I just got myself the 3rd 7900x and will settle on this as I am not going to exchange anymore. The AMD game code promo got me into it as my cost is ~$425US + tax – maybe $40-50 game sale = $430US. Typically, I have Chrome streaming windows open, a VM running downloading stuff while gaming, so 1T boast speed would never happen anyway in my 4K gaming scenario.
The 8 core 7800X3d would be quite limiting with 5Ghz speed, max 5.2 PBO if lucky in comparison. Another deciding factor is the resale value down the road. When Zen5 lands, I assume 7900x MT performance to be slightly better than the 8 cores equivalent while losing ST performance for let say 30%. If the 8 cores sell for $449 again, I can sell the 7900x for $300. If I wait until Zen5 3D where the 8core new is $300+tax, I can sell for $250.
Alternatively, the 7800X3D MT performance would be worse than the 6 core Zen5 assuming the 30% ST increase with clock deficit, the ST would be at 40%. Typically, $299 is the price when people may be interested in 6 cores, so it may be priced at $250. I am more confident to sell the 7900x than the 7800x3d when Zen5 appears.
The 7900X3D is too much compromise since 6core + vcache may not be enough for games and pricey already. 7950X3D is priced out of what I want to spend.
For the testing,
as usual, my 3rd 7900x is failing the AVX test at stock speed again. This one is actually quite bad as I cannot increase PBO to 100 without it BSODing on just general Windows workload. My old one can do +200 despite it failing the AVX test on ccd0.
On this cpu, I settle at PBO=Disabled, CO for general usage at
+10,+5,+10,+5,+10,+5 ccd0
-15,-20,-20,-20,-15,-20 ccd1
At this CO, ccd0 would pass the 1T p95 small AVX, but not the AVX2 or YCruncher 19-ZN2 ~ Kagari. Ccd1 however is able to pass the Kagari tests at these level for 6min+.
For general gaming, ccd0 can be at 0 and ccd1 at all -30 and it did not crash when gaming for hours. I try to find at what CO my ccd0 is required to pass Kagari 1T test but give up as it’s still crashing at +25 CO on core0/1.
ccd0 fail without PBO
ccd1 pass at -CO without PBO
Base on this experience,
1) AMD/Windows mark my crappier CCD as the best cores to use again. I don't have Asus MB so don't know what score the ccd is estimated to be at.
2) AMD need to at least give user the option to set an AVX clock offset in BIOS, if it decides to continue giving out 1 crappy ccd on multi die cpu that doesn't meet its spec so that user is confident they are not silently corrupting data especially for scientific workloads. I know it's not EPYC but failing at stock speed like this feel like AMD is forced to chase performance at the expense of reliability. I feel better when bumping my ccd0 CO up a bit with PBO disabled as I know at stock curve, it's not that stable at AVX.
I got myself a defective CPU again, so 3/3. I will try my luck when Zen5 appears.