Soldato
Scrap Alder Lake after 50 days in the market. Yeah, no.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
AMD demonstrates Ryzen 9 5900X prototype with 3D V-Cache stack chiplet design
- Thread starter RavenXXX2
- Start date
Scrap Alder Lake after 50 days in the market. Yeah, no.
Could be sold as part of the same Alder Lake series /12th gen, so models with additional cache with the E-Core count reduced (or removed entirely) could be sold as a '12700C' or '12900C' (depending on the clock speed). The 'C' would simply mean additional cache, presumably at about the same clock rates as the K CPUs with similar names. These parts would still have 8 performance cores, so this naming scheme /designation would make (some) sense.
We know Intel has designed SKUs for CPU models without any E-Cores included (i5-12600, 12500, and Core i5-12400).
Or, just part of a new Golden Cove generation towards the end of the year (November, like Alder Lake in 2021?), but different to the rumours about 'Raptor Lake', which suggested the opposite, an increase in E-Cores, which have been the least well received part of Alder Lake, and will not help Intel to keep ahead of the Zen3D lineup.
Without getting into another argument / discussion about this - I've been doubtful that Intel would be able to increase the core count and L3 cache using the same fabrication technology as Alder Lake (The upcoming server CPUs will also use the 'Intel 7' process). With evidence of an improvement to the current 'Intel 7' process, I'd be more inclined to think these rumours may have some truth in them.
I think Intel's server CPU launches will take precedence over all other launches (besides the rest of the announced Alder Lake lineup) though, in 2022.
Are there any technical reasons why Intel can't make 8 Core (no E-Cores) Golden Cove cores for desktops, with additional L3 cache? Higher cache amounts will no doubt be possible on the Golden Cove server/HEDT platforms (with much higher Golden Cove core counts).
Last edited:
Soldato
The die is designed for a certain amount of cache, Intel can't just slap more onto it. The only reason AMD can slap more cache onto Zen 3 is because the chiplets were designed from the outset to allow such a thing, even if it was never used. The Alder Lake die is also designed to have 8 E cores, they're just disabled for the lower SKUs. Essentially what you're proposing is Intel redesign the entire thing. That's not gonna happen. Intel are 100% committed to the hybrid design. Alder Lake is a good proof of concept, Raptor Lake should be a nice refinement, but it's Meteor Lake that's going to be the proper test, not only as Intel's ability to make such a design viable without stupid die sizes, stupid power consumption and stupid thermals, but also against Zen 5 and Zen 4C when AMD also implement their own hybrid approach.
Also note that a purely P core die is Sapphire Rapids HEDT, and Intel will not cannibalise their desktop sales by offering a P-only alternative for those who really don't want E cores: pay the money for HEDT or nothing.
Intel are stuck with the design. They can't add more P cores because they don't have the power or thermal envelope to do so, which is why Raptor Lake is supposedly adding more E cores. Raptor Lake is suggested to use "Raptor Cove" for the P cores, which is a tweaked Golden Cove. It's possible Intel could do some last-minute die changes to allow more cache, but I don't think there's any space for it.
Disagree. AMD's presentation so far has only given an average 15% uplift in gaming performance from the stacked cache. That's a specific number for a specific task, and will not be enough to retake the gaming lead from Alder Lake's mid and low-end SKUs. For productivity workloads, it's going to come down to the specific job. Anything that benefits from cache will see a massive jump, anything that doesn't use much cache will see little to no benefit. So V cache alone is not enough to overcome Alder Lake's lead.
Except for the 5950X. That's already mostly tied with the 12900K, slap 15% on top of that and AMD retake/retain the ultimate performance crown.
Glue?
I suppose the other possibility is that, 2022 might be a mostly uneventful year for Intel and AMD (for desktop CPUs), besides Zen3D and the remaining Alder Lake CPUs.
Wonder if Zen3D might have a small increase (like a refined version of Zen 3) in IPC as well as extra cache? That would make it quite a tasty CPU series.
BTW, Merry Christmas all
Last edited:
Remember that 15% was with both CPUs clocked at 4ghz where the extra cache will likely make a larger impact vs both the CPUs running at 4.9ghz if indeed the Vcache versions can match the older variants in clock speed.
Soldato
IPC no, but there have been suggestions of minor speed bumps as TSMC's 7nm is just ridiculously mature now.
Soldato
And that 15% was upto 25% It will be interesting to see Zen 3D performance over a broader range of tests and if AMD have mixed in a little infinity cache secret sauce, for pairing an RDNA graphics card and Zen 3D CPU.
Yeah, sorry forgot what it actually is lol.
I remember that it basically allows much higher Video memory bandwidths...
It's a pretty major limitation, given that (currently) Nvidia is able to produce way more cards than AMD.
Not sure this will have any thing to do with Zen3D though, probably need to wait for RDNA3 cards to see changes to infinity cache design or amounts of cache.
I remember that it basically allows much higher Video memory bandwidths...
It's a pretty major limitation, given that (currently) Nvidia is able to produce way more cards than AMD.
Not sure this will have any thing to do with Zen3D though, probably need to wait for RDNA3 cards to see changes to infinity cache design or amounts of cache.
Last edited:
Secret sauce perhaps, but how much logic is required to keep a cache coherent "inside" (meaning here on the same package), how hard would it be to extend any of these concepts of tags etc. over PCIe bus? On the other hand CPU and GPU don't really care what the other is doing in terms of dirty pages etc., so maybe some way for the Zen3 3D to "give" some cache to an RDNA2 card? Obviously for the RDNA2 card the cache is far farther away than on-die RNDA2 Infinity Cache, but it could be "nearer" having to go to main memory via PCIe (and via really means via the Zen SOC anyhow). Hard to see how how that would help the GPU - certainly in any conventional way - maybe they were able to come up with something? Set your Zen3 3D back to being sans 3D cache but let the RDNA2 GPU have it instead.
@jigger- Infinity cache itself doesn't have anything to do with whether the CPU is made by AMD or Intel though, does it?
It's just a feature of RDNA2 graphics cards.
One improvement for RDNA3, might be instead of 128MB of L3 infinity cache, it could be Level 2 cache instead, which would probably cost more and I assume, be more difficult to produce. But, it would result in lower cache latency and less cache 'misses' (I think).
It's just a feature of RDNA2 graphics cards.
One improvement for RDNA3, might be instead of 128MB of L3 infinity cache, it could be Level 2 cache instead, which would probably cost more and I assume, be more difficult to produce. But, it would result in lower cache latency and less cache 'misses' (I think).
Last edited:
Do we know yet if AMD is planning sneaky (low key) launch of 6nm desktop processors (not the 6nm APUs) towards the end of 2022? Then, maybe follow this up with Zen 4 6 months later?
In theory, these could use an AM5 / DDR5 platform like Zen 4.
The lack of 6nm desktop CPUs has been the most confusing part of their release schedule, at least for a lowly commoner like me. I'd assumed this wasn't happening due to the lack of info on official roadmaps, but who knows?
In theory, these could use an AM5 / DDR5 platform like Zen 4.
The lack of 6nm desktop CPUs has been the most confusing part of their release schedule, at least for a lowly commoner like me. I'd assumed this wasn't happening due to the lack of info on official roadmaps, but who knows?
Last edited:
It does seem a bit unnecessary to release desktop Zen3D CPUs on 7nm, then release 6nm desktop CPUs later in 2022.
It's also problematic/messy in terms of what AMD would call Zen3D based on 7nm.
I'm more inclined to think that AMD will want to reserve their 6nm capacity for new graphics cards, to keep up with Nvidia's 'Ampere Next' cards.
It's also problematic/messy in terms of what AMD would call Zen3D based on 7nm.
I'm more inclined to think that AMD will want to reserve their 6nm capacity for new graphics cards, to keep up with Nvidia's 'Ampere Next' cards.
Last edited:
Caporegime
If you use a horizontally aligned CPU cooler it should help keep the VRMs cooler.
If you wanted a coherent fabric then it probably would. I think you could be thinking of smart access memory.
Soldato
And even that's just a marketing name for Resizeable BAR which Nvidia has now rolled out support for.