- Mar 3, 2017
- 1,777
- 6,791
- 136
Exactly. Moreover, if there is a 32 core processor, then 16 core processor becomes cheaper and accesible to more people. In turn more software will be optimized for higher core count.They are different types of devices, with different types of workload. Nobody is using their mobile phone or iPad to execute the type of high throughput MT workloads that some execute on desktop PCs. Hence the higher need for more MT perf on desktop PCs.
I was just pointing out the fact that there is less need for high MT performance on mobile phones and iPads compared to desktop PCs, since you brought up the comparison.That's not the point and I think you know. You say there is stagnation comparable to Intel. In 2018 a tablet launched that was faster in MT than the best non-HEDT part from Intel only 18 months earlier. That's stagnation of MT performance. If AMD is stagnating in MT then the entire industry is stagnating with them![]()
Let me remind you that Zen 2 increased MT but also increased MSRP from $329 for the 2700X to $750 for the 3950X. How much was the 3700X MSRP? How much is the 7700X MSRP? Are AMD 8 core parts cheaper because of the introduction of 16 core parts?Exactly. Moreover, if there is a 32 core processor, then 16 core processor becomes cheaper and accesible to more people. In turn more software will be optimized for higher core count.
That’s still a substantial portion, especially if you add sales of both the 16C CPUs (7950X and 7950X3D) together. You said even AMD was surprised how well the 3950X (and I assume 16C later versions) are selling.
Also, note that not everyone has to buy the top end SKU for it to be justified. Those that only want 8C/16C can still buy that even if AMD introduce 24C/32C variants.
I don’t see how you can derive any such conclusion from that statistics.24C are not needed for the time, looking at the sales vs core count 8C should still keep being the most relevannt for quite some time, here a sample of german weekly sales, you ll understand that a 24C Zen 5 is surely not on AMD s plan :
![]()
The 7950X and X3D are the highest 1T performance parts too, they sell for more than their core count alone. AMD always likes the two to go together to get that upsell.Fact is still that the AMD CPUs with max current core count of 16C are selling very well, indicating that there is demand for high core count AMD CPUs. The crowd that are buying them would most likely move over to buying 24/32C variants instead if such were available.
The 8C 3800X was $399 and the 12C 3900X was $499 though, so not such a big price difference for that higher core count variant.Let me remind you that Zen 2 increased MT but also increased MSRP from $329 for the 2700X to $750 for the 3950X
Good question. Could also be e.g. 8C Zen5 CCD + 16C Zen5C CCD for 24C variant, or 2 x 8C Zen5 CCD + 16C Zen5C CCD for 32C variant.Unrelated but how do you reckon the 32C variant would be constructed? Is it 4 CCD? or CCDs are now 16 cores? Or is it 2 x Zen 5C CCDs? What would be the price of such a part?
If they can fit 3 CCD why not try 1 x 8C Zen 5 CCD and 2 x 16C Zen 5C CCD. Should be even better for throughput.Good question. Could also be e.g. 8C Zen5 CCD + 16C Zen5C CCD for 24C variant, or 2 x 8C Zen5 CCD + 16C Zen5C CCD for 32C variant.
W.r.t. price, that’s hard to know. Currently 7950X is around $550. So if 16C Zen5 would be $649, possibly 24C would be $799 and 32C $999. Note that this assumes use of some Zen5C cores for 24C/32C variants as mentioned above. If only Zen5 cores the price would be higher.
Yes, that could also be an option, assuming it can fit within the max 170W TDP of AM5.If they can fit 3 CCD why not go for 1 x 8C Zen 5 CCD and 2 x 16C Zen 5C CCD. Should be even better for throughput.
Yeah or why not EPYC. Problem is that they are far too expensive.If you want more than 16c go with threadripper.
Yes. A lot of MT workloads don’t need super high memory bandwidth. But it’ll differ per workload of course.Do you really think 2 channels can properly feed more than 16 cores?
If Strix launches this year, per the AMD chart below, it should be 8x5x series (at least on mobile).@DisEnchantment As it seems likely Zen 5 based Strix/Granite Ridge is going to be 9xxx series now with the 8xxx series being a Phoenix Point respin, should the thread title be renamed?
A rising tide lifts all boatsZen1->Zen2 in 2019 went from 8C to 16C, roughly doubling the CPU perf. So ~20% increase for Zen4->Zen5 would be nothing in comparison. If we'll see 32C on Zen5 AM5 (or some combination of P + E cores reaching similar MT performance core count) then it would be something similarly impressive. Otherwise, at a substantial price increase to $999 for top SKU, Zen5 will be a reverse osbourne to announce info about at this point.
What is with 2 downvotes for stating the obvious date??? Seems like somebody don't really follow Intel's timeline and speculate the date...Of course ARL-S will be launching in Q4 2024. Here's the timeline from Raptor Lake-S onwards:-
1. Raptor Lake-S 13th Gen: Q4 2022
2. Raptor Lake-R 14th Gen: Q4 2023
3. Arrow Lake-S 8+16 (N3B): Q4 2024
4. Arrow Lake-R 8+32 (N3B): Q4 2025
5. Nova Lake-S 16+32 (N2): Q4 2026
Unless some showstopper bugs appear, I think Intel will stick to above timetable. As for Arrow Lake 20A, I think Intel will make it for mobile -H only just like Meteor Lake, this one I am not so sure, it all depends on yield and true PPA of 20A, we shall see...
I can't believe people still arguing whether 16-core is enough in 2024. The real question should be what is required to fit in 32-core in desktop space???
Ask me the same question three years later cause Apple, AMD, Intel and Qualcomm are all gunning to launch new generation of CPU/APU to utilize the doubling memory bandwidth of DDR6. It is the PC market share that four companies are fighting for; so tell me do you think any company will miss the boat in 2026?Would it help gaming much? There's your answer.
Also isn't Zen 5 supposed to be a chonker? That 30% more IPC isn't free.
Platform cost is too much when I only want 32 cores.If you want more than 16c go with threadripper.
Yes, because we had 2 channel DDR4 for 16 Cores. So naturally, 2 channel DDR5 should be enough for 32 Core Zen5. That doesn't mean it is optimal, but should be enough. In addition, these processors have large L3 so it can mitigate memory limitations.Do you really think 2 channels can properly feed more than 16 cores? We see large uplifts with the 64 core TR going from 4 channel to 8 channel so we know 32c would be limited on just 2.
Not if you use a combination of Performance & Efficiency cores, where the latter are also clocked lower.With doubling core counts, so does power requirement
Also if the power target is 230W, 32 Zen4 cores can run comfortably within that limit. Even at 125W, 32 core Zen4 is possible without loosing much of single core performance.Not if you use a combination of Performance & Efficiency cores, where the latter are also clocked lower.
How do you propose solving the bandwidth/core problem if sticking with a common shared socket?So apparently the market wanted 16C already in 2019 after all.
Now that was more than 4 years ago, and we’re long overdue for another core count increase.
You mean memory bandwidth per core? That was already discussed in previous posts. In short, I don’t think it’ll be a problem for most MT workloads.How do you propose solving the bandwidth/core problem if sticking with a common shared socket?