![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Yeah - this seems to be a CPU only for workstation-type workloads and not for gaming. AM6 is the play for anyone that is a gamer. Although there will be a TON of people that treat this as AMD's "Intel KS" part because they want the best now (until next year). It did not last long on Amazon so I wonder if they even made that many.
Exactly this. Zen 6 single-CCD X3D will be the ultimate end processor for gamers on AM5 to upgrade to, not any of the dual-CCD options.
I don't think so.
Well, to amend your statement to be more accurate: We now know that extra v-cache on the second CCD doesn't really help in gaming. The cross-CCD latency penalty still applies.
would it be on the same CCD maybe it would, a game want to live all on the same CCD, this product was never much expected to help gaming (outside some OS scheduler mistake)
On phoronix test suite the 9950x3d->9950x3d2 boost is not that dissimilar to going from a 9900x to a 9950x, you need a workload that love cache and parrallelisable accross multiple ccd, some scientific computing will be able to take advantage.
(can easily be fixed but fails out of the box)It's far better than I expected.
Yes, I expected them to perform worse, but their results are very close.
how so ? about a tie it seem (which is logical), even a bit worse, difference is probably just the clock speed a bit lower (that would be why it is worse for both, same ccd or cross) ?
If you need coherence that would assure the need to always transfer the data and a fast dirty/clean up to date status for everything and of course you cut the effective cache capacity in half (if it is small enough what you can fit it in 50% of the cache, may has well simply run all that task on a single CCD for now, I imagine some special case exist for such mirror mode too if large portion can be considered read-only safely, would be interesting to see)
It's just as expected anyhow, you can still find games where it helps maybe, but do you play them? And how much do you care about them, specifically?
Wendell found Borderlands 4, for example:
View: https://youtu.be/1PcO2k6vprY?t=664
He also highlights some evident driver issues though, since the 8 core part smashes the 16 core one in some titles so, yeah... dual CCD still sucks for gamers
Oh and it also gets really hot and pulls a fair amount of power with all that cache, again not very surprising, but I would actually not want to put it on my entry level AM5 motherboard, I think.
Many of the squares, especially the worse ones in the ligher green color, are 2 - 3 nanoseconds better. And some are more like 4 - 6 ns better.
Yea but they are better on the V1 according to this picture, but maybe that's a mistake.
Ah......yep. I got that mixed up. Sheesh.
The difference is 200 MHz plus another 200 MHz from PBO (if used (and hardly achieved)).
That not a latency benchmark, having more cache can help augment cache it and not needing to go read in ram (or on the other core CCD L3), this can happen with a worse latency easily (which the v2 has, according to your graph)
Yeah I mean CCD to CCD and on V2 everything is in cache, on V1 it just jump between both CCDs.
it is a strange one, was able to push a 5600x at 100% usage (12 threads), gained a lot from going 8p core 8 ecore to 8pcore-16 ecore at launch:
That's a pretty different result from what TechSpot got. TechSpot's result was more along the lines of "meh, don't worry about it". It was basically tied with the 9950X3D in BL4. Of course they tested different settings - medium and badass, while that test was run on "very high".
But the cache is split on the V2, so there is still jumping between both CCDs, anyways. V1 is 32+64MB on CCD0, 32MB on CCD1, V2 is 32+64 on each CCD, that's it.
Intel cores are faster, but AMD cores get the data faster - so in real world scheduler is what made one or the other "quality".
Steven Walton's results sometimes differ from those in other parts of the world and have a good chance of being wrong.
I grabbed one from MC gonna pickup today.
The game probably scales up with extra cores. And having Vcache on both CCD's, means the only penalty is cross CCD communication. With the 9950X3D, you would also have the imbalance of the 2nd CCD not having extra cache. So, it may either try to send to RAM, or simply try and keep everything on the Vcache CCD. Thus, losing the core scaling benefits.
Won't happen. Dual CCD chips will still have per CCD L3 cache. Zen6 is going to up to 12 cores per CCD though, so you'll be able to get a 12-core chip with a single cache or a 24-core setup with 2x 12-core CCDs. Zen6 is also supposed to bring big improvements to the CCD to I/O die interconnect.
