New Zen microarchitecture details

Page 50 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Dresdenboy

Golden Member
Jul 28, 2003
1,730
554
136
citavia.blog.de
How about something Zen related?
NUKDu9S.jpg

More here: https://www.reddit.com/r/Amd/comments/4k8lha/new_q2_amd_investor_presentation_includes_vague/
 

mikk

Diamond Member
May 15, 2012
4,112
2,108
136
Doesn't say much without info about the exact Orochi SKU in this comparison.
 

guskline

Diamond Member
Apr 17, 2006
5,338
476
126
It would be good news to get some solid updates as to where Zen stands in June 2016. I would hope there is enough silicon out there with testers to give us a peek at what might be in store.
 

ElFenix

Elite Member
Super Moderator
Mar 20, 2000
102,414
8,356
126
Y-axis isn't labeled either.

it's 4 and 2, obviously, and 4 is way better than 2. you might say, it's better by 2. which, unless you subscribe to common core, is 2 better!

i think it'd be materially misleading if it's not double at *something*
 
Mar 10, 2006
11,715
2,012
126
it's 4 and 2, obviously, and 4 is way better than 2. you might say, it's better by 2. which, unless you subscribe to common core, is 2 better!

i think it'd be materially misleading if it's not double at *something*

Can't be materially misleading if it's not labeled, but for now I assume 2x multithreaded performance.

Much, much better than what AMD is currently selling into the market.
 

dark zero

Platinum Member
Jun 2, 2015
2,655
138
106
Actually could be:
- 40% ST perf. Improvement. SMT, better cache, die shrink and new instructions will boost them.
- Up to 2X MT perf. Improvement if we compare the Octacores... since now it will be 8 threads vs 16. In fact expecting a 70% MT improvement
 

mikk

Diamond Member
May 15, 2012
4,112
2,108
136
Assuming this is compared to an FX-8150 (CB R15 ~550), the score for Zen is roughly 1100. This is comparable to an Haswell-E based sixcore i7-5820K. But considering Zen has a 2 core advantage it would possibly mean that there is still quite a big gap in Singlethread performance to Intel.
 

Phynaz

Lifer
Mar 13, 2006
10,140
819
126
it's 4 and 2, obviously, and 4 is way better than 2. you might say, it's better by 2. which, unless you subscribe to common core, is 2 better!

i think it'd be materially misleading if it's not double at *something*

Assuming the y-axis starts at zero. But since it's unlabled it could be 10.2 and 10.4.
 

Exophase

Diamond Member
Apr 19, 2012
4,439
9
81
Assuming the y-axis starts at zero. But since it's unlabled it could be 10.2 and 10.4.

Since the other graph on the same page is labeled that way I think it's pretty safe to assume that it also starts at 0. I don't see why else they'd label one and not the other. I refuse to believe that they think they'd get away with trying to pull some nonsense arbitrary comparison this way.
 

ElFenix

Elite Member
Super Moderator
Mar 20, 2000
102,414
8,356
126
Assuming the y-axis starts at zero. But since it's unlabled it could be 10.2 and 10.4.

and i think it would be materially misleading not to be. then again, it's been a long time since i've taken securities law.



Can't be materially misleading if it's not labeled, but for now I assume 2x multithreaded performance.

Much, much better than what AMD is currently selling into the market.

source? that's not my understanding of 10b-5. and, keep in mind, i think it means it's double at *something* performance oriented, but i don't know what that is.


Since the other graph on the same page is labeled that way I think it's pretty safe to assume that it also starts at 0. I don't see why else they'd label one and not the other. I refuse to believe that they think they'd get away with trying to pull some nonsense arbitrary comparison this way.
i don't think it's a safe assumption to make just from one being labelled and the other not. esp. considering that one has 5 lines and the other has 6.
 
Last edited:
Mar 10, 2006
11,715
2,012
126
Since the other graph on the same page is labeled that way I think it's pretty safe to assume that it also starts at 0. I don't see why else they'd label one and not the other. I refuse to believe that they think they'd get away with trying to pull some nonsense arbitrary comparison this way.

Nope, they very clearly took the effort to label the Carrizo slide but intentionally signaled a separation between the two and did not label the second.
 
Aug 11, 2008
10,451
642
126
Assuming this is compared to an FX-8150 (CB R15 ~550), the score for Zen is roughly 1100. This is comparable to an Haswell-E based sixcore i7-5820K. But considering Zen has a 2 core advantage it would possibly mean that there is still quite a big gap in Singlethread performance to Intel.

I have already said it, but this is pretty much what I am expecting: 8 core Zen to compete with 6 core intel HEDT, maybe slightly better in highly threaded workloads, and trailing somewhat in single or lightly threaded. So then it comes down to overclocking and price.

Lets just hope BW-E gives some improvement over HW-E. My feeling is it will be a crap over-clocker and given the small ipc gain might not be faster than HW-E at all, and could even be a slight regression.
 

krumme

Diamond Member
Oct 9, 2009
5,952
1,585
136
Assuming this is compared to an FX-8150 (CB R15 ~550), the score for Zen is roughly 1100. This is comparable to an Haswell-E based sixcore i7-5820K. But considering Zen has a 2 core advantage it would possibly mean that there is still quite a big gap in Singlethread performance to Intel.
Well. Even if cb is a good measure of performance doing it at eg. 95w instead of 140w should make for a difference.

Actually i find anything above 100w in a desktop a little cheasy for 2016. The new i5 quad cores easily keep 45w tdp.

Besides with dx12 comming its a question if the ipc level is not fine as it is.

If zen is good i have no doubt skylake 8c will hit well before half a year and with low tdp variant available. If we look at the small die sizes intel have there is plenty opportunity. The fix cost is sunk. Variable cost is low. What matters is keeping share and get revenue for future investments. Same goes for amd/gf.

Intel big cores are fast elegant and lean. Zen is perhaps a bit slower but more cost effecttive. It could be a fantastic fight. No weak bd or p4 or what ever slow arch amd had in the 90ties. No.

I think we as consumers are in for the best year ever.
 
Mar 10, 2006
11,715
2,012
126
Or they felt that one label was sufficient for both graphs.

Negative, the lines in the first don't match up with the second and they make sure to stick a dotted line in between the two graphs. This is a very intentional omission and if you bought AMD stock on the expectation that this was the case, they would tell you "no, we never said that, you just inferred it."

Anyway, so the only "statements" that AMD made to investors are the following:

1. "Coming to high performance desktops"
2. "Significant performance leap expected"

Accompanying this is a clearly separated, y-axis unlabeled chart that seems to imply that Summit Ridge is 2x Orochi but is never explicitly stated .

As long as statements (1) and (2) are not false: meaning that Zen is in fact at the time that the statement was made "coming to high performance desktops" and as long as AMD internally expects, at the time that the statement was issued to investors, a "significant performance leap" then they are legally free and clear.

And of course they cover their behinds further with the disclaimer slide right there at the end :)
 
Mar 10, 2006
11,715
2,012
126
I have already said it, but this is pretty much what I am expecting: 8 core Zen to compete with 6 core intel HEDT, maybe slightly better in highly threaded workloads, and trailing somewhat in single or lightly threaded. So then it comes down to overclocking and price.

Seems reasonable to me.

Lets just hope BW-E gives some improvement over HW-E. My feeling is it will be a crap over-clocker and given the small ipc gain might not be faster than HW-E at all, and could even be a slight regression.

Intel gave BDW-E samples to reviewers a while ago and the embargo should lift soon.