5800XT Overclocking and Undervolting Thread

[OKC Yeakey Trentadue]

So I got my 5800xt and wanted to see what I could do with it. ITX case with a crappy cooler (NH-L12s missing pieces at the moment) so focus is on efficiency. Started benching with Cinebench 20.

First run was bone stock, JDEC ram. 5882 Multi and 623 Single. Temps were 90* and 71*. Ouch. That's not going to do.
Next I turned on 65w eco as well as XMP. 5346 and 623 points. Temps for both at 71*. Better, but I want to maintain performance.
Then tried above with x43 multiplier. 5627 and 551 with. Temps 69* and 51*. Definitely not wanting to lose single core performance so on to curve optimizer.
Tried stock with with -25 offset. 5924 and 630. Better than stock but temps were still 90* in multi. Going back to Eco...
This time with -20 offset and +200 MHZ. 5415 and 628. Temps both 73*. Best temp/performance combo so far, but need some more work...

 

[freeagentt]

You need a better cooler, one that will manhandle that CPU. Doesn't have to be expensive, but it should be a little more substantial.

 

[pendragon1]

all within spec. if you want cooler, then that ^^^, get a TR ps120/pa120

 

[OKC Yeakey Trentadue]

You need a better cooler, one that will manhandle that CPU. Doesn't have to be expensive, but it should be a little more substantial.

From the wall, i draw 200w then it drops to 185w when thermally limited at STOCK setting. Modified the eco settings and now get similar to stock performance at 135w, again from the wall so massive percentage savings of cpu power if i could isolate that. And at 71* I still have thermal headroom.

Gonna focus on ram not to see how it compares to the 5700g in gaming power usage.
https://hardforum.com/threads/my-3050-6gb-sff-build-series-s-replacement.2037620/

 

[jlbenedict]

So the 65w eco mode is basically a macro..
These are the typical values for AM4 CPU's:

45w CPU - PPT = 60W, TDC = 45A, EDC = 65A
65w CPU - PPT = 87W, TDC = 60A, EDC = 90A
95w CPU - PPT = 128W, TDC = 90A, EDC = 125A
105W CPU - PPT = 142W, TDC = 95A, EDC = 140A

Try using customized values such as the ones that would be for a 95W CPU..along with the best offset value possible ( better binning might get you -25 or even-30 )
This should drop temperatures but not neuter the performance as much as the 65w values.

 

[cyberblogger]

I've been thinking of getting this 5800xt to replace the ryzen 5 2600 that is on the B550 board I just got. But I calculated the heatsink has to have a mass to dissapate 280 watts with 100 watts continuous. So you will need a heat sink with about twice the mass (Weight) as the NH-L12s.

What I picked out for that brand was a NH-D15 G2

 

[OKC Yeakey Trentadue]

So the 65w eco mode is basically a macro..
These are the typical values for AM4 CPU's:

45w CPU - PPT = 60W, TDC = 45A, EDC = 65A
65w CPU - PPT = 87W, TDC = 60A, EDC = 90A
95w CPU - PPT = 128W, TDC = 90A, EDC = 125A
105W CPU - PPT = 142W, TDC = 95A, EDC = 140A

Try using customized values such as the ones that would be for a 95W CPU..along with the best offset value possible ( better binning might get you -25 or even-30 )
This should drop temperatures but not neuter the performance as much as the 65w values.

I did and i got much better results, again using my crappy round aluminium thermaltake.

110w, 85a, 120a with -25 offset (-30 crashes) and +200 mhz.

Multi cb20 of 6020 at 84* and 162w from wall
Single of 642!

So now better MT, ST, power/heat than stock on a small cooler.

Xmp settings with 1600 fclock.

 

[OKC Yeakey Trentadue]

Anyone know a canned benchmark that shows differences in cache? VMmark is all i could find.

 

[freeagentt]

You are limiting its performance by running neutered power settings. These CPUs thrive with extended power limits, boost override, and the curve.

If you cant cool it at stock, then you have some issues.

 

[OKC Yeakey Trentadue]

You are limiting its performance by running neutered power settings. These CPUs thrive with extended power limits, boost override, and the curve.

If you cant cool it at stock, then you have some issues.

What are you talking about? Show me this cpu pushing out way more power than stock by dumping more power in it. I've already achieved over 6000 on CB 20 using around an 85w ppt according to jlbendict's chart.

And the CPU was a replacement for my light gaming Series S build, https://hardforum.com/threads/my-3050-6gb-sff-build-series-s-replacement.2037620/, where low power was one of the focuses.

I also got over 12k cpu score on timespy without getting into memory adjustments.

 

[OKC Yeakey Trentadue]

Started playing with the memory some and got over 12,500 on Timespy CPU. Little more tweaking and I might hit 13,000. Very pleased given the power draw numbers. 56ns memory latency is pretty cool too.

Memory is capable of higher speeds (4400 mhz with my 5700g) but this cpu needed 1.15v soc to make 1966mhz Fclock stable. I didn't expect it to go as high as Renoir but happy with cas 16 at 3800mhz with these voltages.

 

[freeagentt]

GFlops my man. You can try it for yourself with Linpack Xtreme. I run the 10GB load.

 

[OKC Yeakey Trentadue]

GFlops my man. You can try it for yourself with Linpack Xtreme. I run the 10GB loload.

What does that test represent and what is considered good for DDR4?