24 Core Ryzen 9 8960X3D - AMD’s Secret Zen 5 Flagship

Yeah, to h377 with the 3D cache, nobodyknows how to schedule a lopsided processor... your friend was right, just give us the 32 cores on Zen5.
Playing games is a waste of time anyway, and the bogg standard cores can run all your games at top speed ultra settings anyway.
If you put the 3D cache as an off die chip, so it can be used all the time plus the 32 cores, then okay. But not lets go cutting out half the cores just to use the Vcache.

i let others beta test Zen 5 😂 currently i rock a 7800x3d (thank good not to many issues)

Jup the scheduling issues are real, but on my 7950x3d with process lasso completely under my control. I set everything to prefer the frequency cores and use process lasso to force games on the 3d cores. By having all other processes like browsers, discord, recording software running on the frequency cores and only the game on the 3d cores, latency spikes and stutter is greatly reduced. The game basically has 8 dedicated cores upon which nothing else runs.

Well this is totally disappointing.
Only one more CPU upgrade for Zen 5, why even bother. Why even bother having gotten on this platform.
What a rlp0ff.... maybe if Ryzen 9000 and 10,000 processors would of worked on it, it would of been worthwhile.
I'm not even going to bother to go from an 7950x to an 8950x, because there is no compelling reason to do so. Yeah, we want a 32core 8960x.... at maybe a 600 dollar price point.

What about 1 Zen5 normal + 1 Zen5c with X3D (yes the x3d on the c model due to the higher core count)
Me thinks this would be an interestingly performing beast ;)

it will become 7955HS3D

I really don't want mixed cores, I just want simple options. Let me pick the most efficient chip, or the most performant, or the triple cache per each tier.
I know for me normal chips are the best choice. 3D vcache doesn't help me as much as higher boost clocks, and 1MB / thread would hurt me more than extra cores would help.

even if they dont released the 3d, im waitin for it! thx for the awesome video g

color me pink, subscribed

Thinking of Efficiency cores and all these CCores is really a nightmare.

Because we have to deal not just with architecture and cores jumping around, causing a delay/stutter in games and desktop use as a whole like 13000 Intel family, but also dealing with infamous Windows 11 scheduler, that is no good imo.

Same issues with scheduler caused many people people to upgrade from Windows 10, because they had 12th gen and 13th gen Intel issues, and people like me with 13900k still notice these stutters, and I found that more than 1ghz difference in frequencies between P-Cores and E-Cores cause issues.

Overclocking E-Cores to 4.4ghz from stock 4.3 showed immediate decrease in stutters, and overclocking ram further stabilized the system to reduce the stutters, but not fix them as a whole.

This little core structure is detrimental in performance. Most likely causing stutters.
By the way, 13900k does have stutters in games, they are very very micro, but they are there. It seems like FPS drop, but I am convinced it is tied to scheduler and P-Core/E-Core interaction.

The more complex something gets(7950x3d) the less reliable it is. Let's hope AMD smashes it out of the park with the next CPUs

7950x3d is full facepalm.

Let's break it down for an AMD buyer:

Someone wanting the last bit of Multicore Performance -> 7950x NON 3D
Someone wanting the last bit of Gaming Performance -> 7950x3d with Manual Process Lasso optimizations
Someone wanting the last bit of Gaming Performance without haggling manually -> 7800x3d
Somone wanting optimal Gaming AND Multicore Performance without manual haggling? -> 7970x3d with dual x3d. The one chip that doesn't exist.

Pretty sure the customer basis for the fictional 7970x3d would be way higher than the 7950x3d retard chiplet. Stupid decision and AMDs claim can be easily disproven. Some games lose fps heavily compared to a 7800x3d.

i feel cheap with my 5800x3d

I am really struggling to decide whether to upgrade from my 12600K to a 14700K or wait for Zen 5 / Arrow Lake, the latter options costing loads more than an easy drop in upgrade.
From what I can see, using Cinebench R23, 2024 and CPU-Z, Intel seems to win in ever test whether it's single or mutli threated, sometimes only by a few percent but sometimes like 24%!!
And if you are correct in predicting that Arrow Lake is 30-40% faster in single thread (which Intel is already faster than AMD at) and AMD will only get maybe 25% single thread increase it's going to be no contest here, plus if Intel will increases multi threaded performance maybe ~40% for the initial 8p+16e and then maybe ~50-60% for 8p+32e later on and you are saying AMD might only get a 35% increase if they don't release Zen 5c to AMD 5 which is still a massive speculation piece at the moment as no source has given you any indication that is the road AMD is looking to take then I just don't know what it is I am missing when you say this fight will be a close fight, obvious Cinebench and CPU-Z don't represent every application and game out there but AMD seem a little behind right now and you are saying Intel will increase more than AMD soooo...???

That is basically a 13900k, but better in every way.

So no high clocking cores then? Seems like a laptop product.

Missing the attic vibe

AMD monster will only be truly terrifying if it’s also not too expensive. Low prices on Zen 5 = symphony of horror.

Wasn't zen5 not on am5

Man Intel is SO DEAD

what the hell is that name

Don't forget to LIKE the video

That's 6000 MT/s DDR5 speeds in GEAR ONE!! That is where the memory controller runs at the same data rate as the RAM.

Intel CPUs on the other hand can only run DDR5 in Gear Two, which means the memory controller running at half the data rate of the RAM.

DDR5 running at 6000 MT/s in gear one is roughly equivalent to 7800 MT/s being run in gear two.

One thing that they should change in Zen5: move the physical placement of the CCDs on the package to align to the center of the socket (instead of slanting towards one side like Zen4 now), this could make a not-so-insignificant difference for CPU coolers.

To get significantly ahead of their competition, AMD should eliminate the chipset, increase PCIe lanes, and bring quad channel memory to a mainstream platform

My dream for zen 5 is that amd would redesigned the io die and put 8 zen 5 cores there so they could still have 24 cores at the high end with just 2 ccds

This will be my next cpu unless something comes out..

Seems like AMD's C-cores is what Intel's E-cores wish they were

Wow - so much interesting topics to wonder about.

we don't need more cores though we need more pcie lanes. 16 is more than enough cores but two x8 pcie lanes just is not enough.

Increasing multithread performance only at the top end still keeps the intel advantage at the Ryzen 5 and Ryzen 7 tiers.

If AMD wants to surpass Raptor Lake lower end and mid range parts and catch up to Arrow Lake equivalents they will need to use two chiplets in all SKUs and have the same segmentation as intel in regards to big and little cores:

Ryzen 5 being 6 + 4, Ryzen 7 being 8 + 8, Ryzen 9 being 8 + 12 and 8 + 16. 13600K is already 50% faster than the 7600X in multithread performance. AMD really needs to bring 100% gen over gen multithread improvement to this part to leap frog ahead of what Arrow Lake would bring as a 15600K

A whole lot of hope and benefit of the doubt to AMD on this one....

I'd purposely buy the 24 core version if it is a smarter engineering decision than the lackluster 7000X3D chips.

The problem isn't the 7950X3D, it is Windows scheduler. Knowing that a core is part of the the CCD that can boost higher or not is something Windows is already aware of, but the logic in the scheduler doesn't leverage that, hence the crappy performance. We are seeing this issue with Windows schedule since the early days of multi-socket CPU with NUMA where threads were just "thrown" on anycore, regardless of the latency to get there (aka on another socket), vs waiting a few more cycles to just move a thread on another core on the current socket.
When Windows will adopt the Linux kernet to build Win 12, Microsoft will be able to fix this as it is already working properly under Linux.

I’d rather have a all x3d zen5 chip with 16 cores. No need to muck with scheduling and you have plenty of cache for those that need it.

It would be better to just have 2 chiplets with 16 Zen5c cores each + Vcache on both. That would be a high enough uplift that I would consider replacing my aging Threadripper 3970X with that. I don't care about hitting higher clocks, I wouldn't be able to keep that many cores cool at those clocks anyway, even with a custom loop.

Dude, I appreciate all the leaks and the information and speculation, but for the life of me I just dont know what these 'bergamo/turin/raphael/zen 4c' things refer to. Do you have an educational video that covers what these refer to?

This 8960X3D CPU makes no sense:
a) Games that do not need more than 8 cores will not run any faster than a 8800X3D (see 7950X3D vs. 7800X3D)

b) If modern games start using more than 8 cores (and we are very close already):
The scheduler will pick the 8 highest priority threads and puts them on the X3D CCD (so far so good).
Then it takes the lower priority threads and puts them on the Zen5c CCD where they only have 16MB of L3 per CCX. This causes L3 hitrates to collapse in modern games. This is not a big problem for those lower priority threads, but it will cause a lot more DRAM access which will then bottpeneck the X3D cores as well because there are no priorities regarding DRAM access.

You cant stack a regular VCache chiplet on the Zen5c CCD either because they have 2 L3s instead of one plus the VCache would cover the cores on the inner side.

I think what would make sense is:
.) Release the 8950X3D with VCache on both CCDs
.) Release the 8960 in a 8+16 configuration without VCache targeted at content creators.

Nah, I'd rather have the 2x Zen5c plus the stacked cache. Yes, this is an even farther fetch dream.

we need less cores.. more powerful cores.. the e core craze is bs and bad for gaming.

When did you move out of the attic?

I wonder if AMD is going to feel pressured to make this cpu because of the threat of Intel's 8+32

It would be nice if the Radeon team were just as competent as the CPU team, because they're utter dogshit when you do compare. From a marketing perspective they need to sack the likes of Herkelman and Azor and have their CPU counterparts do both sides of the marketing.

AMD's elusive silicone ship Yamamoto. . .

I would definitely buy a 24 cores amd cpu. If they dont I'll go with intel

I also think AMD could copy Intel.

would nice to have 8 + 16 cores CPU from amd like intel hybrid but as a x3d focused on fast CCD one.

According to pcpartpicker, the Ryzen 7 7800X3D costs $70 more than the Ryzen 7 7700X. I would rather save the $70 and buy a more powerful GPU than CPU. The majority, if not the vast majority, of pc builders actually live on a budget.