- Mar 3, 2017
- 1,747
- 6,598
- 136
This thread should be renamed to "Zen 5 mental gymnastics"
Based on what we know about the Zen 5 architecture, as well as the Granite Ridge chip overall, the reasonable guess here is that we’re seeing AMD’s uncore – the memory controllers and attached Infinity Fabric – stretched to its limit. Since AMD reused the original Ryzen 7000 IOD for Ryzen 9000, the only additional resources available to feed the CPU cores is the slightly higher bandwidth of DDR5-5600 memory. All other cache and interconnect bottlenecks remain.
Consequently, when floating point performance in a single thread improves by a hefty 24%, that’s 24% more traffic through the caches and uncore/IOD to keep those CPU cores fed. And those resources aren’t quite there. To be sure, 9950X doesn’t hit a scaling wall here, as multi-threaded floating point performance is still ahead of the 7950X by 19% overall. But it’s just a bit worse – and enough so that we can measure it. I can only surmise that Zen 5 would have gladly taken more memory bandwidth and IF bandwidth if those were available. Thankfully for AMD, Rate results don’t rely on inter-thread communication, so there aren’t any hazards from threads from different CCDs talking to each other over the IF links.
Strix halo to save us all from poor bandwidthExactly what I said a few pages earlier:
Maybe you just need a non-crappy OS?
Interesting that AMD set 9950X PPT down to 200W. I expected it to be 16-17% faster than 7950X. Now it is 9% faster in Computerbase Test with the 200W PPT. Gaming Efficiency 10% better than 14900K and worse than 14600K. Oh Boy, launching X3D next week at Gamescom is basically unavoidable now. ZEN5 isn't bad overall, just disappointing. For gaming however ist actually IS bad.
Wendell says he's seeing inconsistent behavior in Windows:
Results are all over the place. When it works, it seems to work like we were all expecting:
View attachment 105283
When it doesn't, it even has performance regressions.
He mentions running some games in Linux he gets better performance than the Windows version...
I'm not sure what the reason is, but it seems like the verdict for me is still to skip the 9000 series. It's not living up to its potential in Windows and whether that's an AMD, MS, or both issue needs to be investigated.
Y-cruncher developer also mentions 200W limit, like AT. Oh well, it seems such a messThat is not consistent across reviewers. TPU, GN, and HWUB all showed 220 - 230 W of power with the 9950x.
remember we need DDR5 20000MTs or one just Strix Halo platformZen5% memory bandwidth.
The uncore sucks, long liveNehalemStrix Halo.
Won't it bottleneck on GMI links bandwidth first?remember we need DDR5 20000MTs or one just Strix Halo platform
That is not consistent across reviewers. TPU, GN, and HWUB all showed 220 - 230 W of power with the 9950x.
It is running into amperage limits ala Zen 3.Y-cruncher developer also mentions 200W limit, like AT. Oh well, it seems such a mess
Maybe you just need a non-crappy OS?
Not fully. At least in y-cruncher case, depending on subtest it either runs first into the 200W limit or the 160A limit, what suggest that 200W was a default limit at some specific point in time but might have been lifted with newer bios version or something like that. Y-cruncher author had the data ready 2 weeks ago and only published it today. Might be that reviewers that have 230 W measured on stock are therefore using different mobo/bios combo than Alex was.It is running into amperage limits ala Zen 3.
PBO is your friend.