New Zen microarchitecture details

Page 33 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

PPB

Golden Member
Jul 5, 2013
1,118
168
106
You were doing great, and then in came the meta commentary. You destroyed your own posts' credibility.

The only company you should get emotional about is the one you work for.

No one is getting emotional. Laughter counts as an emotion? Then probably I do get emotional reading your "posts" :D

Still nothing to prove Zen will fail? How.... unexpected I see.
 

unicutesx

Junior Member
Apr 5, 2016
3
0
0
Question: is it possible that AMD sends ES with different clocks?
I saw that report from bitsandchips about Zen ES having 3GHz but we're supposed to get Summit Ridge ES next week and the info AMD sent us last month suggests it's a lower clock.
 

cytg111

Lifer
Mar 17, 2008
23,174
12,835
136
Question: is it possible that AMD sends ES with different clocks?
I saw that report from bitsandchips about Zen ES having 3GHz but we're supposed to get Summit Ridge ES next week and the info AMD sent us last month suggests it's a lower clock.

Sure and unlocked.
 

DrMrLordX

Lifer
Apr 27, 2000
21,620
10,828
136
I'm afraid your data is incorrect.

http://www.pcper.com/reviews/Proces...-Rounds-Out-Kaveri-Line/Results-Cinebench-R15

in CBR15, A6-7400K which is 1M2C, has only 70% additional performance(87 vs 149) when multithreaded. Not to mentioned 7400k has higher turbo of up to 3.9Ghz in single thread when there's almost none in MT. I think it just has roughly 60-65% yield which means almost 35% penalty.

Eh? I've poked around with my own SR (2m/4t 7700k) and found scaling to be much better than that. Recently I did some Geekbench 3 and got scaling of 78%. CBR10 scaled better than that if I recall correctly.

40% average throughput increase would put Zen in Skylake territory. Don't go there, it's a house made of treacherous candy.

I still stand by my statement. If Zen's only 40% faster than XV in ST IPC then it'll be ass unless they have the world's best SMT implementation. A core that wide should be quite a bit faster in fp workloads, not to speak of all the improvements in the cache subsystem and the front end.

And didn't we put Cinebench to rest, after Agner demonstrated that it's little more than an intel marketing benchmark?

I find R10 to be very kind to AMD processors. Later versions, maybe not so much. 11.5 is quite bad actually.
 
Aug 11, 2008
10,451
642
126
No one is getting emotional. Laughter counts as an emotion? Then probably I do get emotional reading your "posts" :D

Still nothing to prove Zen will fail? How.... unexpected I see.

You mean like you dont have any proof that it will be a success either? Pretty hard to prove anything about a product that isnt available for testing. And I dont think the poster you are feuding with ever claimed to have proof Zen will be a failure. He is just speculating like everyone else in the myriad of Zen threads we have had so far. Nice way to discredit him though, asking for "proof" of something that is obviously unprovable.
 

Dresdenboy

Golden Member
Jul 28, 2003
1,730
554
136
citavia.blog.de
How slow zen will be:
We start with fx8350 base of 6.85 CB11.5 MT multiply it by XV IPC increase 1.1 and announced Zen improvements of 1.4 and we have 10.5 CB11.5 MT score which is more than i7 6700k.

Let me do the calc for you.
According to this test, CZ vs PD in CB11.5 is +20% ST, +26% MT at the same clock. Mem is slow, so there is an additional error in the calculation (aside from unknowns like exact IPC gain in CB and power consumption - and the clock frequency of course).
So I set
XV_IPC_INCREASE = 1.2 for ST, 1.26 for MT
ZEN_IPC_INCREASE = 1.4 (actually not known)
CMT_EFFICIENCY = 0.85 (PD's value for CB)
SMT_EFFICIENCY = 1.2 (should be doable, but not known)
and use Zen_CB_est_MT = fx8350 base * XV_IPC_INCREASE * ZEN_IPC_INCREASE / CMT_EFFICIENCY * SMT_EFFICIENCY for MT scores

Zen_CB_est_MT = 6.85 * 1.2 * 1.4 / 0.85 * 1.2 = 16.2, if it would run at the same (turbo) clock frequency.

Now I'm lost.
So if 40% increase in Zen IPC is core (2 threads) in relation to a 1 thread of XV?
Based on this 2 core/4 threadZen CPU will have CB11.5 MT perf around 2.8 compared to 3.39 of a8-7600. So, 2C/4T Zen will be 15% slower than 2 module XV in MT tasks?
40% IPC increase given as throughput increase is uncommon, I think. And comparing cores to modules is nonsense, as said before.
 

CHADBOGA

Platinum Member
Mar 31, 2009
2,135
832
136
I am putting out my expectations for key performance characteristics of Zen. Why does the fact that I'm not hyping Zen to the moon bother you so much?

I get that some "computer nerds" want to see AMD do really well again, but I think it's important to manage expectations based on reality, not based on fantasy and what one "hopes" or "wishes" will happen.

There were people around these parts and basically across the entire internet who were pumping Bulldozer as a Sandy Bridge killer, and there were a few lone voices of reason who cautioned that maybe it wouldn't be the big deal that some thought. Those people were bullied then and they are being bullied now on the eve of Zen's launch.

Fortunately, I have a pretty thick skin and will not be bullied into silence.
Then please start posting at iHub again.

Flummox's diarrhoea of the mouth needs wiping. ;)
 

Abwx

Lifer
Apr 2, 2011
10,937
3,440
136
Zen_CB_est_MT = 6.85 * 1.2 * 1.4 / 0.85 * 1.2 = 16.2, if it would run at the same (turbo) clock frequency.


40% IPC increase given as throughput increase is uncommon, I think. And comparing cores to modules is nonsense, as said before.

Actualy it depend of the comparison metric...


At 4GHz the FX score 6.92, that s 0.4325/GHz/module whenever EXV score 0.525/GHz/module, that s 21% better throughput/Hz per module.

Now Zen use one such FPU/core, so it should score like a whole EXV module, that is 4.2/GHz for a 8C, and this of course include SMT since that s the total throughput of the FPU.
 

CHADBOGA

Platinum Member
Mar 31, 2009
2,135
832
136
Well at the very least I hope that Zen can prove to be a real contender against the mid-range i5's.(And if it can be more than that, awesome)

When I look at Intel's i5 pricing, only the i5-K makes any sense to me.
 

MajinCry

Platinum Member
Jul 28, 2015
2,495
571
136
Well at the very least I hope that Zen can prove to be a real contender against the mid-range i5's.(And if it can be more than that, awesome)

When I look at Intel's i5 pricing, only the i5-K makes any sense to me.

Hell, even then, the i5 ks only made sense back in the days of Sandybridge and Ivybridge. A Skylake i5 now costs, what, £175? £180? The i5 2500k used to be £140.

Though you can now get an i5 2500k for £90 off ebay. Fairly tempting, if it weren't for the fact that the motherboards are bloody old 'n' expensive.
 
Aug 11, 2008
10,451
642
126
News flash, prices go up. Desktop i5 prices actually seem quite reasonable to me. If anything is overpriced, it is the i7 where it basically costs a hundred dollars or more to have hyperthreading turned on.
 

RussianSensation

Elite Member
Sep 5, 2003
19,458
765
126
News flash, prices go up. Desktop i5 prices actually seem quite reasonable to me. If anything is overpriced, it is the i7 where it basically costs a hundred dollars or more to have hyperthreading turned on.

MicroCenter has i5-6600K for $209.99 vs. i7 6700K for $339.99. If anything, glorified quad i7 is overpriced and it's obvious when comparing it to i7 5820K or upcoming i7 6800k. The issue is over the next 5 years, basically how long modern CPUs last, there will be plenty of cases where the i5 will bomb. Because the probably of that happening against an i7 is 100%, that's why people will keep paying $100-130 more to get the i7. However, if Zen can deliver 4 core HT for $209 or 8-core HT for $349, well then the pressure on Intel to release mainstream Icelake with 6-cores in 2018 will become stronger than ever. It is a pretty big rip-off to ask $349 for a 122mm2 die. Just wait until GP104/Vega drop and compare the level of die size, PCB/memory/cooling, etc. The main reason people keep paying these prices is because Intel CPUs last so long (total cost of ownership per year is super low), and because the next generation Intel CPUs barely improve performance year over year. This looks like it'll slow down even more as Intel will now release a true next generation architecture only every 3 years after abandoning Tick-Tock. That means Skylake 2015 -> Icelake 2018 -> 2021. That realistically means for Sandy owners who skipped Haswell and got Skylake, they won't upgrade until 2021. That's why it's really important for Zen to start off with a solid fundamental base since they'll have another 2 years or so for Zen+ to prepare for Icelake.
 
Last edited:

ElFenix

Elite Member
Super Moderator
Mar 20, 2000
102,414
8,356
126
You mean like you dont have any proof that it will be a success either? Pretty hard to prove anything about a product that isnt available for testing. And I dont think the poster you are feuding with ever claimed to have proof Zen will be a failure. He is just speculating like everyone else in the myriad of Zen threads we have had so far. Nice way to discredit him though, asking for "proof" of something that is obviously unprovable.

everyone? no. some are actually coming with specific information. not many, granted, but some.

unfortunately most of this thread is filled with either blue sky or fud. and those people should be discredited.
 

myocardia

Diamond Member
Jun 21, 2003
9,291
30
91
I don't know what Erenhardt was estimating earlier, but for the purpose of this dirty guesstimate, please keep in mind Zen 8c/16t @ 3Ghz will likely have double the throughput of 4 construction modules (8t) at same frequency.

Maybe not quite 100% more, but yeah I agree, a very large amount more throughput, at equal frequencies. I'm actually hoping for the 8C/16T Zens to have 2.5-2.6 Ghz base frequency, with turbo boost to the 3.0 Ghz area, if not ~3.2 Ghz for 1 or 2 cores. If it has those kinds of speeds, at roughly 100 watt TDP and somewhere between Sandy Bridge and Ivy Bridge single threaded performance, it will be absolutely perfect for video encoding/transcoding, and that is why I'm interested in Zen.
 

jhu

Lifer
Oct 10, 1999
11,918
9
81
Maybe not quite 100% more, but yeah I agree, a very large amount more throughput, at equal frequencies. I'm actually hoping for the 8C/16T Zens to have 2.5-2.6 Ghz base frequency, with turbo boost to the 3.0 Ghz area, if not ~3.2 Ghz for 1 or 2 cores. If it has those kinds of speeds, at roughly 100 watt TDP and somewhere between Sandy Bridge and Ivy Bridge single threaded performance, it will be absolutely perfect for video encoding/transcoding, and that is why I'm interested in Zen.

Or you could get 2 of those cheap 8 core Sandy Bridge Xeons from eBay right now.
 

myocardia

Diamond Member
Jun 21, 2003
9,291
30
91
Or you could get 2 of those cheap 8 core Sandy Bridge Xeons from eBay right now.

I don't live at the north pole, I live in Texas. That's the reason I'm more than willing to pay extra, to avoid extra heat. We run our air conditioners 9-10 months a year here. It's more or less the equivalent of living in Phoenix, but a couple of degrees cooler around the clock.
 

Azuma Hazuki

Golden Member
Jun 18, 2012
1,532
866
131
Or we can wait for the sodding benchmarks...proof of the pudding in the eating and all that.
 

Dresdenboy

Golden Member
Jul 28, 2003
1,730
554
136
citavia.blog.de
Actualy it depend of the comparison metric...
Your numbers would suggest ~17pts for 4GHz, pretty close to my number, although I assumed simply the Vishera clocks (whatever they were at ST/MT).

What I said about core vs. module was just an answer to the CMT module to SMT core comparisons mentioned by Erenhardt. This way we could see the module as a core (like AMD's engineers did) and see any construction core at half the amount of cores plus CMT. So Vishera would be a 4C again and XV just a dual core like an i3. Then of course a 4C Zen wouldn't scale that nicely.
 

Sheep221

Golden Member
Oct 28, 2012
1,843
27
81
News flash, prices go up. Desktop i5 prices actually seem quite reasonable to me. If anything is overpriced, it is the i7 where it basically costs a hundred dollars or more to have hyperthreading turned on.
It's true that it is abit overpriced but where you really need the 8 threads, various scientific/rendering/simulation applications benefit greatly from more threads rather than IPC on one physical core, so in those scenarios you have to pay the premium, where the extra cost will return quite quickly.
I don't live at the north pole, I live in Texas. That's the reason I'm more than willing to pay extra, to avoid extra heat. We run our air conditioners 9-10 months a year here. It's more or less the equivalent of living in Phoenix, but a couple of degrees cooler around the clock.
SBs are not getting that toasty, and if yes they are much easier to cool down.
 

myocardia

Diamond Member
Jun 21, 2003
9,291
30
91
SBs are not getting that toasty, and if yes they are much easier to cool down.

I'd hardly call an 8 cored, 32nm CPU "cool". I believe Intel rates them @ 120 watts, as a matter of fact. Okay, they claim 115 watts, which I doubt very much is even one watt less than 120, which is awesome for a 2.6 Ghz, 16-threaded, 32nm CPU. BTW, how easy are Zens to cool down, compared to 16-threaded Sandy Bridges?

Oh, and no matter how easy or hard a particular CPU happens to be to keep cool, makes zero difference in how much they heat the air in the room in which they reside. See, since computers do zero actual work (they move nothing, at any rate,for any distance), 100% of the watts that they use, not including spinning fans or spinning disks, is put into the room as heat. When you don't live up near the north pole, having one room of your house 10 or 15°F warmer than the others all summer long is something that you try to avoid, even if it costs you a bit of extra money when buying components.
 

Doom2pro

Senior member
Apr 2, 2016
587
619
106
I'd hardly call an 8 cored, 32nm CPU "cool". I believe Intel rates them @ 120 watts, as a matter of fact. Okay, they claim 115 watts, which I doubt very much is even one watt less than 120, which is awesome for a 2.6 Ghz, 16-threaded, 32nm CPU. BTW, how easy are Zens to cool down, compared to 16-threaded Sandy Bridges?

Oh, and no matter how easy or hard a particular CPU happens to be to keep cool, makes zero difference in how much they heat the air in the room in which they reside. See, since computers do zero actual work (they move nothing, at any rate,for any distance), 100% of the watts that they use, not including spinning fans or spinning disks, is put into the room as heat. When you don't live up near the north pole, having one room of your house 10 or 15°F warmer than the others all summer long is something that you try to avoid, even if it costs you a bit of extra money when buying components.

In any house heating their home other than with electricity in the winter, a computer putting out X watts of heat is X watts of heat not needed to heat with Propane, Natural Gas, Fuel Oil, etc...

In the summer it swings around the other way, every Watt is more thermal energy needed to be pumped out by Air Conditioning.
 

jhu

Lifer
Oct 10, 1999
11,918
9
81
I'd hardly call an 8 cored, 32nm CPU "cool". I believe Intel rates them @ 120 watts, as a matter of fact. Okay, they claim 115 watts, which I doubt very much is even one watt less than 120, which is awesome for a 2.6 Ghz, 16-threaded, 32nm CPU. BTW, how easy are Zens to cool down, compared to 16-threaded Sandy Bridges?

Oh, and no matter how easy or hard a particular CPU happens to be to keep cool, makes zero difference in how much they heat the air in the room in which they reside. See, since computers do zero actual work (they move nothing, at any rate,for any distance), 100% of the watts that they use, not including spinning fans or spinning disks, is put into the room as heat. When you don't live up near the north pole, having one room of your house 10 or 15°F warmer than the others all summer long is something that you try to avoid, even if it costs you a bit of extra money when buying components.

People keep mentioning 140W for 8 core Zen. I'm not sure where this comes from.
 

Dresdenboy

Golden Member
Jul 28, 2003
1,730
554
136
citavia.blog.de
People keep mentioning 140W for 8 core Zen. I'm not sure where this comes from.
Me too. I've never seen a mention on any HEDT related (leaked) slide, or heard so in an interview.

I've only seen 95W for SR so far

Also wouldn't 140W be worse for OC'ing?
 

coercitiv

Diamond Member
Jan 24, 2014
6,187
11,855
136
People keep mentioning 140W for 8 core Zen. I'm not sure where this comes from.
Socket AM4 leaks, it's said to support 140W+ TDP. (hardly as relevant as the mentioned 95W TDP, but hey...)