OccamIT | 2025-05-21 10:09:02 UTC | #1 Not sure if this is the correct place to drop this information. If not I would be greatful if a mod can move to the appropriate location. Received my unit yesterday and have done some basic tweaking and benchmarking. Out of the box config was in line with other benchmarks I have seen. I have done some very basic tinkering with AI but as I am brand new to the local LLM scene I wont publish these as I'm sure my resuls will improve and be more relevent when I start understand the ecosystem a little better. Using Universal XTU I have set the following: Temp & STAPM Temp 99c STAPM Power 140w Slow Power 140w Slow Boost 128s Fast Power 140w Fast Boost 120s Curve Optimiser -50 for All core and iGPU CB23 40700 GB6 SC 2970 MC 17500 Fire Strike 27561 Time Spy 11299 GeekBench results not really affected by the changes but all the others saw massive improvemets. ------------------------- jode | 2025-05-22 02:50:09 UTC | #2 Hi @OccamIT ! Welcome to Level1Techs! thanks for sharing! ------------------------- MakubeX | 2025-05-29 21:07:24 UTC | #3 I received mine as well. I'm about to embark on the journey of setting up Arch Linux and run LLMs using ROCm. Here we go! ------------------------- Drew_M | 2025-06-15 19:34:43 UTC | #4 I just received mine about to install Ubuntu 25.04 today. Is anyone else running linux. I know there was some issues with the wifi but I have read on reddit a bios update has fixed that. Other than a few review videos and a handful of reddit posts not much information on people running linux. ------------------------- OccamIT | 2025-06-19 16:36:51 UTC | #5 I have a dual boot setup with the version on Ubuntu that AMD reccomend for ROCm. I have failed misserabley to get it working so far. I know just enough Linux to be dangerous but this is a brand new area of tech for me so the learning curve is steap. I have been working away for the past three weeks so will pick up on my adventures shortly. Haven't tested the wifi in Linux as I have it connected via ethernet. That works great as far as I can tell. ------------------------- Drew_M | 2025-06-26 20:41:17 UTC | #6 You might be better off trying fedora. Ubuntu 25.04 was working great till I tried updating the graphics drivers and I started running into issues. ------------------------- ulzeraj | 2025-06-26 21:07:11 UTC | #7 @wendell was doing some benchmarks on this chip on a smaller mobile package while the GMTEK Evo-X2 can sustain 120W and burst up to 140W. I wonder how better will those perform. [quote="MakubeX, post:3, topic:230996, full:true"] I received mine as well. I’m about to embark on the journey of setting up Arch Linux and run LLMs using ROCm. Here we go! [/quote] At the moment Vulkan seems to be performing better than ROCm on this chip. ------------------------- wendell | 2025-06-27 02:22:50 UTC | #8 have you had a chance to test vllm? if so can you post your setup, working on the z flow and hp both right now with oddly differing results you doing therock or rocm docker? ------------------------- Drew_M | 2025-06-27 17:50:32 UTC | #10 [quote="Drew_M, post:9, topic:230996, full:true"] If you were to message gmktec on FB I am almost positive they would DHL you a review unit within 72 hours. They have been sending them out to some youtubers(ETAPRIME). My unit is working great the only real issues are the lack of transparency with BIOS updates as they rely on the MB manufacturer and I am not sure they have a single native english speaker. Most of my communication with the company has been poorly translated messages. [/quote] If you were to message gmktec on FB I am almost positive they would DHL you a review unit within 72 hours. They have been sending them out to some youtubers(ETAPRIME). My unit is working great the only real issues are the lack of transparency with BIOS updates as they rely on the MB manufacturer and I am not sure they have a single native english speaker. Most of my communication with the company has been poorly translated messages. ------------------------- Alex_5 | 2025-08-20 08:20:41 UTC | #11 Got mine this month. Installing Linux (ubuntu) was easy, and everything works. However, the fan starts spinning for no reason, even when idle, and the load is often 0.4 (40%), when doing nothing while the cpu's are at only 0.1%. So, something strange happening there. Likely a kernel issue. Installing *amdgpu-dkms* or *amdgpu-install* resulted in some kernel issues and failed. So hopefully theres a fix soon. ------------------------- wendell | 2025-08-21 00:03:31 UTC | #12 bios has a lot of fan control options and thresholds for how the fan is meant to behave fwiw ------------------------- SlayMoreGames | 2025-10-31 14:03:45 UTC | #13 So I got one last night and started playing around with it and noticed a strange behavior running on windows and wanted to see if it happens on linux as well. I set the vram for the GPU in the bios to 96 and tried loading a 32gb model then unloaded it and tried a 64gb model. Watching the 32gb model load the cpu ram spiked like it needed to load to cpu then pass to the GPU, when it was loaded all in the GPU the CPU ram dropped. Same thing with the 64gb model but the model would not load all the way when it would max out the CPU trying to load to the GPU, even with any loading to the CPU turned off on the model loading settings. This morning I tried changing the ram config in the bios to auto and still seeing the same thing. ------------------------- wendell | 2025-10-31 14:34:38 UTC | #14 what software? ------------------------- SlayMoreGames | 2025-10-31 16:04:38 UTC | #15 @wendell LM studio side note you are really making me want to put this with a rtx9000 in my toolbox. ![20250301_105400|412x550](upload://voB5Y4TfLuLR945idUeLLEAEoK3.jpeg) edit* I just reset the system and bios to auto then loaded a 72gb model / after context 118 or so with 84gb set to load in the GPU and it loaded same CPU ram creep and drop and it is showing 64gb in the GPU vram but took the 84. ![image|704x550](upload://3rnYZSHVZftPrQ5fDf9Bt4sAfI5.jpeg) ------------------------- SlayMoreGames | 2025-11-01 14:14:17 UTC | #16 Now I'm wondering if the strange behavior I was getting is not software related but a problem with the hardware because I woke up this morning and got a new behavior. It's dead, no powering on, no lights. I just tried a different power brick I have for a laptop (power output from that is a little less) still no lights, no power. After one day of use, I don't know what to do with this thing grrrrrrr. ------------------------- wendell | 2025-11-01 15:43:23 UTC | #17 noooo. keep us posted on how the rma process goes ------------------------- SlayMoreGames | 2025-11-01 16:34:23 UTC | #18 volt tested the brick and it was good, looking at the motherboard the p-mode button legs in the board have flux pools and swollen sauder. Should I try a different brand? I really want to train the image to 3d AI model i've been making. ------------------------- wendell | 2025-11-01 16:57:05 UTC | #19 check voltage of the cmos battery? There might be some pads there that'll clear the cmos ------------------------- SlayMoreGames | 2025-11-01 17:01:55 UTC | #20 I did :( ------------------------- windfire86 | 2025-12-04 07:36:38 UTC | #21 I'm not a hardware guy, but hope yours ended up getting fixed! I am running Kubuntu 25.10, so the AMD offical ROCm installation stuff is out... but it is too much hassle anyway. Instead, I am running LM Studio with all the Vulkan and ROCm runtimes. Without, them, regardless of what setting I had in BIOS for graphics memory, I couldn't run the new quen-3-next-80b. With it, I can, and it runs at about the same speed as the 30B, as far as I can tell. Next will be linking it to Home Assistant... more on that later. You'll notice no benchmarks. I don't care about tokens, etc. I just want a big enough model to be reasonably accurate answering questions and running my lights and HVAC! And funny is good! ------------------------- Marc_Newman | 2026-02-14 00:04:06 UTC | #22 I repasted my EVO-X2 with PTM9750 and it reduced temperatures from 95-100 to low 80s while gaming. Huge difference. Also the stock NVME heatsink was garbage, one of the rubber band pieces of aluminum, I got one with a heatpipe from Amazon and it fit great, running in the 40s mostly now. -------------------------