AMD Ryzen 3000 Builders Thread

[lopri]

My 3700X has not degraded. It can't be because I have not touched vcore at all up to this point. It would be a disaster if a CPU degraded under stock vcore within a month of use.

@DrMrLordX: I was able to enter BIOS with bclk 103, by raising VDDIO under "AMD Overclocking" section in the BIOS. It failed to enter Windows. I really hate this BIOS because same values can be changed at multiple places (board overclocking section, AMD CBS, AMD Overclocking) and I have no idea which values take precedence until I test it out with non-stop reboots. So far, it looks like memory timings set in the board's overclocking section are preferred over AMD Overclocking's, but for voltages AMD Overclocking takes over. I am not entirely sure.

On a bright side I really like how smooth the system feels during web surfing. This matches my expectation from launch review benches and I am pleased about it.

 

[Kenmitch]

On a bright side I really like how smooth the system feels during web surfing. This matches my expectation from launch review benches and I am pleased about it.

lol

I surfed the web last night with all cores locked to 3.2 GHz on my 3700x and didn't notice anything unusual. It wasn't intentional as I was seeing how my 3700x scales in CB R20 MT and had worked my way down from all core 4.2 GHz to 3.2 GHz. I had a brain fart and wound up leaving it locked to 3.2 GHz for the remainder of the night.

 

[IEC]

Re: the supposed degradation

Andrei of AT says the following:

The performance degradations have nothing to do with the new power plan and idle behaviour.

You can verify this by simply installing just the new 5.0.0.0 power plan exclusively on top of the 1.07.07 chipset driver package. Doing this will result in the new idle behaviour without the performance regressions.

The performance regressions are likely caused by the RDRAND / Destiny 2 temporary fixes, or other changes in the driver package. AMD was quite clear it's a temporary release until it's fixed in the upcoming AGESA.

Furthermore, if you DO use the new driver package, the High Performance plan will have almost the same fast ramp-up behaviour as the previous Balanced plan.

You can see this demonstrated here: https://images.anandtech.com/doci/14688/Ramp-Ryzen-Perf.png

You can use this if you wish, but it doesn't really have any impact on performance.

I updated my article accordingly: https://www.anandtech.com/show/1468...s-for-ryzen-3000-more-relaved-cppc2-upscaling

I posted this because this post from the other thread just enraged me that it got upvoted as the top comment.

It's very simple to just A/B test things and get to proper conclusions. Don't just circle-jerk around whatever somebody is posting without actual proof.

Source:

https://www.reddit.com/r/Amd/comments/ck70nz

 

[DrMrLordX]

@IEC

Huh! I wonder if the same thing has been hurting my CBR20 scores? We'll see if things get better on future AGESA versions.

 

[lopri]

What is 5.0.0.0 Power Plan and where do I get it?

 

[lopri]

As an aside I now can somewhat reliably (i.e. usually after a couple of tries) boot @1900 fclk / uclk. I am not sure what made it possible but I think it was enabling "OC Mode" under "AMD Overclocking" in BIOS.

 

[lopri]

Another thing I am now confident about: Do not give excessive voltages to the IMC. (usually labeled vSOC) You may achieve a higher memory overclocking or tighter timings, but it will lower the core performance. In my testing 1.028V was just enough for 1900 uclk/fclk and 3800CL16. Every chip is different so find an optimal SOC for your voltage by downvolting step by step. (Default is 1.1V)

Note the shrinking power budget for the IMC as core load increases. The same principle, but in the other direction, applies - as the IMC load increases the core will have less power at its disposal, thereby reducing performance.

 

[DrMrLordX]

@lopri

Pretty sure that only applies when running default, though it might explain why boost performance is so bad when people OC their memory.

 

[lopri]

Yes, under default PPT and PB2. Manual overclocking it might be different.

 

[PotatoWithEarsOnSide]

Is it normal for Ryzen Master to display peak clocks over a prolonged period that do not correspond with the on-screen display I get with Afterburner/Rivatuner?
The latter seems to suggest cores almost always clocked near max, whereas the former doesn't even suggest that those clocks were hit at all.

 

[DrMrLordX]

Is it normal for Ryzen Master to display peak clocks over a prolonged period that do not correspond with the on-screen display I get with Afterburner/Rivatuner?
The latter seems to suggest cores almost always clocked near max, whereas the former doesn't even suggest that those clocks were hit at all.

Have you tried corroborating clockspeed data between Ryzen Master and HWiNFO64? They seem to line up pretty well.

 

[StinkyPinky]

So i took out my samsung b die from my 8700k rig (4x8gb 3200cl14) and tried them on my 3900x rig. At xmp profile settings at 3200c14 the latency is 75.9ns. That seems quite high? Would the 4 dimms impact that?
Edit: It's down to 72.6 with a bump to 3400/c14 and a voltage bump.

 

[lopri]

Secondary timings might need some tuning.

 

[DrMrLordX]

Secondary timings might need some tuning.

Indeed.

@StinkyPinky

XMP settings are designed primarily for Intel memory controllers unless you get stuff like FlareX. I would recommend using something like Ryzen Timing Calculator to dial in some better subtimings.

 

[EXCellR8]

I used the calculator when I had the Ripjaws kit and the values seemed to work well. Was never able to climb beyond 3200 though.

Haven't gotten around to the AGESA update on my board yet but I'll see if I can get it flashed tonight. I'd also like to add the other 16GB kit but I figure it's best to see how the system runs on the BIOS update before adding a bunch more RAM.

 

[MBrown]

For streaming, does anyone have any information on the differences between 3700x and 3900x on the gamer's side? Will there be noticeably less frame spikes and stutters with the 3900x over the 3700x? Everything I've seen so far has just been 3900x vs 9900k.

 

[Kenmitch]

For streaming, does anyone have any information on the differences between 3700x and 3900x on the gamer's side? Will there be noticeably less frame spikes and stutters with the 3900x over the 3700x? Everything I've seen so far has just been 3900x vs 9900k.

Look on Youtube. I believe hardware unboxed did one 3700x vs 9700k with 3900x vs 9900k which included games, streaming, and content creation. Watched it last night.

 

[lopri]

On paper 3900X has no disadvantage against 3700X in games. 3900X has more cores and a higher boost frequency. And since both the inter-CCX and inter-CCD communication occur through Infinity Fabric, any thread-hopping beyond a single CCX should incur an equal penalty for both 3900X and 3700X. 3900X also might benefit from bigger L3 but I am not sure about that one because the L3 might not be shared among CCX. In any case, that is another potential benefit in favor of 3900X.

 

[guachi]

Hardware Unboxed has a great video a few days ago comparing the 3900X, 9900K, and 3600 with three video cards at multiple resolutions.

He only tested four games. The 3900X was faster with the 2080, 2070S, and 5700 but the scores with a 580 were the same no matter what CPU was used. But the 3900X wasn't so much faster it really mattered. So I'd imagine the difference between the 3900X and 3700X would be even smaller no matter what GPU you used.

Basically, the 3600 is the king of gaming unless you are already at the top end, so only get a faster CPU if you need it for something other than gaming.

 

[SlowSpyder]

Upgraded my kid's PC to a Ryzen 3600 from a 1500X. Not much to report, but it is a noticeable upgrade. Only numbers I ran were the CPUZ bench, and it wasn't scientific as I updated his bios between the 1500X and the 3600 and my 2600X scores are from a completely different PC (scores rounded to nearest whole number below). Nice to see the 3600 happily up and running on his 370 chipset. Now just have to upgrade that seven year old Radeon 7970 (still chugging along strong!)

1500X
Single: 387
Multi: 2025

2600X:
Single 465
Multi: 3650

3600
Single: 505
Multi: 4100

Now my kid has a faster CPU than me, and that just won't do. Think I'll be looking at a 3700X soon.

 

[Fanatical Meat]

Upgraded my kid's PC to a Ryzen 3600 from a 1500X. Not much to report, but it is a noticeable upgrade. Only numbers I ran were the CPUZ bench, and it wasn't scientific as I updated his bios between the 1500X and the 3600 and my 2600X scores are from a completely different PC (scores rounded to nearest whole number below). Nice to see the 3600 happily up and running on his 370 chipset. Now just have to upgrade that seven year old Radeon 7970 (still chugging along strong!)

1500X
Single: 387
Multi: 2025

2600X:
Single 465
Multi: 3650

3600
Single: 505
Multi: 4100

Now my kid has a faster CPU than me, and that just won't do. Think I'll be looking at a 3700X soon.

Nice upgrade, what type of memory are you using?

 

[SlowSpyder]

Nice upgrade, what type of memory are you using?

G.Skill TridentZ RGB DDR4 3000 (2x8GB) in my PC (R5 2600X).

Corsair Veneance LED DDR4 3200 (2x8GB) in my kid's machine (R5 3600).

But now that you mention it, I just updated the bios (4x, gotta love bridge bios to bridge bios to get from 1.x to 5.x) and ran with whatever settings it took. Chances are I can up the memory speed, it is probably at 2133 or something like that currently.

 

[Fanatical Meat]

G.Skill TridentZ RGB DDR4 3000 (2x8GB) in my PC (R5 2600X).

Corsair Veneance LED DDR4 3200 (2x8GB) in my kid's machine (R5 3600).

But now that you mention it, I just updated the bios (4x, gotta love bridge bios to bridge bios to get from 1.x to 5.x) and ran with whatever settings it took. Chances are I can up the memory speed, it is probably at 2133 or something like that currently.

This is my #1 complaint with AMD, memory is too fussy and appears you have to really mess with it to get the best performance. Good work on the above too, post an update when you mess with it.

 

[VirtualLarry]

This is my #1 complaint with AMD, memory is too fussy and appears you have to really mess with it to get the best performance.

Setting XMP is messy?

BTW, with my Gigabyte B350 Gaming 3 ATX board, and a single OLOy 16GB DDR4-2667 stick, apparently it has JEDEC (is JEDEC-specced) 2667 profiles, so I didn't even have to set XMP to get max memory speed. I tweaked it up to 1.35V from 1.20V, and up to DDR4-3000 manually in BIOS, and it seems BOINC PrimeGrid stable, for over an hour or two so far.

 

[SlowSpyder]

This is my #1 complaint with AMD, memory is too fussy and appears you have to really mess with it to get the best performance. Good work on the above too, post an update when you mess with it.

I think the first gen Ryzens were pretty picky, but from what I've read a 570 board with a 3xxx processors should be fairly trouble free, supposedly.