buildzoid

did you try 16.12 driver? it can have better scaling for more than 2 GPUs. Also re-enables HBM OC.

Async BCLK + PBO so the CPU-z clock readouts are all messed up. RAM was at 6000

New 24Gb Hynix chips?

Around 19C.

missing CPU-z RAM tab

I tried upto 0.95V on the 0.75V rail before the card died and as far as I can tell raising 0.75V doesn't help overclocking at all. So don't bother with that mod.

mod location for the 0.75V rail on the red devil cards. This might help with cold behaviour. Stock resistance is around 2K to GND and 0.75V so you'll want to use a 20K or larger trimmer.

I really don't like the idea of 1 CPU being elligible for 2 different sets of globals. So the 12900K should only be in 1 category. Ranking CPUs by thread count would be extremely disruptive to the current ranking system however I don't feel like it's fundamentally unfair. Especially since the whole point of hyperthreading is to make a single CPU core perform more like 2. The idea behind mixed core CPUs is the same. I feel like the distiction between ranking by threads vs cores is basically: Best performace in X thread benchmark VS Best performing X core CPU A 7700K is one of the fastest quad core CPUs, however it isn't the fastest at 8 thread benchmarks. The current system on HWbot screws over CPUs like the 7600K, 8350K, 9600K, 9700K because they aren't "proper" 4/6/8 core CPUs. The 12900K has 16 cores so for the same reason that a 9700K is grouped with the 9900K the 12900K should be grouped with the 5950X and other 16 cores. Also if we get an 8C/16T+20C CPU in the future that's not really an issue as it would be in the 28C rankings. If it's slower than say 28C/56T or 16C/32T+12C CPUs in things like cinebench it's back to the same situation as the 9700K vs 9900K and I fail to see the problem.

eh I'm planning a dedicated video just about benching the 6900XT.

Here's a link to the full res version of the desktop wallpaper: https://imgur.com/gallery/3Wk2mOk

pin nearest to Vcore VRM is GND

Found the I2C header pads. No need to solder wires really close to the VRM controllers. PCB picture from: https://www.techpowerup.com/review/powercolor-radeon-rx-6800-xt-red-devil/3.html In other news the mem cap mod is not as effective as it intially seemed. It's not +50MHz more like +20MHz.

Stock (1.35V) Vmem. The stock cap configuration is kinda shit. Just compare the back of a 2080 Super to a 6900XT

I added 4x4.7 uF 0402 MLCCs per VRAM chip and 3x470 uF to the Vmem VRM. Card now seems to scale all the way to 2200 mem. Starts to lose performance at 2220. Anyone else want to test?

All the voltage rails I can find on my card. IDK why there's 2 1.8V rails. The BIOS chips probably run on 1.8V and the GDDR6 obviously has a 1.8V VPP but that could be done with 1 reg not 2. the 0.75V and 1.8V rails use MP8633 regulators. https://www.monolithicpower.com/en/documentview/productdocument/index/version/2/document_type/Datasheet/lang/en/sku/MPQ8633A/document_id/1052/

I tried it. It helps a little but seems to make core stability slightly worse.

I run like 2450 min clock with 2820 max clock. With 2600min and 2820max the scores are actually worse. Also if you set min clock too low that's also bad. IMO just running the 2 in lock step is wrong.

So far my exp with the 6900XT Red Devil Ultimate(XTXH core) on air is: Raising min clock too much tanks both FPS and stability. Mem clock significantly above 2150 causes performance loss. Major artifacting and black screens at 2250. (I'd guess this can clear up with more Vmem, VDDCI or VSOC depending on where in the chain the errors are coming from) Even with the 1.2V limit you'll want at least a 410W power limit for FSE. So far I've just been using a power play table with a 511W power limit based on the 6900XT OCF BIOS.

you really need to get and use PScheck if you want to improve your valids. Also a board with 1core per CU mode(old ASUS, any Gigabyte/Asrock).

this bench is so freaking hot. ~370W VCCIN power in HWinfo

I agree that if you want a all core valid ranking it should just be separate from the existing CPU-z valids.

Apparently CPU-z wasn't bin to win enough as is. So next are we gonna ban maxmem for memory frequency valids and require all mem channels instead of just 1?

damn

Baby-J was on a bugged memory ratio AFAIK

yep video about it soon