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

[Fjodor2001]

Maybe it used to be that way, but I'm not sure if it is any longer when it comes to CPUs/computers.

Certainly not for desktop computers, where I assume a 5-8 year cycle is more normal. The reason being that the improvement rate have gone down so much and there is little reason to upgrade (laptops excluded).

I'd imagine the same being true for server CPUs, although there have been some more progress there (mainly more cores). I doubt upgrading every 2-3 years is economically justified.

 

[Edrick]

On desktop, yes. But that was just because desktop Broadwell only got the tiny niche SKUs 5675/5775C. I.e. desktop Broadwell was more or less skipped.

On mobile, Skylake lauched much later than Broadwell. Because there we got a full Broadwell lineup.

Ok, mobile skylake launched 8 months after broadwell. Granted that is better than the desktop, but if skylake-E launches 8 months after broadwell-E, that puts us in Q4/2016. So I think Q1 2017 is very likely.

 

[Arachnotronic]

I think there are posters here that are genuinely hoping for Intel to make competitively stupid moves in the hopes that AMD can capitalize on such stupidity. It is genuinely amusing to watch.

 

[Fjodor2001]

I think there are posters here that are genuinely hoping for Intel to make competitively stupid moves in the hopes that AMD can capitalize on such stupidity. It is genuinely amusing to watch.

Care to be more precise?

 

[Fjodor2001]

Ok, mobile skylake launched 8 months after broadwell. Granted that is better than the desktop, but if skylake-E launches 8 months after broadwell-E, that puts us in Q4/2016. So I think Q1 2017 is very likely.

Intel Broadwell Mobile: September 5, 2014
Intel Skylake Mobile: September 1, 2015

So 12 months, not 8.

=> The likelihood of of Skylake-E launching just a few months after Broadwell-E are slim.

 

[Arachnotronic]

Care to be more precise?

Sure. Intel is very well aware of AMD's public Zen product launch plans (there are literally teams of people within Intel paid -- and paid well -- to do such competitive analysis), so to think that Intel will sit idly by while it allows AMD -- a competitor that Intel worked so hard to defeat -- to regain share in rather profitable portions of the PC processor market seems a little bit naive, IMO.

Intel will put out products as quickly as it possibly can to in order to respond to any competitive threat from Zen-based processors, IMO.

 

[scannall]

Sure. Intel is very well aware of AMD's public Zen product launch plans (there are literally teams of people within Intel paid -- and paid well -- to do such competitive analysis), so to think that Intel will sit idly by while it allows AMD -- a competitor that Intel worked so hard to defeat -- to regain share in rather profitable portions of the PC processor market seems a little bit naive, IMO.

Intel will put out products as quickly as it possibly can to in order to respond to any competitive threat from Zen-based processors, IMO.

Only if the numbers worked though. If AMD doubled its market share, it still wouldn't show on Intel's bottom line. So going crazy, or pushing products that aren't ready isn't going to happen. Unlikely they would bother going price war either, as that would cost more than just ignoring AMD at this point.

 

[Fjodor2001]

Sure. Intel is very well aware of AMD's public Zen product launch plans (there are literally teams of people within Intel paid -- and paid well -- to do such competitive analysis), so to think that Intel will sit idly by while it allows AMD -- a competitor that Intel worked so hard to defeat -- to regain share in rather profitable portions of the PC processor market seems a little bit naive, IMO.

Intel will put out products as quickly as it possibly can to in order to respond to any competitive threat from Zen-based processors, IMO.

So you're basically suggesting they would drop the price of the HEDT CPUs? Or is there any other option that can be realized in short time?

However that contradicts the predictions by ShintaiDK who have said that the current monopoly-like situation actually results in lower prices. If anything, a competitive AMD Zen should raise prices.

Anyway, things to consider:

* Last time AMD was competitive, Intel chose not to lower prices of server CPUs and engage in a price-war. They'd rather give up market share than margins. Why would they act differently this time?

* Even if Intel lowers price of their CPUs, I think AMD can grab a decent market share and make nice profit from AMD Zen. I don't think Intel will go for extremely lower prices, or even negative profit. That would be insane, and would kill Intel's profit too.

Finally, regardless of how much Intel decides to lower price, I'm just happy if we get some competition going. It will benefit all consumers. Except for those stuck in a fanboy camp.

 

[frozentundra123456]

Yea, look how well they have responded to ARM.

In any case, nobody really knows what Zen will be like, or even when it will be available, hype train projections not withstanding.

By the same token, if Zen is competitive with intel, nobody really knows how intel will respond. I mean, they have had plenty of chances to increase mainstream core count and stubbornly resisted doing so. And I am not sure they really can wring anything more out their notebook optimized processes except adding more cores. Maybe making edram more available, but that might eat into their precious margins, just like adding more cores. So if Zen is competitive with intel's best (highly unlikely IMO), but if it is, I am not sure Intel has the willingness and/or ability to do much about it.

Edit: I am basically saying what Scannall said. I am not sure intel can easily increase performance, and I think they might cede a few percent of the market to AMD rather than get involved in a price war.

 

[Arachnotronic]

So you're basically suggesting they would drop the price of the HEDT CPUs? Or is there any other option that can be realized in short time?

Don't drop prices; increase value! Bring in higher core counts from the bench of Xeon E5 processors to keep ASPs high. This leads to slightly higher production cost obviously, but it's better than losing the sale completely to AMD.

Lowering prices is a sucker's game. It's always better to try to increase the value you deliver to your customer at a given price point.

 

[Fjodor2001]

Don't drop prices; increase value! Bring in higher core counts from the bench of Xeon E5 processors to keep ASPs high. This leads to slightly higher production cost obviously, but it's better than losing the sale completely to AMD.

But this would mean that Intel will be eating into their higher end server CPU segment (where higher core counts matter most), lowering margins and profit there.

Also, I'm not sure the HEDT market is ready for 12 or 16 core CPUs. Do you think those considering an 8 core Intel HEDT CPU today will pay significantly more to get a 12/16 core one?

And anyway, IIRC, AMD also has 16+ core Zen CPUs planned. So they can also just respond by increasing core count if needed to "increase value". Then we'd get a core-war, instead of a price-war.

Lowering prices is a sucker's game. It's always better to try to increase the value you deliver to your customer at a given price point.

Sure, if you have that option. But you may not always have it, and then you have to fall back to lowering price, or you'll lose market share.

 

[TheELF]

Maybe it used to be that way, but I'm not sure if it is any longer when it comes to CPUs/computers.

Certainly not for desktop computers, where I assume a 5-8 year cycle is more normal. The reason being that the improvement rate have gone down so much and there is little reason to upgrade (laptops excluded).

I'd imagine the same being true for server CPUs, although there have been some more progress there (mainly more cores). I doubt upgrading every 2-3 years is economically justified.

Yeah I guess you're right,the point still is that intel won't loose any money by skipping a gen because companies will upgrade to whatever is newest at whatever point they do upgrade.

 

[jpiniero]

And anyway, IIRC, AMD also has 16+ core Zen CPUs planned. So they can also just respond by increasing core count if needed to "increase value". Then we'd get a core-war, instead of a price-war.

AMD would never recoup the costs. Take a look at the Furies. They probably won't take a bath on it but they aren't selling enough to avoid losing money.

 

[cbn]

Also, I'm not sure the HEDT market is ready for 12 or 16 core CPUs. Do you think those considering an 8 core Intel HEDT CPU today will pay significantly more to get a 12/16 core one?

We have to remember that the eight core CPUs have SMT. So that makes a eight core a sixteen thread already.

For a Server though, increasing core count and die size should help. This provided single thread doesn't fall below a certain level. (SIDE NOTE: Intel appears to have hit this wall already. Proof is the low base clock speed (2.3 Ghz) on the E5 2699 v3 Xeon and the reduced progression of core count reported for Broadwell (24) and Skylake (28) Xeons.)

 

[jpiniero]

Proof is the low base clock speed (2.3 Ghz) on the E5 2699 v3 Xeon and the reduced progression of core count reported for Broadwell (24) and Skylake (28) Xeons.)

If you look at the E7, it would be:

10
15
18

If they went to 24 and 28 next that wouldn't be so bad.

 

[Fjodor2001]

AMD would never recoup the costs.

How did you arrive at that conclusion? And how would it differ from Intel?

 

[AMD64Blondie]

Bah humbug.Never switching back to AMD.

I'm going Intel from here on out.(currently rocking a i5 4670.)

 

[cbn]

If you look at the E7, it would be:

10
15
18

If they went to 24 and 28 next that wouldn't be so bad.

10 cores for Westmere EX (130W TDP, 2.4 Ghz base clock), 15 cores for Ivy Bridge EX (155w TDP, 2.8 Ghz clock), 18 Cores for Haswell EX (165W TDP, 2.5 Ghz base clock).

And then for Broadwell and Skylake we have to factor in 14nm is 1.5 node jump from 22nm. So 28C is actually not that big of an increase when a person looks at it that way. This, contrast, to Westmere EX to Haswell EX which was a 80% increase in cores for a 1 node jump.

Westmere to Haswell= 80% increase in cores for a 1 node jump (32nm Toc to 22nm Toc)
Haswell to Skykake = 55% increase in cores for a 1.5 node jump (22nm Toc to 14nm Toc)

P.S. On Skylake, unless Intel increases TDP I'm thinking base clocks will probably drop again. Maybe 200 Mhz, but then IPC is higher so I guess it balances out somewhat.

 

[tential]

Bah humbug.Never switching back to AMD.

I'm going Intel from here on out.(currently rocking a i5 4670.)

?
Why?
I mean, I just happen to be getting a new processor around Zen's release most likely, or Zen+ release. And there is no way I'd lock myself into an "Intel only" processor. Who cares what the brand is. Best processor at the time for the price is what I get.

It could be a processor called Zulu XyZ you're a dirtbag loser 2000, by a no name brand, if it was the best processor for the money I had, that's what I'd get.

Edit: Zen does seem interesting in my unraid box. Using the large number of cores, I could use Zen to power a lot of functions done by outside PCs currently, all within my server box in virtualized spaces. So I do think there will be interesting applications for enthusiasts with Zen. It may not be a processor we buy for gaming, but we may be able to find interesting uses for it. I still bet Intel will be the best option, but I could see a situation where I end up getting both for different purposes. This all depends on Intel's core counts though at the time though and performance! If Zen+ didn't have a significant corecount or IPC or any advantage that would let me see a different application for it, then forget it.

 

[cbn]

Edit: Zen does seem interesting in my unraid box. Using the large number of cores, I could use Zen to power a lot of functions done by outside PCs currently, all within my server box in virtualized spaces.

Yeah, that is something I want to do also:

http://lime-technology.com/virtualization-host/

So far I still behind on a lot work, but I have been wanting to get this going on my FM2+ board. Hopefully by Christmas I have something up and running. If not then hopefully in January.

 

[jihe]

That Zen will do fine and get Haswell or maybe slightly more IPC level performance? I thought that too!

I fully expect that to be on AMD's powerpoint slides.

 

[VirtualLarry]

Then we'd get a core-war, instead of a price-war.

I had hoped that is what would have happened, when Thuban dropped. Sadly, Intel mainstream CPUs were and are still quad-cores.

Either way, bring on the "core wars". I do DC, so I'll happily soak up as many cores as they want to give us.

 

[myocardia]

From where have you heard that Intel would develop a new Skylake core for Skylake-E that significantly differs from the one in 1151 desktop Skylake? They've not done that before on the E series.

Intel has never made any E series CPU that did not start its life as a Xeon core, all the way back to the first E series, the Sandy Bridge-E. This is why there has never been a single E series CPU that used the same socket as their consumer CPU of the same generation. They have always used the same socket as the Xeons of that generation. Crazy, huh?

 

[Yuriman]

Frankly, I'd probably buy a quad core Zen, if such a thing exists and is competitive. I have little use for more than 4 cores. I'm a light gamer, use my machine for the occasional transcoding and compression, nothing too heavy these days. If Zen isn't compelling enough, I'll likely end up with a Kaby Lake or Cannonlake i3 as a replacement for my i5. The only reason I'd be tempted by a more expensive chip is if the integrated graphics allow me to game comfortably enough to ditch the discrete GPU.

As long as Intel is still selling mostly dual cores, and with quads making up the "upper mainstream", things will continue to run very well on a modern i3.

 

[Vortex6700]

Intel's response when they get beat by amd in performance isn't to lower prices, it's to pay companies to use thier products exclusively and threaten to short ship any oem that doesnt agree to that strategy. This practice is illegal in both the UK and America, but that matters just about as much as restrictions on covert surveillance.