I know we're all sick of me blabbing on and on about AMD/Intel but I think AMD has X3D in the can. They are probably optimizing their microcode and firmware. I suspect, and predict (total guess... prediction based on gross speculation) that AMD will wait until Intel releases the Arrow Lake refresh before releasing X3D to steal their thunder. Also, Windows 11 2H24 should be on us soon and everyone hopes that restores some order to the benchmark chaos.
One last point. Many of these benchmarks are presented incorrectly. I recently saw a 7800X3D benchmarked against a 9700X and used to declare the 9700X a failure with 3% performance uplift. They need to compare 7900X to 9900X, not 7900X3D to 9900X. This is particularly important for gaming benchmarks where the X3D brings a significant uplift. Anyway, things should be more clear with the next Windows cumulative patch. There are leaks suggesting the new version of X3D will work across all cores and can also be implemented with two X3D chiplets to double the cache. I'm not sure if that would be one large cache, usable by any or all cores, or if each 3D cache would be directly tied to a core chiplet. It has been described as the former but the latter is more plausible. The 7000 series X3D added 30% performance uplift to some games. Even if the 9000 series X3D brings 15% performance uplift, the Arrow Lake refresh is probably not going to win many benchmarks when compared to the cache heavy Zen 5. AMD is also talking about X3D being relevant to non-gaming tasks where previous generations they were direct in saying X3D was only helpful to gamers. Meanwhile, talk of AMD discounts, poor sales, and amazing deals have not materialized at my go-to vendors. I wish they would because I'd like to buy a whole bunch of PC gear I don't need.
We're never sick of your analysis, Tom. Pretty much everything you have posted is what I am gathering as well. If 9950X3D has 3D V-Cache on both CCDs like most believe, it will take all my strength not to update my X99 setup. I just want more PCIe slots and lanes. That still pisses me off.
$AMD is down a tick to $164.96. Perhaps this is due to the Intel announcement later today. Perhaps it is related to the onslaught of negative AMD talk on social media. Rumors make it sound like AMD has X3D standing by to announce and deploy shortly after Intel makes their next move. Everyone I know (not a huge group, to be fair) is solid in the AMD camp. Intel would have to beat AMD by 10% to draw any of them over. I think legions of YouTube talking heads declaring Intel's superiority when Intel only leads in two extremely niche benchmarks is not working as well as it traditionally did.
The ultimate CPU doesn't exi.... https://wccftech.com/amd-5th-gen-epyc-turin-cpus-launch-192-cores-500w-5-ghz-outperforming-xeon/
I must say, I don't understand EPYC Turin. Consider the Turin 9015: Zen 5, 8 cores, 3.6GHz base clock, 4.1GHz boost clock, 125W. Ryzen 9700X: Zen 5, 8 cores, 3.8GHz base clock, 5.5GHz boost clock, 65W. Both CPU complex built on the same process node. The advantage of Turin seems to be double the cache and more memory bandwidth but X3D will eliminate the cache advantage and mitigate the memory bandwidth advantage. EPYC requires ECC. Ryzen supports ECC. Something is going on with EPYC that I don't currently understand.
It's all about that RAM and PCIe lanes: https://www.amd.com/content/dam/amd.../amd-epyc-9005-series-processor-datasheet.pdf Why spend a fortune on a monster if all you need is tons of RAM and massive connectivity? The memory controller and PCIe connections have got to add to the overall wattage of the package.
MSI leaks of X3D performance benchmarks are showing X3D to be uninspiring. Time will tell. I will take a guess that Arrow Lake R will have very little performance uplift, like Zen 5. X3D will also have very little performance uplift. Meanwhile, Phoronix benchmarks will show beautiful gains from generation to generation in the linux world.
Zen 5 X3D parts will be released on November 7. These parts are going to reviewers now. I expect the embargo will be lifted on November 7. Trader alert.
I read the 9800X3D and 9950X3D will both be released on the 7th. I'm not saying that is correct. Lord knows, AMD coverage is as accurate as a Michael J. Fox self portrait.
In support of what you have written, there are also reports of 9900x3d and 9950x3d arriving in 2025. X3D seems to bring a big TPU penalty. 9950X3D should be in the 230W range.
Ive been thinking about this. What work load is the 9950X3D suited for? I suspect the fastest single thread part in the lineup will be the 9800X3D, due to the 5.2GHz boost clock. X3D seems to have a high power cost so it will not be anywhere close to the leading edge of the instruction per Watt graph. Its not like any productivity user needs more compute than a Ryzen 5300. So, what is the application of the 9950X3D? Balls out multi-threaded science and analysis where time is more important than cost? Code breaking, perhaps? That, and nerds who have no need for the power but want the best part because it's cool AF. Let's cut the crap, RTN. I just called you and I out. We're both getting 9950X3D. Neither of us has a requirement for this part. I will appreciate more power to make CAD more responsive. Its not like 9th gen Intel platform is slow, although it does slow down noticeably as models gain complexity. Assemblies, in particular, can tax the system. I've noticed it rarely uses more than a single CPU, although recomputing a model does nail up all 8 cores and can take a few seconds. I'd be better off with a 9800X or possibly 9800X3D.
EPEEN, BRAH!!! I do not upgrade nearly as often as I used to, so I want the fastest sane setup that could last me 5-10 years. I do a fair bit of productivity apps as well as the occasional game. No, I do not neeeed it, but I make enough to justify the want. I really wish AMD would ditch this crippled socket and go a little further. Power users (that are not crazy enough to go Threadripper Pro) are not happy with either AMD or Intel. HEDT has been overlooked for far too long. Dunno, maybe I will wait until a new socket. That'll be a while since AMD stated it'll be at least 2026-2027 before that happens. Stupid AMD. Maybe Intel will get it's head out of its ass by then.
