR20 or Blender. The graph shows the 3600 holding 4.2 GHz up to 12T, but the reality is that the chip doesn't perform as well in MT scenarios despite all cores reporting clockspeeds 75MHz higher with PBO. Remember how undervolting would permit higher clocks but with lower performance? PBO (and stock) may be boosting into "false clock" territories where the voltage it's providing is low enough that the CPU operates with a lower current limit. Which is funny, since PBO is supposed to uncap current limits . . .

edit: the Corona Render results also don't fit the mold, which is weird.

Good catches. PBO really is worthless currently. Wonder what the limit that caused the performance to degrade?

In case of undervolting it appears to be the Clock Stretcher kicking in trying to fend off the voltage starving (which also means that when trying to run Ryzen 3k at lower wattage for better efficiency it's better to use PPT to limit TDP instead undervolt):

But I can't imagine that to be the issue when PBO overclocking. Possibly the 4.2GHz boost is too short lived on many cores?

 

[DrMrLordX]

This anomaly reminds me of the Stilt's post about how a certain mobo vendor is manipulating certain hidden voltage parameters to fool the cpu into thinking it's consuming less than it is. Wouldn't that make the cpu boost clocks a tad higher without the necessary energy to allow the boosted cores to perform optimally at those clocks, resulting in 'lower' numbers for those clocks?

I think it may actually be the opposite effect. The_Stilt was saying that Asus (at least) was fiddling with power limit detection so it could feed more power (current, essentially) which would permit higher boost clocks and higher benchmark performance - assuming the CPU was kept cool enough to not hit clockspeed walls inflicted by temperature/hotspots (see der8auer's video from launch day on the 3900x to observe the temp-based clock limits).

In this case, it appears that the 3600 is trying to boost to 4.2 GHz on all cores at an abnormally-low (but still mostly stable) voltage. Either that or we're seeing the clockspeed reporting software being fooled by boost algorithms that can shift clockspeed/voltage too quickly for the software to report all the actual values.

Possibly the 4.2GHz boost is too short lived on many cores?

It may be clockspeed jitter happening so fast that monitoring software can't pick it up (see above).

 

[moinmoin]

Either that or we're seeing the clockspeed reporting software being fooled by boost algorithms that can shift clockspeed/voltage too quickly for the software to report all the actual values.

It may be clockspeed jitter happening so fast that monitoring software can't pick it up (see above).

That's definitely also an issue. We already had reports that monitoring software prevents Ryzen 3k from idling due to polling happening too often and aggressively. As I recall AMD's Robert "solved" this by advising not to run more than one monitoring software at once, and that (back then) he suggests HWinfo and Ryzen Master to read out the frequency. Computerbase then pointed out that using Ryzen Master increases power usage by 10% so skews the picture as well. Ideally there would be some way to directly read out the info straight from Zen's SCF, Ryzen Master may already do that but clearly is too heavy.

 

[IEC]

ASUS bug was most likely unintentional per The_Stilt and was acknowledged and fixed by ASUS within days of launch. May have made a difference for R5 3600, but doesn't seem to make much difference for 3900X given the broken boost clock situation.

 

[PotatoWithEarsOnSide]

AMD_Robert suggested CPU-Z and Ryzen Master
He specifically said that HWinfo was causing Ryzen to spike due to high polling rates.

 

[moinmoin]

AMD_Robert suggested CPU-Z and Ryzen Master
He specifically said that HWinfo was causing Ryzen to spike due to high polling rates.

Thanks for the correction, so I remembered those two backward.

 

[Abwx]

Not Zen 2 but still Ryzen 3000, short review of the 3200G/3400G :

https://www.computerbase.de/2019-07/amd-ryzen-3200g-3400g-test/

 

[B-Riz]

I can't remember a major release being simultaneously so impressive yet janky lol. I built an office system with 3600 (vanllla, X seems wasted) and a leftover B350 board (Asus Prime 350 Plus, updated to latest bios using my OG 1700). Worked ok for about a day, and while installing Office 365 it started freezing up. Turned it off via the PSU switch when nothing else would respond, and the CPU is just dead as a doornail now. Stock volts, stock cooler from my 2700X, EVGA 500W PSU, Evo 960 250GB SSD, 8GB single stick DDR3000, and an old 750ti 2GB.

Everything else works, fired right back up with the old 1700. Wth.

Odd, my Amazon Warehouse B350 Plus and 3900X work great. But the the *thermal* limits of the board limit the max perf of the chip.

Gammaxx 400, AS Ceramique 2, 3900X, Auto CPU Voltage, Auto PBO, Asus Prime B350-Plus, 5007 BIOS, 16GB Hynix 2933 OC to 3200, Corsair AX860i

https://imgur.com/a/sNc1ZLP

I can drop the 3600 in it later and see how it performs.