keks2k

Let's hope they're drawing the right conclusions from that feedback.. 🤞

This! I'm wondering if we could get the dGPU heatsink in order to gain a little headroom. At least for short spikes it should help, maybe even a bit more. If only I didn't like the overall machine. The Lenovo Legion 7i was actually able to run it on sustained loads around 130W. Downsides for me were the fans never really turning off and running with a high pitched sound. Plus the slightly rickety keyboard The Lenovo P16 also got returned as a) the bios was completely locked, so no UV and stuff, b) memory only ran at 4000 and c) I didn't get used to its fans either. 😔 BTW. is anyone running the 7770 on Linux, or are you all Windows-bound? I'm getting thrown back to 55W rather quick, whereas on Windows it stays around 85W so there are some things going on with their drivers (or maybe even DTT).

I just stumbled across this one: Perfectly nails it! He could've dived into the improper cooling a bit more, not only the high voltages (the llc observation?) and the underrated power brick, as he pointed out to have even repasted one of his units, though. But that's just nitpicking I guess..

Did you actually receive your machines with a recent bios version / did you see performance impacts during updates? The first 7770 board came with 1.5.6, now I got 1.2.1. and wondering, if I should update, as going down from 1.5.6+ is blocked.

You really make me jealous.. but getting closer! 🤪 So you're repasting your machine on a regular base?

Just a quick recap: 1. Heatsink/cooling system has been replaced, now it's the Sunon variant - which to me are still audible while spinning up (I think that's what @win32asmguy is referring to with the "creaking" noise). To me it sounds a bit more silent but I guess that's also a matter of taste. I just don't care.. 😉 Result: core temp distribution changed. Max. delta of 20 became 12, but there's probably way more to those deltas than only the paste job (thinking of core quality/leakage/contact pressure), so I'm not too sure if that's a good or a bad thing. After all: minimal CB benefits. 2. Mainboard has been replaced now, too and results got better, reaching 21k first run in CB23 with the usual suspects of uv, ia/llc, PL1/2 tuning etc.. maybe I'll invest a little more time, but as it's settling to stable 19k even after 10 minutes, I'm now on the verge of "being okay with it". Guess that magic 20k barrier was crucial to my inner monk. 3. Turns out decreasing cache ratio to 32 actually improves performance -at least for CB23- by enabling voltage reduction by -100mV, which was not possible before. In fact, 32 seems to be a sweet spot, as higher and lower values reproducibly decrease CB results. Could be related to the integrated memory controller. Would be highly interested if some of you share these experiences?! Questions: - Has anyone of you seen a setup var for the cache ratio? Haven't seen anything obvious in the decompiled bios. Otherwise I'd have to figure out, how to set it from linux. Putting rock-stable values to bios vars would just be pretty convenient. - Same question regarding offset voltages.. 🙂 Finally, as some others already pointed out (thanks for the nice rant @ATAN) : it's such a shame Dell didn't add proper cooling into that otherwise real beauty.. Thinking of one or two more heatsinks/-pipes and maybe another fan. Why, Dell..?

i felt the same way first day or two until i got the undervolting going, will you copy my exact undervlting setting and share your CB results? (note that i also locked all my cores @4.7) Jep, tried them, among other things, as a base at some point, but voltages are way too aggressive: as I mentioned earlier, I had to cap my cache multiplier to 41 - which takes away quite some performance in CB. Also liked/tried the idea of changing frequency settings: cap at 4.7/8/9GHz, disabled some e-cores (e.g. 4/8 and stuff), capped e-cores at 3 GHz, ... but no luck. Only hope would be LM - but the longer I take it into consideration, the more I dislike it somehow. Will take a night over it. See what the sales rep comes back with tomorrow (it's almost midnight here).

that's what I thought as well, plus maybe potential power connectors. And the sales rep told me that one might add in a dGPU later on - in that case you would need the yellow/brownish connectors anyway (not sure if it would make sense to supply those to any dGPU-less system in this case)

Hehe, thanks for the encouraging words! I've already really spent quite some time on it with uncountable runs, but especially the observation with cache frequency / voltage correlation makes me worry about a bad go from silicon lottery in my case. 😟 So that's why I'm unsure if putting LM on it would be a good idea, with returning/replacing it in my mind.

Jep, it's the Delta fan version I received. Going to drop them a message regarding the different fan suppliers and ask for possibilities / replacement. Will report back the outcome. Regarding the iGPU: yes, I did that on purpose as a) I don't need any graphics power and b) were hoping that thermals are better, as the cooling system doesn't have to dissipate extra heat from the separate circuitry (dGPU, VRMs, memory). Even took care of power draw of the internal graphics during testing by also reducing its voltage - but to no avail: results didn't change, which is what I mostly were hoping for. As for the internals:

Hi guys, just wanted to join the discussion, which I've been silently following for quite some time now, as my 7770 (with only iGPU) finally arrived - and delights me with bad thermals, too. I already played quite a lot with: - undervolting, - power limits, - frequencies (capping e-cores in some combinations achieved some positive effect), - disabling cores - fixing the IA setting (1.7 mOhms - thanks for pointing that out!), but all that effort with only so-so results. After all, the best score I ever got was 19k on a first run CBR23 at roughly 80 watts, which then heavily dropped further. That's pretty disappointing to me. One of my devices weak points seems to be the cache: I had to cap its multiplier to 41 in order to achieve only -0.04 volts on the P-cache (and therefore the CPU, which seem to be kind of locked/inter-related). Temp-wise I observe a P-core delta of 20 degrees after a repaste with MX-2. Before it was around 16, like @win32asmguy got on his machine. So my job seems to have had at least some negligible effect into the right direction. One thing I noticed: The cooling plate and CPU die don't seem to be nicely aligned. There seems to be much more pressure in the middle and towards one side in respect to the other (see attached pics; less pressure on the left side on both attempts). Tomorrow I'll be receiving some LM paste, but currently not quite sure if I really want to go that route; or if I should try to get the otherwise lovely device replaced.. otoh/from what I'm reading here, I'm not that convinced I'll be more lucky next time. What do you think?

I ordered a 7770 in early September and the expected delivery date went through the roof (to 2 months) after my purchase, despite being labelled "available quickly", so I contacted the sales rep and she told me that it was due to delivery problems of the bottom cover with the maintenance flap (to access the primary SSD without disassembling the case). She offered to cancel my first order and replace it with a new one, this time without the flap => and there we go: it arrived only 12 days later. The part in question was "354-BBFJ" which got replaced by "354-BBFI". ..now I have to fight bad thermals, which I'll join the discussion in the other thread here for. 🤬