Discussion Ryzen 3000 series benchmark thread ** Open **

[IEC]

I heard you guys like benchmarks.

Despite ASUS having an annoyingly bad 5007 BIOS based on AGESA 1.0.0.2 for the X470 Strix-F, I've managed to figure out a few things that might be helpful:

Specs: 3900X / X470 Strix-F / Samsung B-die 3600 CL15 kit

Gleanings from a few hours of testing:
1) The CPU is showing the same pre-AGESA 1.0.0.3 AB low boost clocks as reviewers ran into, so don't expect proper benchmarking especially for ST/lightly threaded workloads
2) This particular board + BIOS has cold boot DDR voltage set at 1.2V. So you guessed it, I boot into safe 2133 by default, need to set 3200 + voltage, reboot, then actually set my memory speed and timings (thanks ASUS)
3) 3733 CL16 auto subtimings (atrocious subtimings, btw) is as far as I have gone so far with a 1:1 fabric at 1867 (OC past official 1600 "limit") - 3800 caused me to hang at POST. Not sure if CPU limit or buggy BIOS.
4) Undervolting my CPU by -0.100V up to -0.150V offset appears to have done very little to ST/MT clocks, suggesting whatever is broken with boost in AGESA 1.0.0.2 makes it think it is hitting the FIT silicon fitness limits even if it isn't. Bummer, now I have to wait for a BIOS with AGESA 1.0.0.3AB+ to bench this puppy so I can hit my proper boost clocks...

But for now, at least I can post a baseline AIDA64 memory and cache benchmark at 3733 1:1 IF:


67.3ns. Catch me if you can.

"Sweet spot" confirmed. I bet I can shave off some more if I can tweak those atrocious subtimings...

 

[DrMrLordX]

@BTRY B 529th FA BN nice find. Pity they had to disable a chiplet to get there. Still ought to be an amazing gaming machine at those speeds though.

 

[DrMrLordX]

But for now, at least I can post a baseline AIDA64 memory and cache benchmark at 3733 1:1 IF:

Have you tried higher clocks while maintaining 1:1 IF? Might require more vSoC.

 

[IEC]

Have you tried higher clocks while maintaining 1:1 IF? Might require more vSoC.

Hangs at post requiring clear CMOS, AUTO vSoC. Not going to push the limits too hard while this BIOS is buggy. Going to wait until 1.0.0.3AB or better to push the limits a bit more.

 

[DarthKyrie]

I heard you guys like benchmarks.

Despite ASUS having an annoyingly bad 5007 BIOS based on AGESA 1.0.0.2 for the X470 Strix-F, I've managed to figure out a few things that might be helpful:

Specs: 3900X / X470 Strix-F / Samsung B-die 3600 CL15 kit

Gleanings from a few hours of testing:
1) The CPU is showing the same pre-AGESA 1.0.0.3 AB low boost clocks as reviewers ran into, so don't expect proper benchmarking especially for ST/lightly threaded workloads
2) This particular board + BIOS has cold boot DDR voltage set at 1.2V. So you guessed it, I boot into safe 2133 by default, need to set 3200 + voltage, reboot, then actually set my memory speed and timings (thanks ASUS)
3) 3733 CL16 auto subtimings (atrocious subtimings, btw) is as far as I have gone so far with a 1:1 fabric at 1867 (OC past official 1600 "limit") - 3800 caused me to hang at POST. Not sure if CPU limit or buggy BIOS.
4) Undervolting my CPU by -0.100V offset appears to have done very little to ST/MT clocks, suggesting whatever is broken with boost in AGESA 1.0.0.2 likely causes it to hit the FIT silicon fitness limits. Bummer, now I have to wait for a BIOS with AGESA 1.0.0.3AB+ to bench this puppy so I can hit my proper boost clocks...

But for now, at least I can post a baseline AIDA64 memory and cache benchmark at 3733 1:1 IF:
View attachment 8319

"Sweet spot" confirmed. I bet I can shave off some more if I can tweak those atrocious subtimings...

It looks like I was right about the MBs below X570 not getting the new AGESA yet and will clock like crap until they do receive a new BIOS like I was telling VirtualLarry yesterday. Hopefully, those will be released next week. Out of curiosity does that MB support PCIe 4.0 with that BIOS?

 

[IEC]

Let me reboot and check (and maybe tweak some memory subtimings)

P.S. Cinebench R20* MT over 7000 while full tilt package power <145W. Niiiiiice.

Edit: Might be missing it, but not seeing any Gen4 PCIe option. I remember seeing this on my X370 Gigabyte K7 board so YMMV. Also, manually setting subtimings failed, but I probably just goofed something up so I'll bother with it later. 3733 CL16 will have to do for now.

 

[DarthKyrie]

Have you tried higher clocks while maintaining 1:1 IF? Might require more vSoC.

The BIOS in the 300 and 400 series MBs are all borked as far max performance on the Ryzen 3000 series, it's like what was seen in the reviews that had to be rerun with the new AGESA.

 

[DarthKyrie]

Let me reboot and check (and maybe tweak some memory subtimings)

P.S. Cinebench RT MT over 7000 while full tilt package power <145W. Niiiiiice.

Now that is sweet. Can Intel do this at anything below 250W?

 

[Markfw]

Let me reboot and check (and maybe tweak some memory subtimings)

P.S. Cinebench RT MT over 7000 while full tilt package power <145W. Niiiiiice.

Same as mine I did not enumerate, but I had to manually set ram voltage to 1.375, speed to 3600 and clocks to 15,15,15 (I let the rest default) It defaulted to 26 !

Edit this is on 4233 RAM ! and even after that, it went back up to 16 cas.

 

[SK10H]

Since my Ubuntu VM isn't working at the moment, I test running Win10 1903 x64 in a VM, connected using PIA with AES256, RSA4096 etc as it's single thread bound. Download max out at 10MB/s using utorrent on 2700x default boast to 4.35.
With 3600x, it seems to max out at 13.7MB/s.

 

[BTRY B 529th FA BN]

@DrMrLordX Yeah, that raises my interest in the 3950X. A 3900X doing 4.7GHz on all cores is awesome, even if it's just one chiplet. I never use HT with a gaming rig so if 8 cores can do something like that whether it being the 3800X, or the 3950X with one chiplet then I'll be grinning all the way.

 

[ClockHound]

BTRY B 529th FA BN: Just the beginning. Hope the 3800X uses a chiplet binned as high.

3950X: multi-core workhorse by day and hard clocking gaming monster by night. Binning 'n grinning.

 

[TheGiant]

I am the author. I cannot hit that 4800~4900 range without ASUS' MultiCore Enhancement, which as far as I'm aware goes outside of Intel's spec to overclock the CPU.

well I dont think it overclocks the CPU, it just removes the artificial limit of 95W, which noone follows- some of the fanboys

noone exactly follows the limit (except WS and server bioses)
TDP became a tool of lie
the absolute difference in power with MCE conditions vs 95W with 9900K is higher because that CPU simply can clock that high and its non overclocked all core turbo is 4,7GHz
the power to achieve it (like 170W) is the price for 14nm tech

so you see not ryzen 2700X, not 8700K doesnt hold to its TDP
by the way its not nice, but official- Intels TDP is just base clock all core power, not turbo
with new CPUs like mine 3900X, AMD overclocks them by design with pretty intelligent algorithm
simply said- there is no overclocking with modern CPUs, yours will run based on your cooling and temps- simply what your cooling can handle
there is ofc another question- what of the CPUs that dont reach their TDP (7700K, 8600K ...) - should we overclock them to reach that Watts?

 

[DrMrLordX]

@TheGiant

What do you mean 2700x doesn't hold to its TDP? Your graph shows system power consumption. Not package power or anything like that.

https://www.anandtech.com/show/12625/amd-second-generation-ryzen-7-2700x-2700-ryzen-5-2600x-2600/8

 

[TheGiant]

@TheGiant

What do you mean 2700x doesn't hold to its TDP? Your graph shows system power consumption. Not package power or anything like that.

https://www.anandtech.com/show/12625/amd-second-generation-ryzen-7-2700x-2700-ryzen-5-2600x-2600/8

exactly
if you look at that chart above, you see 8600K(TDP 95W) and R2600 (TDP 65W) consume about the same power
R2600 holds by MB settings to its TDP 65W
30W above is 2600X (95W) and 95W restricted 9900K, which looks logical 65+30=95- the same components, pretty much the same chipset power (x470 and z390)
and then you see 8700K (95W spec ofc intel base clock, everyone seems ignoring that, reality =crap) and 2700X (105W), but 30W above, which you cant explain by VRM loss or power efficiency of the same power supply

MB just allows it to consume more, same like the Intel chips

I cant find the article now, but there was 105W restricted 2700X (it was like 3,8 GHz all core turbo) and higher limited (the same like techpot article)

the same is now with R3 line, where AMD allows the CPU to consume more
it is just limited now with more parameters- current, wattage, temps, VRM temps etc
AMD does it very clever with R3 lineup- my 3900X I just leave on default, no point on finding the best clocks- setting to constant it limit low thread performance and with higher power
I have to find a good way to undervolt it as my i5 6600K

that is how I see it

9900K just CAN clock so high and with 14nm consume so much that it is much more visible than with other chips

 

[DrMrLordX]

and then you see 8700K (95W spec ofc intel base clock, everyone seems ignoring that, reality =crap) and 2700X (105W), but 30W above, which you cant explain by VRM loss or power efficiency of the same power supply

And yet AnandTech showed the 2700x staying within it's TDP.

Jury's still out on whether Matisse will stay within its TDP since Asus (and possibly others) appears to be playing around with modified microcode - see The Stilt's comments on "the thing".

 

[TheGiant]

And yet AnandTech showed the 2700x staying within it's TDP.

Jury's still out on whether Matisse will stay within its TDP since Asus (and possibly others) appears to be playing around with modified microcode - see The Stilt's comments on "the thing".

well my 3900X while doing handbrake x265 wall power=249W
the same with my 6600K 4,4GHz underwolted wall power 109W

 

[dlerious]

Science Studio tested 3900X on A320. It booted and ran Aida64 but couldn't reach boost clocks.

 

[Zucker2k]

well I dont think it overclocks the CPU, it just removes the artificial limit of 95W, which noone follows- some of the fanboys

noone exactly follows the limit (except WS and server bioses)
TDP became a tool of lie
the absolute difference in power with MCE conditions vs 95W with 9900K is higher because that CPU simply can clock that high and its non overclocked all core turbo is 4,7GHz
the power to achieve it (like 170W) is the price for 14nm tech

so you see not ryzen 2700X, not 8700K doesnt hold to its TDP
by the way its not nice, but official- Intels TDP is just base clock all core power, not turbo
with new CPUs like mine 3900X, AMD overclocks them by design with pretty intelligent algorithm
simply said- there is no overclocking with modern CPUs, yours will run based on your cooling and temps- simply what your cooling can handle
there is ofc another question- what of the CPUs that dont reach their TDP (7700K, 8600K ...) - should we overclock them to reach that Watts?

Lots of people are in denial over Ryzen+ power consumption....

So it turns out the 95W 9900k was not only the best performing desktop chip until a week ago, but even more efficient than most give it credit for....

 

[DrMrLordX]

Lots of people are in denial over Ryzen+ power consumption....

Like AnandTech reviewers?

@dlerious

At least it works in A320, which is surprising. @Shivansps should be thrilled.

 

[PotatoWithEarsOnSide]

Lots of people are in denial over Ryzen+ power consumption....

So it turns out the 95W 9900k was not only the best performing desktop chip until a week ago, but even more efficient than most give it credit for....

At stock, the 9900K is very efficient. 4.2GHz all-core within 95w.
However, getting the darn thing to run within spec is a bloody nightmare since it involves turning off all of the spec overrides.

 

[Fanatical Meat]

Like AnandTech reviewers?

@dlerious

At least it works in A320, which is surprising. @Shivansps should be thrilled.

Yay! I can buy a $500-$750 chip and save a few bucks running it on a $40 motherboard.
Yay!

#sarcasm

 

[DrMrLordX]

Yay! I can buy a $500-$750 chip and save a few bucks running it on a $40 motherboard.
Yay!

And, as predicted, it will throttle like a sonofagun. I had actually expected some throttling on B350 as well, but right now clocks across the board seem lower than I expected, so it's difficult to actually get Matisse chips up to a clockspeed where it can torture 6-phase VRM setups.

 

[TheGiant]

And, as predicted, it will throttle like a sonofagun. I had actually expected some throttling on B350 as well, but right now clocks across the board seem lower than I expected, so it's difficult to actually get Matisse chips up to a clockspeed where it can torture 6-phase VRM setups.

I did too
the fact that an average b350 board (asrock pro4 b350) can handle 3700X and most likely 3900X is a massive OverDelivery by AMD

 

[Edgemeal]

DAWBENCH TESTED

http://www.scanproaudio.info/2019/0...00x-dawbench-tested-3-is-it-the-magic-number/