New Zen microarchitecture details

Page 43 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
Ok, doubling the per thread allocation increased the performance by 2.7% (1.82fps to 1.87fps). Not much, as expected and I need to test if it does the same on Haswell too.

EDIT: All threads used, thread allocation untouched: 6.11fps.

I'd say it is just that >= Haswell performance has improved due the new optimizations implemented in newer encoder versions.
 
Last edited:

RussianSensation

Elite Member
Sep 5, 2003
19,458
765
126
That's a Sandy Bridge you're looking at there. Skylake is probably ~34% faster per clock than Sandy Bridge.

It's not that much on average but your point is valid.

getgraphimg.php


The comparisons of Zen to older Intel CPU architectures such as Sandy, Ivy or even Haswell are mostly irrelevant. That is because by the time Zen launches in 2017, Skylake will be 1.5 years old (August 2015), will have been superseded by Kaby Lake, and Cannonlake will be around the corner. It's ludicrous to suggest that all this time people wait for Zen but then start comparing Zen to older Intel CPUs such as Haswell-E (August 2014 series) or Skylake (August 2015 series). Even 2016 Broadwell-E and Kaby Lake won't be the main competition to Zen for that long in 2017. As long as Zen is a 2017 launch product, its main competition will become Skylake-E and Cannon Lake.

Besides, someone who bought Haswell-E in 2014-2015 or Skylake in 2015 has enjoyed all that performance until the day Zen launches. If we buy the argument that 2017 Zen is worth waiting for, why not wait another year for Icelake? These types of arguments become circular in nature.

That's why I didn't even bother waiting for Zen. With how well Intel's modern CPUs hold value, there is no point at all waiting for Zen. One can just as easily purchase i7-6700K/i7 6800K BW-E and if Zen is really that good, just sell the Intel platform and upgrade to Zen in 2017. If Zen flops, you get to enjoy a proper fast CPU that's fast enough to drive Polaris, Vega, Pascal, instead of still using some outdated Nehalem, i5 Sandy or worse Bulldozer/Vishera, etc.

Also, Intel's weakest CPUs are in the sub-$250 segment but if Zen is initially only launching as an 8-core, it seems AMD is targeting the high-end. AMD's biggest problem is that many of the high-end enthusiasts will have upgraded to Intel with 2013-2015 Haswell/E, 2015-2016 Skylake/Kaby Lake/BW-E. It's not a surprise that Intel is experiencing record i7 sales in modern times. It becomes a tricky situation for anyone who wants a new 14nm GPU in 2016 because if you want the upcoming cards launching in June-July, no CPU from Nehalem, Sandy i5 (or below) or AMD series is fast enough to max them out at 1080p 60Hz. That means if someone is waiting until 2017 for Zen, they cannot even upgrade to 14nm cards until then too. That's a crappy compromise if you ask me. :thumbsdown: That's why AMD delaying Zen to 2017 is a horrible mistake.
 
Feb 19, 2009
10,457
10
76
@RS

Zen's market segment are for those on i3 and i5 (as well as those using AMD CPUs). This segment don't go higher than $399.

AMD has WSA and they need to move volume. Zen I bet is going to be even smaller than Polaris 10. They have plenty of margins to be flexible in pricing. If they want to shake things up, give 5960X caliber at <$399.

If they can do it for Polaris 10 SKUs at potentially <$299 with a bigger 14nm FF die and 8GB vram along with all the PCB, they can do it for a small Zen chip.
 

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
Did I miss something here . . . ?

I wouldn't be surprised if that would basically be the case... :sneaky:

AMD said Zen will launch in 2016, however they didn't say on which platform... There are three other Zen platforms besides AM4, which appears to be the runt like I said before. I would guess they launch the server platforms first and only then AM4.

That's just a guess, however it wouldn't make sense to launch Bristol Ridge in Computex (?) if Zen was to be released only few months later. Bristol Ridge is an old product already (Carrizo), but still releasing it for AM4 isn't free.
 

DrMrLordX

Lifer
Apr 27, 2000
21,583
10,785
136
I dunno, Zen for desktop makes more sense in 2016 than Zen for server. Validation and all that. Plus if it's a boutique release, that let's them run up prices and build hype with it, which is harder to do if you're putting it on a stodgy server/workstation platform that doesn't even (officially) support overclocking, non-ECC RAM, etc. Zen ships for revenue in 2017, but in 2016 it's a flagship showcase product.

My thought is that Bristol Ridge is there only to get AM4 launched and the bugs worked out of early board revisions. It paves the way for a smooth(er) Summit Ridge launch to AM4 in Oct-Dec 2016.
 
Mar 10, 2006
11,715
2,012
126
I wouldn't be surprised if that would basically be the case... :sneaky:

AMD said Zen will launch in 2016, however they didn't say on which platform... There are three other Zen platforms besides AM4, which appears to be the runt like I said before. I would guess they launch the server platforms first and only then AM4.

That's just a guess, however it wouldn't make sense to launch Bristol Ridge in Computex (?) if Zen was to be released only few months later. Bristol Ridge is an old product already (Carrizo), but still releasing it for AM4 isn't free.

They've said it's for desktop first, server version coming in 2017.
 

majord

Senior member
Jul 26, 2015
433
523
136
It's not that much on average but your point is valid.

getgraphimg.php


The comparisons of Zen to older Intel CPU architectures such as Sandy, Ivy or even Haswell are mostly irrelevant. That is because by the time Zen launches in 2017, Skylake will be 1.5 years old (August 2015), will have been superseded by Kaby Lake, and Cannonlake will be around the corner. It's ludicrous to suggest that all this time people wait for Zen but then start comparing Zen to older Intel CPUs such as Haswell-E (August 2014 series) or Skylake (August 2015 series). Even 2016 Broadwell-E and Kaby Lake won't be the main competition to Zen for that long in 2017. As long as Zen is a 2017 launch product, its main competition will become Skylake-E and Cannon Lake.

Besides, someone who bought Haswell-E in 2014-2015 or Skylake in 2015 has enjoyed all that performance until the day Zen launches. If we buy the argument that 2017 Zen is worth waiting for, why not wait another year for Icelake? These types of arguments become circular in nature.

That's why I didn't even bother waiting for Zen. With how well Intel's modern CPUs hold value, there is no point at all waiting for Zen. One can just as easily purchase i7-6700K/i7 6800K BW-E and if Zen is really that good, just sell the Intel platform and upgrade to Zen in 2017. If Zen flops, you get to enjoy a proper fast CPU that's fast enough to drive Polaris, Vega, Pascal, instead of still using some outdated Nehalem, i5 Sandy or worse Bulldozer/Vishera, etc.

Also, Intel's weakest CPUs are in the sub-$250 segment but if Zen is initially only launching as an 8-core, it seems AMD is targeting the high-end. AMD's biggest problem is that many of the high-end enthusiasts will have upgraded to Intel with 2013-2015 Haswell/E, 2015-2016 Skylake/Kaby Lake/BW-E. It's not a surprise that Intel is experiencing record i7 sales in modern times. It becomes a tricky situation for anyone who wants a new 14nm GPU in 2016 because if you want the upcoming cards launching in June-July, no CPU from Nehalem, Sandy i5 (or below) or AMD series is fast enough to max them out at 1080p 60Hz. That means if someone is waiting until 2017 for Zen, they cannot even upgrade to 14nm cards until then too. That's a crappy compromise if you ask me. :thumbsdown: That's why AMD delaying Zen to 2017 is a horrible mistake.

Please send me a Kaby lake Sample, so I can add it to the bunch... Maybe a Zen sample too?

In other words, what else are we supposed to make comparisons to?


As for moving beyond to Kaby Lake etc, we don't even know if it will bring any IPC uplift letaloe how much, so cannot even speculate here. Nor do we know when Cannon lake will hit the market. You could just as easily say, "Cannon lake won't be Zen's main competiton because by then Zen+ will be just around the corner"

.. and so on.

As it stands, with the information at hand, Skylake and Kabylake will be Zen's direct competition, simple as that.

As for waiting for Zen, well I actually don't think anyone has suggested people 'wait' for zen for the sake of zen itself. The only reason anyone suggests this to people, is for potential it has to distrupt the pricing of hex and octocore CPU's in general. Without knowing final performance, or AMD's approach to pricing (which will lbe influcenced by many things - Performance, yields, and whether they want to go for market share or maximize margins) . It's so far out though I don't think anyone would wait if they needed the performance now, and had the $$
 

el etro

Golden Member
Jul 21, 2013
1,581
14
81
I wouldn't be surprised if that would basically be the case... :sneaky:

AMD said Zen will launch in 2016, however they didn't say on which platform... There are three other Zen platforms besides AM4, which appears to be the nly few months later. Bristol Ridge is an old product already (Carrizo), but still releasing it for AM4 isn't free.

I think that BR exists simply because don't makes any economical sense to port Excavator to 14LPP, and the process was not really ready a few months ago.
Another thing i think is that AM4 haves a good planned longevity, so make BR pin to pin compatible with the very first Zen APU is a nice idea, it gives users great future upgrade possibility, like Intel did with LGA775 socket.
 

Dresdenboy

Golden Member
Jul 28, 2003
1,730
554
136
citavia.blog.de
Please send me a Kaby lake Sample, so I can add it to the bunch... Maybe a Zen sample too?

In other words, what else are we supposed to make comparisons to?


As for moving beyond to Kaby Lake etc, we don't even know if it will bring any IPC uplift letaloe how much, so cannot even speculate here. Nor do we know when Cannon lake will hit the market. You could just as easily say, "Cannon lake won't be Zen's main competiton because by then Zen+ will be just around the corner"

.. and so on.

As it stands, with the information at hand, Skylake and Kabylake will be Zen's direct competition, simple as that.

As for waiting for Zen, well I actually don't think anyone has suggested people 'wait' for zen for the sake of zen itself. The only reason anyone suggests this to people, is for potential it has to distrupt the pricing of hex and octocore CPU's in general. Without knowing final performance, or AMD's approach to pricing (which will lbe influcenced by many things - Performance, yields, and whether they want to go for market share or maximize margins) . It's so far out though I don't think anyone would wait if they needed the performance now, and had the $$

How would a change in the market's pricing structure affect prices of used Intel CPU's? I think, they would land at their future market prices minus some offset similar to todays or significantly lower (depending on supply). Thanks to the low IPC inreases during the last years this might be some psychological frog in boiling water phenomenon (as in stock markets while prices are slowly declining). What happens, if we "throw the frog into hot water" with prices?
 

rtsurfer

Senior member
Oct 14, 2013
733
15
76
It's not that much on average but your point is valid.


The comparisons of Zen to older Intel CPU architectures such as Sandy, Ivy or even Haswell are mostly irrelevant. That is because by the time Zen launches in 2017, Skylake will be 1.5 years old (August 2015), will have been superseded by Kaby Lake, and Cannonlake will be around the corner. It's ludicrous to suggest that all this time people wait for Zen but then start comparing Zen to older Intel CPUs such as Haswell-E (August 2014 series) or Skylake (August 2015 series). Even 2016 Broadwell-E and Kaby Lake won't be the main competition to Zen for that long in 2017. As long as Zen is a 2017 launch product, its main competition will become Skylake-E and Cannon Lake.

Kaby will come around August 2016. If Zen is early 2017, the imminent threat would be upcoming Skylake-E, which is not coming till probably next computex. CannonLake is way out of the picture.
 

mikk

Diamond Member
May 15, 2012
4,112
2,108
136
Kaby will come around August 2016.


KBL-U probably but I doubt this for desktop, Q4-Q1 2017 according to Roadmaps. Zen should outperform Skylake without problems in some of the benchmarks that are multithreaded because of 4 cores vs 8 cores.

Zen APU might be different if it comes in H2 2017 and in case Intel can hold their Cannonlake H2 2017 target for mobile.
 

AtenRa

Lifer
Feb 2, 2009
14,000
3,357
136
The comparisons of Zen to older Intel CPU architectures such as Sandy, Ivy or even Haswell are mostly irrelevant. That is because by the time Zen launches in 2017, Skylake will be 1.5 years old (August 2015), will have been superseded by Kaby Lake, and Cannonlake will be around the corner. It's ludicrous to suggest that all this time people wait for Zen but then start comparing Zen to older Intel CPUs such as Haswell-E (August 2014 series) or Skylake (August 2015 series). Even 2016 Broadwell-E and Kaby Lake won't be the main competition to Zen for that long in 2017. As long as Zen is a 2017 launch product, its main competition will become Skylake-E and Cannon Lake.

1. ZEN will launch the same Quarter (4Q 2016) as Kabylake.
2. Desktop Cannonlake may not even launch in 2017.
3. Broadwell-E platform is more expansive than AM4, also BD-E will be 140W TDP vs 95W TDP for ZEN. BD-E will be faster but more expensive and with high power consumption.

Besides, someone who bought Haswell-E in 2014-2015 or Skylake in 2015 has enjoyed all that performance until the day Zen launches.

Quad-Core Skylake doesnt have the same MT performance as 8-Core ZEN. So no you havent had that performance in 2015, especially if 8-Core ZEN will launch at $300-400 price.


Also, Intel's weakest CPUs are in the sub-$250 segment but if Zen is initially only launching as an 8-core, it seems AMD is targeting the high-end.

I have a feeling ZEN will target $200 to $400 price segment. That is from Core i5 KabyLake ( 6-Core ZEN ??) to Core i7 KabyLake (8-Core ZEN ??) to Core i7 Broadwell-E 6-Core (8-Core ZEN ??).

Lower than Core i5 is the APU domain.

AMD's biggest problem is that many of the high-end enthusiasts will have upgraded to Intel with 2013-2015 Haswell/E, 2015-2016 Skylake/Kaby Lake/BW-E. It's not a surprise that Intel is experiencing record i7 sales in modern times. It becomes a tricky situation for anyone who wants a new 14nm GPU in 2016 because if you want the upcoming cards launching in June-July, no CPU from Nehalem, Sandy i5 (or below) or AMD series is fast enough to max them out at 1080p 60Hz. That means if someone is waiting until 2017 for Zen, they cannot even upgrade to 14nm cards until then too. That's a crappy compromise if you ask me. :thumbsdown: That's why AMD delaying Zen to 2017 is a horrible mistake.

Every current CPU can use Polaris and GP104. You may not get 100fps but only 90fps or 70fps but you will get far higher fps than stay with your Core i5 + Kepler. And since you gain a lot of performance with slower CPUs from DX-12 games, you may upgrade your GPU to Polaris/VEGA/Pascal even with a FX8350 or slower CPU.
 
Aug 11, 2008
10,451
642
126
@RS

Zen's market segment are for those on i3 and i5 (as well as those using AMD CPUs). This segment don't go higher than $399.

AMD has WSA and they need to move volume. Zen I bet is going to be even smaller than Polaris 10. They have plenty of margins to be flexible in pricing. If they want to shake things up, give 5960X caliber at <$399.

If they can do it for Polaris 10 SKUs at potentially <$299 with a bigger 14nm FF die and 8GB vram along with all the PCB, they can do it for a small Zen chip.

Yea, I can just hear the AMD executives talking at the bar. Wow, we have had to price our products really cheap for years now. Sucks doesnt it?? Yea, sure does. But the products just were not that good.

Have you heard that Zen is going to be as good as Intel's thousand dollar chip at 70% of the TDP? Amazing isnt it?

Yea, sure is amazing performance. I have a great idea. Lets price it really cheap. OK, Lisa said we should remain the bargin alternative didnt she? She and the shareholders will love that idea.
 

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,686
136
It becomes a tricky situation for anyone who wants a new 14nm GPU in 2016 because if you want the upcoming cards launching in June-July, no CPU from Nehalem, Sandy i5 (or below) or AMD series is fast enough to max them out at 1080p 60Hz. That means if someone is waiting until 2017 for Zen, they cannot even upgrade to 14nm cards until then too. That's a crappy compromise if you ask me.
The same 2017 when big Pascal and Vega launch? You know, those chips even you consider the true 14nm flagships.

If you want a new system in 2016, you don't wait for Zen, there's plenty of good available alternatives. But let's not invent other reasons just for the sake of argument.
 

Dresdenboy

Golden Member
Jul 28, 2003
1,730
554
136
citavia.blog.de
Yea, I can just hear the AMD executives talking at the bar. Wow, we have had to price our products really cheap for years now. Sucks doesnt it?? Yea, sure does. But the products just were not that good.

Have you heard that Zen is going to be as good as Intel's thousand dollar chip at 70% of the TDP? Amazing isnt it?

Yea, sure is amazing performance. I have a great idea. Lets price it really cheap. OK, Lisa said we should remain the bargin alternative didnt she? She and the shareholders will love that idea.
+1 for this funny story. ;) Surely it won't be cheap (as already said). I wonder if the assumed different price levels correlate with different performance expectations out there.
 

Dresdenboy

Golden Member
Jul 28, 2003
1,730
554
136
citavia.blog.de
8C vs 4C is a difference to 4C vs 4C.
Which use cases do you see there? What does an APU buyer (incl. full systems) expect? For general apps and older games the 8T and improved ST performances are much better than with BR. These games are also mostly DX11 based. 28nm APU gfx isn't that much behind 14nm Intel gfx, probably more due to the limiting CPU performance.
So if 4 additional cores aren't that big of a gain, it's better to spend these 30+ sqmm on GPU CUs and cache.
 

mikk

Diamond Member
May 15, 2012
4,112
2,108
136
You don't understand my point. Without a core count advantage CPU performance is all up to IPC and x Mhz, means it's harder for AMD to compete with Intel. If a 4 core version is enough for mobile is a different question.
 

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
Anything but a single CCX (4C/8T) for Raven Ridge is highly unlikely. On RR they have the bandwidth for 10-12 GCN CUs so there is not much point in pairing more CPU resources with a such GPU.
 

Vortex6700

Member
Apr 12, 2015
107
4
36
If AMD 4c/8t has marginally lower IPC but beats out the i5 xxxxk in MT due to HT, would everyone here go with the i5 option at the same price?
 

hojnikb

Senior member
Sep 18, 2014
562
45
91
If AMD 4c/8t has marginally lower IPC but beats out the i5 xxxxk in MT due to HT, would everyone here go with the i5 option at the same price?

Well, depends on what you use your PC for. Maybe AMD performans worse in games. Or maybe motherboards are costlier.

It's platform cost, that matter, not just cpu.
 

JDG1980

Golden Member
Jul 18, 2013
1,663
570
136
Yea, I can just hear the AMD executives talking at the bar. Wow, we have had to price our products really cheap for years now. Sucks doesnt it?? Yea, sure does. But the products just were not that good.

Have you heard that Zen is going to be as good as Intel's thousand dollar chip at 70% of the TDP? Amazing isnt it?

It won't be as good as Broadwell-E or even Haswell-E. It will probably have IPC about on par with Sandy Bridge, and reasonably competitive clock speeds (3.0 - 4.0 GHz, depending on various factors). That means it will fill a market niche similar to what Thuban did in the Nehalem days: more cores, with each individual core having somewhat less (but still quite decent) IPC, and clock speeds a bit lower than Intel's best.

Thuban started out at $295 for the top SKU and $199 for the lower clocked version. I wouldn't be surprised to see similar pricing on Zen. Remember, it has to compete not only with Intel's i5 and i7 lineup, but also with used Sandy Bridge-E chips which have recently gained a following in the enthusiast market.
 

jhu

Lifer
Oct 10, 1999
11,918
9
81
It won't be as good as Broadwell-E or even Haswell-E. It will probably have IPC about on par with Sandy Bridge, and reasonably competitive clock speeds (3.0 - 4.0 GHz, depending on various factors). That means it will fill a market niche similar to what Thuban did in the Nehalem days: more cores, with each individual core having somewhat less (but still quite decent) IPC, and clock speeds a bit lower than Intel's best.

Thuban started out at $295 for the top SKU and $199 for the lower clocked version. I wouldn't be surprised to see similar pricing on Zen. Remember, it has to compete not only with Intel's i5 and i7 lineup, but also with used Sandy Bridge-E chips which have recently gained a following in the enthusiast market.

Used Sandy Bridge E is irrelevant. Not even Intel competes with those on price.