Hey - a summary of my experiences in a SFF build.
I switched from the 7600X to a 7800X3D in my Dan A4 with a Noctua L9a with Duct.
Running with PBO -35 stable, CPB enabled.
Everything went great - the CPU is super efficient in gaming, hitting not more than 50W and 75C in most of my titles with 120Hz FPS limit - Cinebench yields around 94% of multicore performance with the cpu sitting at 89C with sustained power draw of 82W like expected - so fine so good. 1 core, 2 core and 4 core performanc eis above average - 8 core is just a bit too much for the cooler so it has to calm down abit but 94% is absolutely acceptable and i have no regular all core workloads.
But TLOU shader compiling absolutely kills the CPU. The power draw barely hits 70W but the temps shoot through the roof - even surpassing the hard 89C limit - Tdie shoots up to 95C within 15s leading to a hard shutoff - very bad feeling. It seems that Tcase is used for the temp limit and Tdie is able to surpass that. Also there is a seperate temp limit for the Vcache - at 70C or just below of that - i won't figure it out again exactly (hard shutoff)
I had to set 80C as thermal limit in BIOS - so Tcase stays at 80 - Tdie reaches 85C and the Vcache stays below 65C.
I still get 99% of gaming performance and all the cache benefits - I game mostly games like ANNO series, Cities Skylines , and so on...
Also very interesting to me was that the die temps are not directly related to the power draw - it seems that special calculations produce significant more heat than others. I know that phenomenon from Intel chips and AVX code pretty well but have never seen such significant real life differences. Some examples without lowered temp limit:
Hope you find it interesting and i am heavily interested in further experiences.
I got very bad responses in the AMD subreddit - people just don't get the point of making a SFF build work that shouldn't work or just do not read the post at all.
Link:
Often i got links to Noctuas website that the L9a is not capable of handling the baseclock of the 7800X3D - which is just untrue. With CPB disabled (CPU stays at baseclock) temps never even exceed 80C with out any limits - just boosting is the problem. I figured out that limiting the target temp is way more effective than limiting all core boost clock or manipulating single core boost limits - this way i just get the maximum performance possible without any temp problems anymore.
I switched from the 7600X to a 7800X3D in my Dan A4 with a Noctua L9a with Duct.
Running with PBO -35 stable, CPB enabled.
Everything went great - the CPU is super efficient in gaming, hitting not more than 50W and 75C in most of my titles with 120Hz FPS limit - Cinebench yields around 94% of multicore performance with the cpu sitting at 89C with sustained power draw of 82W like expected - so fine so good. 1 core, 2 core and 4 core performanc eis above average - 8 core is just a bit too much for the cooler so it has to calm down abit but 94% is absolutely acceptable and i have no regular all core workloads.
But TLOU shader compiling absolutely kills the CPU. The power draw barely hits 70W but the temps shoot through the roof - even surpassing the hard 89C limit - Tdie shoots up to 95C within 15s leading to a hard shutoff - very bad feeling. It seems that Tcase is used for the temp limit and Tdie is able to surpass that. Also there is a seperate temp limit for the Vcache - at 70C or just below of that - i won't figure it out again exactly (hard shutoff)
I had to set 80C as thermal limit in BIOS - so Tcase stays at 80 - Tdie reaches 85C and the Vcache stays below 65C.
I still get 99% of gaming performance and all the cache benefits - I game mostly games like ANNO series, Cities Skylines , and so on...
Also very interesting to me was that the die temps are not directly related to the power draw - it seems that special calculations produce significant more heat than others. I know that phenomenon from Intel chips and AVX code pretty well but have never seen such significant real life differences. Some examples without lowered temp limit:
| OCCT | 88W | 85C |
|---|---|---|
| Cinebench | 82W | 89C - how it should be |
| TLOU shader compiling | 65-70W | 95C - hard shutdown |
Hope you find it interesting and i am heavily interested in further experiences.
I got very bad responses in the AMD subreddit - people just don't get the point of making a SFF build work that shouldn't work or just do not read the post at all.
Link:
Often i got links to Noctuas website that the L9a is not capable of handling the baseclock of the 7800X3D - which is just untrue. With CPB disabled (CPU stays at baseclock) temps never even exceed 80C with out any limits - just boosting is the problem. I figured out that limiting the target temp is way more effective than limiting all core boost clock or manipulating single core boost limits - this way i just get the maximum performance possible without any temp problems anymore.
