New Zen microarchitecture details

Page 153 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

raghu78

Diamond Member
Aug 23, 2012
4,093
1,475
136
http://www.pcworld.com/article/3155...hitecture-is-expected-to-last-four-years.html

When asked how long Zen would last, compared to Intel’s two-year tick-tock cadence, Papermaster confirmed the four-year lifespan and tapped the table in front of him: “We’re not going tick-tock,” he said. “Zen is going to be tock, tock, tock.”


Papermaster confirms Zen will last for 4 years with significant enhancements on a annual cadence. 2017-2020 is going to be exciting. Zen is just the beginning. Intel will have a tougher time in 2018-2019 against Zen successors. The next tock from Intel is unlikely to make it before late 2019. Intel 10nm seems to be delayed by atleast 6 months. Cannonlake for desktops/notebooks will probably land in early 2019 and Icelake by late 2019.
 

CatMerc

Golden Member
Jul 16, 2016
1,114
1,153
136
You know, I always thought about that. If Zen reaches current Intel IPC, what does that mean for Zen+? Would that mean AMD actually >beating< Intel IPC? Even 10% over current Zen would be enough for that, and they probably have some low hanging fruits that didn't make it to Zen due to time or transistor budget reasons.

The next 4 years are going to be a tough time for Intel. Jim Keller's wizardry.
 

ShintaiDK

Lifer
Apr 22, 2012
20,378
145
106
See, AM4/Zen actually has a better PCIe/SATA config then LGA-115x, since it has 16+4(8+2 for BR) lanes directly from the CPU, and two SATA ports wired directly into the CPU as well. On 115x, everything either have to pass over the DMI link to the PCH, or you have to sacrifice 8 lanes from the graphics card.

That's my opinion anyway.

You can pick between 4 PCIe lanes or 2 PCIe and 2 SATA. You cant get 4 PCIe and 2 SATA.

So if we exclude graphics, for Summit Ridge you end up with either:
X370 config 1: 2 PCIe 3.0 lanes, 6 SATA and 8 PCIe 2.0
X370 config 2: 4 PCIe 3.0 lanes, 4 SATA and 8 PCIe 2.0

Compared:
Z170 20 PCIe 3.0 lanes and 6 SATA.
Z270 24 PCIe 3.0 lanes and 6 SATA.

X370 should simply have been PCIe 3.0 and more lanes. It should also have been a 28mn chipset instead of 55nm.
 

iBoMbY

Member
Nov 23, 2016
175
103
86
X370 itself is connected to Ryzen via PCIe 3.0 x4 - everything connected to X370 is sharing this bandwidth. IMHO they should've used something else to connect to a chipset - they'd probably 100 options to offer more bandwidth.
 

Insert_Nickname

Diamond Member
May 6, 2012
4,971
1,693
136
You can pick between 4 PCIe lanes or 2 PCIe and 2 SATA. You cant get 4 PCIe and 2 SATA.

Poor wording on my part. My point still stands. The AM4 platform is more flexible in that respect.

In my opinion.

So if we exclude graphics, for Summit Ridge you end up with either:
X370 config 1: 2 PCIe 3.0 lanes, 6 SATA and 8 PCIe 2.0
X370 config 2: 4 PCIe 3.0 lanes, 4 SATA and 8 PCIe 2.0

Compared:
Z170 20 PCIe 3.0 lanes and 6 SATA.
Z270 24 PCIe 3.0 lanes and 6 SATA.

X370 should simply have been PCIe 3.0 and more lanes. It should also have been a 28mn chipset instead of 55nm.

Everything other then the PCIe x16 slot for graphics on LGA-115x still has to go through the DMI 3.0 link to the PCH. PCIe, SATA, LAN etc. You can't cram the bandwidth requirement of 24 PCIe 3.0 lanes though a PCIe 3.0 x4 link. Simple as that.

Besides X370 is equal in link width to Zx70. Both use a PCIe 3.0 x4 link. I'd even go so far as to say AMD is again being even more flexible then Intel, since that PCIe 3.0 x4 link doesn't even have to be used for an FCH...

http://www.anandtech.com/show/10705...and-am4-analysis-a12-9800-b350-a320-chipset/4

Manufacturing process doesn't matter much for what is a desktop chipset. Mobile is already covered by BR/RR integrated FCH.
 

TheELF

Diamond Member
Dec 22, 2012
4,026
753
126
You know, I always thought about that. If Zen reaches current Intel IPC, what does that mean for Zen+? Would that mean AMD actually >beating< Intel IPC? Even 10% over current Zen would be enough for that, and they probably have some low hanging fruits that didn't make it to Zen due to time or transistor budget reasons.

The next 4 years are going to be a tough time for Intel. Jim Keller's wizardry.
Just because AMD found a couple of cases where zen performance matches broadwell performance does not mean that they matched the IPC.
It just means that you will get similar performance in some cases.
We will have to wait for single thread benches to figure out where zen ipc is placed.
 
Reactions: frozentundra123456

CatMerc

Golden Member
Jul 16, 2016
1,114
1,153
136
Just because AMD found a couple of cases where zen performance matches broadwell performance does not mean that they matched the IPC.
It just means that you will get similar performance in some cases.
We will have to wait for single thread benches to figure out where zen ipc is placed.
CPC's numbers imply that it's in more than a "couple" cases
 

USER8000

Golden Member
Jun 23, 2012
1,542
780
136
Just because AMD found a couple of cases where zen performance matches broadwell performance does not mean that they matched the IPC.
It just means that you will get similar performance in some cases.
We will have to wait for single thread benches to figure out where zen ipc is placed.

Apart from the CPC preview which seems to show something similar too.
 

DrMrLordX

Lifer
Apr 27, 2000
22,027
11,607
136
Bad memory...

No worries, though a lot of 845 buyers would have liked the 'k' on the end for OC flexibility.

Anyway they don't have higher clocks that corresponding APU parts, so they are harvested and probalby not very much quality silicon (aka: high leakage or low max OC)
eg: the 880k is 4/4.2 in 95W and there is an APU with 4/4.3 in 95W, GPU included...

Yeah they are harvested. Quality is all over the map. Just because the iGPU is fused off doesn't mean the rest of the chip is garbage though. Some people have very good 860ks for example, while others have some truly awful ones.

It seems like the earliest batches of the harvested chips had the best chances of doing well.
 

TheELF

Diamond Member
Dec 22, 2012
4,026
753
126
Canard showed ~15% below broadwell...
They stated that their sample is limited to 3.3Ghz and it's doubtfull that the stock broadwell part clocks more then that for all core turbo.
And if you compare to the i5-6500 that does 3.3Ghz all core turbo for sure it looks even worse, 12 threads more ( +300% ) only give you ~80% more performance,a 4/8 part at the same clocks would be ~10% slower then a skylake i5 and that's before you subtract the SMT benefit.

We really have to wait for the final product and comprehensive benchmark suites to find out where it stands.
 

AtenRa

Lifer
Feb 2, 2009
14,003
3,361
136
You can pick between 4 PCIe lanes or 2 PCIe and 2 SATA. You cant get 4 PCIe and 2 SATA.

So if we exclude graphics, for Summit Ridge you end up with either:
X370 config 1: 2 PCIe 3.0 lanes, 6 SATA and 8 PCIe 2.0
X370 config 2: 4 PCIe 3.0 lanes, 4 SATA and 8 PCIe 2.0

Compared:
Z170 20 PCIe 3.0 lanes and 6 SATA.
Z270 24 PCIe 3.0 lanes and 6 SATA.

X370 should simply have been PCIe 3.0 and more lanes. It should also have been a 28mn chipset instead of 55nm.

There are 2x SATA-6 AND 4x USB 3.1 directly on the CPU in ZEN.

That means they dont have to go through the 4x PCIe connecting the Chipset with the CPU.
It also means the rest of the 4x SATA-6 and the rest of the USB/M.2/U.2 etc connections found on the Chipset can use all the 4x PCIe Gen 3.0 Bandwidth from CPU to Chipset and back for themselves.


Intel SandyBridge/Kabylake CPUs doent have any SATA or USB in the CPU. That means that everything on the Z170/270 has to share the DMI Gen 3.0 Bandwidth to connect to the CPU.
That means that at some time you will be Bandwidth constrained because you only have 4x PCIe Gen 3.0 Bandwidth between the CPU and the Chipset (Z170/270) for all the SATA, USB, M.2, U.2 etc.

So AMDs X370 may have less PCIe lanes but it can use more devices simultaneously than Intel Z170/270 before it will be Bandwidth starved.






 
Last edited:
Reactions: Dresdenboy

iBoMbY

Member
Nov 23, 2016
175
103
86
Still, Kaby Fake (and its predecessors) is based on a years old platform, and from a brand new product in 2017 I would expect something better. What Ryzen/AM4 brings to the table I/O wise is enough for me personally, but only barely. All they would've needed was a faster link to the X370 - maybe just some improvised, preliminary, PCIe 4.0 x4 like connection (or maybe Infinity Fabric links), and they could offer a lot more.
 

BeepBeep2

Member
Dec 14, 2016
86
44
61
Canard showed ~15% below broadwell...
They stated that their sample is limited to 3.3Ghz and it's doubtfull that the stock broadwell part clocks more then that for all core turbo.
And if you compare to the i5-6500 that does 3.3Ghz all core turbo for sure it looks even worse, 12 threads more ( +300% ) only give you ~80% more performance,a 4/8 part at the same clocks would be ~10% slower then a skylake i5 and that's before you subtract the SMT benefit.

We really have to wait for the final product and comprehensive benchmark suites to find out where it stands.
Assuming IPC is the same as those tests, if Ryzen ships at 3.6 / 3.9 (or 4.0), that is ~15% higher than the CanardPC sample which would put performance around or ahead of the 6900K @ 3.2 / 3.7.

Their gaming tests also showed the i5 6500 5% slower than the 6600K with an 8-10% clock decrease and same core count, so it seems their gaming tests were very dependent on single thread performance. I would imagine in this case as well, that if you increase clocks on Ryzen 15-20% it will match or beat the 6900K (10%) in gaming according to their performance tests.

Even if AMD doesn't match Intel on IPC, if they match Broadwell-E's performance and power consumption that will be an extremely significant accomplishment.
 
Last edited:
Reactions: Dresdenboy and KTE

ecogen

Golden Member
Dec 24, 2016
1,217
1,288
136
Canard showed ~15% below broadwell...
They stated that their sample is limited to 3.3Ghz and it's doubtfull that the stock broadwell part clocks more then that for all core turbo.
And if you compare to the i5-6500 that does 3.3Ghz all core turbo for sure it looks even worse, 12 threads more ( +300% ) only give you ~80% more performance,a 4/8 part at the same clocks would be ~10% slower then a skylake i5 and that's before you subtract the SMT benefit.

We really have to wait for the final product and comprehensive benchmark suites to find out where it stands.

Lol what? It was 10% below the 6900k and above the i5-6500 in games, which are the best thing from their benchmarks we can use to at least ballpark estimate IPC. And you can bet that 6900k was boosting to at least 3.5Ghz in gaming situations.
 

iBoMbY

Member
Nov 23, 2016
175
103
86
I did't know the Z270 was an year old platform...

Not "year". "years". It is basically just the 6th revision of the same platform introduced with Sandy Bridge. And Kaby Fake/Z270 is still using the same DMI 3.0 interface revision, which was introduced with Z170. That's Intel on full snooze-mode for you.

AMD had the chance to introduce something new, maybe even revolutionary, but they decided really good and fast I/O isn't worth any consideration.
 

AtenRa

Lifer
Feb 2, 2009
14,003
3,361
136
AMD had the chance to introduce something new, maybe even revolutionary, but they decided really good and fast I/O isn't worth any consideration.

They give you 2x SATA + 4x USB 3.1 or 1 x4 NVMe directly on the CPU, this is seriously fast I/O, something Intel doesnt have in Socket 1151.
For the vast majority of the users the AM4 +ZEN platform is aiming for (mostly gamers), they have faster I/O and more Bandwidth per device than Intel Socket 1151.
 

leoneazzurro

Golden Member
Jul 26, 2016
1,051
1,711
136
Not "year". "years". It is basically just the 6th revision of the same platform introduced with Sandy Bridge. And Kaby Fake/Z270 is still using the same DMI 3.0 interface revision, which was introduced with Z170. That's Intel on full snooze-mode for you.

AMD had the chance to introduce something new, maybe even revolutionary, but they decided really good and fast I/O isn't worth any consideration.

I don't undestand you criticising AMD while the competitor, according even to your words, is doing worse.
 

iBoMbY

Member
Nov 23, 2016
175
103
86
I don't undestand you criticising AMD while the competitor, according even to your words, is doing worse.

Yes, I'm criticizing AMD, because they could do a lot better. I will get a Ryzen, and it will be okay. But still I'm very disappointed about the I/O options of Ryzen. They want to (also) compete in the high-end, and they only bring mid-level I/O. They should've at least given it 32 PCIe lanes. If they bring anything like SP3 or SP4 for workstations, you would need at least a 2x Summit Ridge MCM to have more I/O, which would be nice, but that would be at least at Intel 2011-3 prices.
 

leoneazzurro

Golden Member
Jul 26, 2016
1,051
1,711
136
Yes, I'm criticizing AMD, because they could do a lot better. I will get a Ryzen, and it will be okay. But still I'm very disappointed about the I/O options of Ryzen. They want to (also) compete in the high-end, and they only bring mid-level I/O. They should've at least given it 32 PCIe lanes. If they bring anything like SP3 or SP4 for workstations, you would need at least a 2x Summit Ridge MCM to have more I/O, which would be nice, but that would be at least at Intel 2011-3 prices.

It is perfectly possible that professional solutions will have also different solutions, their specifications are not publicy available yet.
 

jpiniero

Lifer
Oct 1, 2010
15,170
5,702
136
Like I'm totally expecting the vast majority of OEMs to not use any PCH with AM4. I think it's fine.

Yes, I'm criticizing AMD, because they could do a lot better. I will get a Ryzen, and it will be okay. But still I'm very disappointed about the I/O options of Ryzen. They want to (also) compete in the high-end, and they only bring mid-level I/O.

They aren't going after workstation at this point, and for enthusiasts (gamers) what Ryzen gives between what's onchip and the external PCH is more than adequate unless you use SLI/CF. If the MT performance is there, and the ST really in the Haswell range, I don't think it will be much of a problem.

Maybe they will do a separate socket for 2 die models later on.
 

Tup3x

Golden Member
Dec 31, 2016
1,080
1,079
136
Do we even know if AM4 is limited to those specs? Maybe there's gonna be a X390 chipset for the 2 die versions for exactly that reason.
I could be wrong but I think there aren't enough pins to allow that so it would have to be a new platform.
 

imported_jjj

Senior member
Feb 14, 2009
660
430
136
Yes, I'm criticizing AMD, because they could do a lot better. I will get a Ryzen, and it will be okay. But still I'm very disappointed about the I/O options of Ryzen. They want to (also) compete in the high-end, and they only bring mid-level I/O. They should've at least given it 32 PCIe lanes. If they bring anything like SP3 or SP4 for workstations, you would need at least a 2x Summit Ridge MCM to have more I/O, which would be nice, but that would be at least at Intel 2011-3 prices.

You might be a bit confused.
Ryzen should offer 32 PCIe lanes for dual GPU at x16, not sure if we have confirmation but that's what we assume.
In addition to that, the CPU + chipsets offer modularity. They got some IO from the CPU and some from the chipset. Ofc if mobo makers want more IO, they can add that with a 3rd party chip,like they've always been doing on most platforms.
It's much better than Intel's main platform but less than the server chips sold in consumer at insane prices. EDIT: if 2x16 PCIe 3.0 is not enough, maybe it would be better to wait for next year and hope they got PCIe 4.0.

And it is about cost because AMD would have to be insane to not place this against Kaby Lake but at 2x the cores.
It's clearly a tight die aimed at high volumes.
At 999$ they would sell 100k units in a year.
At 99$ to 349$ for quad to octa and one or two SKUs above that, they would ship tens of millions of units since everybody would upgrade.
They could even go to 49$ for quads but it is likely that ,this year, most that buy bellow 99$ would make an effort and get the quad. Next year,when Zen+arrives,they can drop the prices for this gen to 49$.

No 4 chans memory ,no huge L3 cache, nice IO but not excessive.They are gonna hit Intel where it hurts, across the board and serve every customer that doesn't need a silly integrated GPU.
If Zen delivers and AMD has sufficient supply, it's gonna be weird. The massive upgrade cycle it will cause....

Some will cry about margins and you can look at that in a couple of ways.
How big is the die, Intel's 4 cores APUs have a GPU that's bigger than the 4 cores+cache and it's 120+mm2.They could have given us more cores a long time ago, they just don't want to.
Lets say Zen is 200mm2 or less and even if we are generous in all areas if you do the math for costs - die size, wafer cost, yield, test and packaging, box contents(if not tray), transport and so on - you get to 40$ or less for a 6 and 8 cores die. Even if the average retail for the 6-8 cores ends up at 250$ , there is ample room for AMD to have great margins while also feeding the distributors and retailers.
4 cores would cost a bit more on a per core basis but still lots of room especially if they don' t push bellow 99$ for cheapest SKU.
Or you can think of it this way,Polaris 10 is used in the 470D to 480 8GB and has an area of 232mm2. But with GPUs you also got a bunch of GDDR5, the actual board, the AiB partners so the costs are much higher.
If Zen lands somewhre between 150 and 200m2 why be extra extra greedy and not price it against Kaby Lake.
AMD needs to build the brand too and giving us prices we like is very easy today while also making a LOT of money. Intel will flood the airwaves with ads, AMD can't do that but they can offer reasonable prices and gain exposure.It will help their sales in other segments too and help longer term.
Every price point they don't address is business left on the table that feeds Intel.
Intel can react but will take a bit of time so AMD should exploit the opportunity and once Intel reacts, they can manage it as Intel can't afford to kill its margins while also killing average selling prices.

I see that many are so conditioned by Intel's marketing, that 8 cores for bellow 500$ is nuts to them. It's not ,8 cores is the equivalent of a quad core APU in costs.

Anyway , we'll see what AMD does ,if they play it smart or they waste the opportunity. They should have supply if they also use Samsung but, if there are shortages, it would take quite a bit of time for supply to catch up.
 
Last edited:
Reactions: psolord
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |