[Techpowerup] AMD "Zen" CPU Prototypes Tested, "Meet all Expectations"

[JDG1980]

It's hard to imagine that Zen will have IPC any lower than Sandy Bridge. Remember, that was a 2011 design made on 32nm, and even given the long lead time of CPU architectures, AMD had to have known going in that their new design needed to be at least that good, if not better. Most likely, we'll see IPC somewhere in the Sandy Bridge to Haswell range. The official claim of 40% IPC improvements over Excavator (which equates to ~61% better than Piledriver) would also put the result slightly above SB (which I calculate as having 58% better IPC than Piledriver, based on comparing the FX-8300 to the equivalently clocked i7-2700K on five ST benchmarks on this site).

The real wild card is clock speed. Zen will be expected to make most of its money in the server market, where core counts and perf/watt are more important than raw GHz. I wouldn't be surprised to see the flagship 8-core Zen sporting a base clock speed of 3.00 GHz, with higher turbos allowed when not all cores are in use. Keep in mind that the 8-core Haswell HEDT chip (i7-5960X) also has a 3.00 GHz base clock. Some enthusiast-focused motherboards will probably be built for a thermal design limit of over 95W, which will allow overclocking, perhaps to near 4.00 GHz.

A hypothetical Zen HEDT chip with 8 cores @ 3.00 GHz (OC to ~4.00 GHz) and IPC in the SB/IB range will probably need to have a maximum price tag of $499 to be able to take sales from Intel's HEDT lineup. They are expected to be released in H2 2016, and that means they'll have to compete with Broadwell-E. Server parts will sell at a discount to Intel's parts because they will not be quite up to par with Intel's best products, but they will at least be in the same ballpark and should be able to win back lost market share at much better margins than AMD is currently getting with their absurdly bad construction core lineup.

 

[nismotigerwvu]

K8 was only better compared to P4. Not to P-M.

And K5/K6 had higher IPC than P5 (but didn't clock high enough to take the absolute performance crown, especially so for K5) and K7 had higher IPC than P6/early Netburst. That's quite a long run if you're looking from the launch of K5 through the launch of Yonah. I know you're default response will be that the K5 wasn't performance competitive with the Pentium models of the time. That's true, it wasn't, but you're only comparing IPC and also claiming that Yonah was better than K8, which in this case, yes it had higher IPC but would have been absolutely smashed in absolute performance since you are comparing low TDP mobile chips to full on desktop class ones.

 

[ShintaiDK]

And K5/K6 had higher IPC than P5 (but didn't clock high enough to take the absolute performance crown, especially so for K5) and K7 had higher IPC than P6/early Netburst. That's quite a long run if you're looking from the launch of K5 through the launch of Yonah. I know you're default response will be that the K5 wasn't performance competitive with the Pentium models of the time. That's true, it wasn't, but you're only comparing IPC and also claiming that Yonah was better than K8, which in this case, yes it had higher IPC but would have been absolutely smashed in absolute performance since you are comparing low TDP mobile chips to full on desktop class ones.

I am not so sure about that. The Performance rating was outright disasterous. It took K6 to actually compete with Pentiums.

 

[looncraz]

The official claim of 40% IPC improvements over Excavator (which equates to ~61% better than Piledriver) would also put the result slightly above SB (which I calculate as having 58% better IPC than Piledriver, based on comparing the FX-8300 to the equivalently clocked i7-2700K on five ST benchmarks on this site).

Your calculation is very flawed. That web-site doesn't use fixed clockrates and you can't determine the clockrates used for any given test.

Haswell has almost exactly 40% higher IPC over Penryn, and Excavator's average IPC is almost exactly the same as Penryn. It's hard to compare architectures this far apart, but Intel's claims alone illustrate the point while also showing that all of my benchmarks are representative.

Zen, in theory, should almost exactly match Haswell, if the 40% IPC increase is a 40% performance increase. That, actually, is not a certainty. 40% more instructions retired in a certain time window doesn't always equate to 40% more performance - it depends on where the bottlenecks are in the code.

A slow fdiv, for example could kill most gains for some program, whereas another program may love the changes and see a doubling of performance.


EDIT:

Also, on a side note, I discovered something in the Zen patch I didn't recognize before: both AGUs appear to be for load/store, and are not dedicated to one or the other.

I'm mapping out all the instruction assignments and will post my results (in a spreadsheet).

Well, I did it, LOL!

http://looncraz.net/ZenAssignments.html

 

[TechGod123]

Your calculation is very flawed. That web-site doesn't use fixed clockrates and you can't determine the clockrates used for any given test.

Haswell has almost exactly 40% higher IPC over Penryn, and Excavator's average IPC is almost exactly the same as Penryn. It's hard to compare architectures this far apart, but Intel's claims alone illustrate the point while also showing that all of my benchmarks are representative.

Zen, in theory, should almost exactly match Haswell, if the 40% IPC increase is a 40% performance increase. That, actually, is not a certainty. 40% more instructions retired in a certain time window doesn't always equate to 40% more performance - it depends on where the bottlenecks are in the code.

A slow fdiv, for example could kill most gains for some program, whereas another program may love the changes and see a doubling of performance.


EDIT:

Also, on a side note, I discovered something in the Zen patch I didn't recognize before: both AGUs appear to be for load/store, and are not dedicated to one or the other.

I'm mapping out all the instruction assignments and will post my results (in a spreadsheet).

Well, I did it, LOL!

http://looncraz.net/ZenAssignments.html

What exactly does that mean? Is that a good thing?

 

[el etro]

What exactly does that mean? Is that a good thing?

The 2 "full"(load and store) AGUs says it is a good thing.

 

[sm625]

Its legitimately disappointing to see how many people think its gonna be a bulldozer attempt. I expected more objectivity from the members of this forum.

Because the focus seems to be on bulldozer's module design. The problem is that even if you take an 8 core bulldozer and turn off half of each module, it is still slow as dirt compared to pretty much every i5. An i5-6600k scores almost 8000 in passmark. The aforementioned bulldozer configured as 4 "real" cores only scores around 4500 in passmark. This is with all the modular design constraints removed. So even putting aside the module vs SMT design issues, AMD still needs to pull roughly 80% more performance out of their hat. How?

 

[AtenRa]

Because the focus seems to be on bulldozer's module design. The problem is that even if you take an 8 core bulldozer and turn off half of each module, it is still slow as dirt compared to pretty much every i5. An i5-6600k scores almost 8000 in passmark. The aforementioned bulldozer configured as 4 "real" cores only scores around 4500 in passmark. This is with all the modular design constraints removed. So even putting aside the module vs SMT design issues, AMD still needs to pull roughly 80% more performance out of their hat. How?

ZEN architecture is not based on Bulldozer CMT but in SMT like Intels. Also each ZEN Core has 50% more execution resources (4x ALUs + 2x AGUs vs 2+2) than what Bulldozer Integer Core has. How do you believe they reached 40% higher IPC than Excavator ??
The problem is not in the IPC, but if the architecture and the 14nm LPP process can achieve 4GHz or above at the designated TDP.
Personally im expecting AMD not aiming for the absolute top performance against Intel but mostly on Perf/Watt. Im expecting ZEN to have lower absolute performance than Desktop Skylake but it may have better Perf/watt.

 

[Arachnotronic]

ZEN architecture is not based on Bulldozer CMT but in SMT like Intels. Also each ZEN Core has 50% more execution resources (4x ALUs + 2x AGUs vs 2+2) than what Bulldozer Integer Core has. How do you believe they reached 40% higher IPC than Excavator ??
The problem is not in the IPC, but if the architecture and the 14nm LPP process can achieve 4GHz or above at the designated TDP.
Personally im expecting AMD not aiming for the absolute top performance against Intel but mostly on Perf/Watt. Im expecting ZEN to have lower absolute performance than Desktop Skylake but it may have better Perf/watt.

Wow. Intel has literally been spending billion$ on trying to eke out all of the performance/watt it can from each architecture and you think AMD is going to go from the lousy Bulldozer lineage to a new architecture built on a relatively shoestring budget on a foundry process optimized for mobile products and wind up with better perf/watt in key PC/server use cases? OK.

 

[LTC8K6]

ZEN architecture is not based on Bulldozer CMT but in SMT like Intels. Also each ZEN Core has 50% more execution resources (4x ALUs + 2x AGUs vs 2+2) than what Bulldozer Integer Core has. How do you believe they reached 40% higher IPC than Excavator ??
The problem is not in the IPC, but if the architecture and the 14nm LPP process can achieve 4GHz or above at the designated TDP.
Personally im expecting AMD not aiming for the absolute top performance against Intel but mostly on Perf/Watt. Im expecting ZEN to have lower absolute performance than Desktop Skylake but it may have better Perf/watt.

Why should we believe that they have reached 40% higher IPC than Excavator?
We don't even know what they mean by that.

Most likely, the 40% should be preceded by "up to".

Why should I believe that any more than I believe that Intel now has "3X the battery life" in mobile devices?

 

[Shehriazad]

Wow. Intel has literally been spending billion$ on trying to eke out all of the performance/watt it can from each architecture and you think AMD is going to go from the lousy Bulldozer lineage to a new architecture built on a relatively shoestring budget on a foundry process optimized for mobile products and wind up with better perf/watt in key PC/server use cases? OK.

Of course they will....after all the legend Jimmyboy himself designed Zen all on his own (he's legendary like that!) so there is no way it won't at least have +60% IPC and run @ 5 Ghz while only consuming 95(0)watt xD





p.s. We need a sarcasm textmarker.

 

[Arachnotronic]

Of course they will....after all the legend Jimmyboy himself designed Zen all on his own (he's legendary like that!) so there is no way it won't at least have +60% IPC and run @ 5 Ghz while only consuming 95(0)watt xD





p.s. We need a sarcasm textmarker.

:thumbsup:

 

[Fjodor2001]

Intel has literally been spending billion$ on trying to eke out all of the performance/watt it can from each architecture and you think AMD is going to go from the lousy Bulldozer lineage to a new architecture built on a relatively shoestring budget on a foundry process optimized for mobile products and wind up with better perf/watt in key PC/server use cases?

Intel has also spent billions of dollars on R&D for mobile, and how has that turned out for them? Spending huge amounts on R&D is no guarantee for success.

 

[ShintaiDK]

Spending huge amounts on R&D is no guarantee for success.

So is not spending R&D. You never get what you didn't pay for.

 

[dark zero]

R&D doesn't determine if they improve or not... Some companies does wonders with not so much R&D.

Another thing is that Intel is going to Apple path and starts to make their own SW incompatible or with heavy punish to the rest.

And that 40% I feel that only applies to ST... On MT I expect at best 15%

 

[ShintaiDK]

Some companies does wonders with not so much R&D.

Compared with their competition? Who?

 

[Arachnotronic]

Intel has also spent billions of dollars on R&D for mobile, and how has that turned out for them? Spending huge amounts on R&D is no guarantee for success.

No, but I don't know if you've noticed but Intel has been basically whipping AMD's behind for the last 10 years.

Intel literally puts its best people and virtually unlimited funds to defending its PC/server share and growing the TAMs in those areas. Do you really think the ragtag team of engineers that's left at AMD -- the ones that Apple, Samsung, Qualcomm, etc. didn't bother to poach -- are going to disrupt Intel's two largest most profitable businesses so easily?

 

[nemesismk2]

bull dozer was great and was quickly made into vishera which was great and still is. yes i own a amd fx 8350 and amd fx 4300.

 

[Fjodor2001]

No, but I don't know if you've noticed but Intel has been basically whipping AMD's behind for the last 10 years.

Intel literally puts its best people and virtually unlimited funds to defending its PC/server share and growing the TAMs in those areas.

And all those billions of dollars are giving us measly 5% yearly performance increases. Says something about Intels' R&D inefficiency if anything. Look at Apple instead if you want to see good results per R&D dollar spent.

Intel is good as marketing and making the most of their monopoly-like market position though.

 

[Arachnotronic]

And all those billions of dollars are giving us measly 5% yearly performance increases. Says something about Intels' R&D inefficiency if anything. Look at Apple instead if you want to see good results per R&D dollar spent.

Whatever you say

Can't wait for Zen to arrive. I was on these forums when Derpdozer was being hyped to the moon and I will be sure not to miss the discussion when Zen ultimately turns out to be a disappointment.

 

[Fjodor2001]

Compared with their competition? Who?

Just to name a few: Skype, Spotify, Mojang, King, ARM, Apple, Qualcomm. The first four have with small R&D resources produced hugely valuable companies that later have been sold for tens of billions of dollars. And the last three were small companies too once and with small R&D budgets, but now they are very successful multi billion dollar companies.

There are lots of other similar examples.

 

[Fjodor2001]

Can't wait for Zen to arrive. I was on these forums when Derpdozer was being hyped to the moon and I will be sure not to miss the discussion when Zen ultimately turns out to be a disappointment.

Assuming Zen is not successful, what will you and the consumers gain from that since you are longing so much for that to happen?

Are you saying the consumers are better off with a monopoly after all? I.e. ShintaiDK style?

 

[ShintaiDK]

Just to name a few: Skype, Spotify, Mojang, King, ARM, Apple, Qualcomm. The first four have with small R&D resources produced hugely valuable companies that later have been sold for tens of billions of dollars. And the last three were small companies too once and with small R&D budgets, but now they are very successful multi billion dollar companies.

There are lots of other similar examples.

You obviously didn't understand the question at all. Or you just didn't want to because it would ruin your Zen hype.

 

[Fjodor2001]

You obviously didn't understand the question at all.

You obviously didn't understand my answer at all. Or you're trying to spin your original question into something else:

Some companies does wonders with not so much R&D.

Compared with their competition? Who?

My answer clearly responded to that and proved your claim to be incorrect. So how would you like to re-spin your question now?

 

[Sweepr]

Are you saying the consumers are better off with a monopoly after all? I.e. ShintaiDK style?

Can you please point out what this has to do with this thread, or his post, because honestly I don't get it. Why do you change subject every now and then only to talk about Shintai?