Talk:Zen 3

Latest comment: 1 year ago by AP 499D25 in topic V-cache size confusion

Discussion of process

edit

The confusion over what AMD meant by "7nm+" represents a considerable portion of the coverage on it to this point, and it is relevant that there was some confusion. As this becomes less significant to the article it can be trimmed later. I'm not sure I see this as "useless" information at this time. —DIYeditor (talk) 14:38, 17 March 2020 (UTC)Reply

Feedback from New Page Review process

edit

I left the following feedback for the creator/future reviewers while reviewing this article: Nice work. Of course the sources and article and sources will need to be broadened as time progresses..

North8000 (talk) 19:36, 23 April 2020 (UTC)Reply

Zen 4 is announced successor to Zen 3

edit

Not sure the logic of removing that. --Castaa (talk) 00:01, 30 June 2020 (UTC)Reply

Definition of kilobyte

edit

Who even added this unnecessary and redundant note to every Zen page? Programmers and engineers define kilobyte as 1024 bytes for decades and it is kibibyte needs clarification not kilobyte. You are free to add this "i" between K and B if you are very pedantic but there is no need to treat readers as idiots and write this note everywhere. No one is going to sue Wikipedia if we don't add "do not iron clothes while you're wearing them" in the Iron page. 109.252.89.157 (talk) 12:46, 18 October 2020 (UTC)Reply

Improvements Section

edit

Should the "Improvements" section under features be deleted? It doesn't really seem necessary as all the improvements are stated in the paragraphs above it. Awesomeplaya211 (talk) 15:43, 20 November 2020 (UTC)Reply

Zen 3 core complex core number?

edit

"The base core chiplet has a single eight-core complex" Since there's e.g. a Ryzen 5 5600X with 6 cores (and Ryzen 9 5900X with 12 cores), there are obviously also Zen3 CCX with a different number of Cores than 8 (probably less than 8 cores)... — Preceding unsigned comment added by 84.188.92.132 (talk) 08:15, 17 March 2021 (UTC)Reply

Should Zen 3+ have a new page?

edit

Zen 3+ has significant differences from Zen 3, as much as from Zen 2 to Zen 3 if not more. Even if Warhol Zen 3+ range gets cancelled from AMD's plans, Zen 3+ page needs to be there for Rembrandt. When Zen 4 launches, for the budget market more 6000 series processors will arrive. For all these, Zen 3+ should be considered successor for Zen 3 and have a new page while Zen 4 be successor for Zen 3+. — Preceding unsigned comment added by 2405:204:522A:8EB4:55F9:1637:5C07:E1DF (talk) 07:09, 29 May 2022 (UTC)Reply

Oppose – Zen 3+ is a slight architectural overhaul CPU-wise with 6 nm processing. It's good here as it is, but could use some expansion. --Zac67 (talk) 07:21, 29 May 2022 (UTC)Reply
Zen+ was an optimization on Zen with minor changes. However it's considered successor to Zen and predecessor to Zen 2. Why not follow the same with 3+? Makes it easy for users to follow through the generations amidst all the mid-cycle refreshes and naming conventions. 2405:204:522A:8EB4:55F9:1637:5C07:E1DF (talk) 08:35, 29 May 2022 (UTC)Reply
Zen+ was not a new generation, we don't need to repeat that foolish Wiki article. Zen+ is still gen1 Zen but served in 2nd generation Ryzen/Epyc.Zen3+ is still 3rd generation zen but serving in 6th generation Ryzen.--Denniss (talk) 10:53, 29 May 2022 (UTC)Reply
No, it shouldn't. It's more of an update/refresh, than a significant new architecture. That Zen+ article could be merged into Zen 1st gen article with how little new/different info it has from Zen 1st gen.
It could, however, be added as a "Zen 3+" section on the Zen 3 article, much like how Raptor Cove doesn't have its own article, instead it is a section on the Golden Cove article. I feel like we should do that with Zen 3+.
The Zen 3 article currently notes Zen 3+ as being simply Zen 3 built on 6nm, but from what I've seen in marketing slides and some articles I've read, it's a bit more than that. It has quite a number of architectural power efficiency improvements, resulting in a good performance-per-watt increase over Zen 3, even though the performance increase isn't all that impressive. Example source. — AP 499D25 (talk) 04:14, 21 April 2023 (UTC)Reply

V-cache size confusion

edit

@Zac67 In response to this edit:

I changed it to 96MB because it states "per CCD". So each CCD that contains V-Cache contains a total of 96MB of L3 cache, not 64.

If we're going to write it as 64MB then it should be made much clearer that it is additional 64MB on top of the standard 32, e.g. make it so it says "(+64 MB with 3D V-cache)" instead of the "64 MB per 3D V-Cache CCD" that it says right now.

On another note, the standard 32MB cache size states "per CCX" but the V-Cache size states "per CCD". I think we should only use either one or the other here, not both, for consistency. I'd say "per CCX" since CCX refers to the logical layout of the chip, and CCD refers to a physical chiplet; and we're talking logical arrangement of the chip here regarding the stacked cache.

Hope that makes sense. — AP 499D25 (talk) 14:03, 21 August 2023 (UTC)Reply

@AP 499D25: Hmm – for me, the 32 MB are on the CCD and the 64 MB V-Cache die stacked on top is not part of that CCD. But I may be wrong. How about something along "96 MB per CCD with 3D V-Cache die"? I think calling it "3D V-Cache CCD" confuses readers. --Zac67 (talk) 20:01, 21 August 2023 (UTC)Reply
I see where you are coming from. Indeed, "3D V-Cache CCD" looks a bit ambiguous, "96 MB per CCD with 3D V-Cache" is how I'd write it. Maybe the "die" at the end is necessary, but I don't want it to be too long, since this is infobox content anyway.
For what it's worth, I skimmed through several articles covering the V-Cache technology from AMD, AMD themselves call the cache on top of the CCD a "cache die".[1] However, I have counted three sources that call it a layer,[2][3][4] one that calls it simply cache,[5] and three that call it a chip or chiplet.[6][7][8] None of them explicitly call the 64MB stacked cache layer a "CCD". — AP 499D25 (talk) 10:42, 24 September 2023 (UTC)Reply

References