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

[mysticjbyrd]

Why would I be "afraid" of Zen being a good chip? I just get tired of the endless, ever inflating predictions from the AMD fans in these forums. And BTW, I own no Intel stock, and have absolutely no financial interest in Intel. Just because someone does not buy into endless hype does not mean they have a financial interest in a competitor. You might also be surprised to know that I used to be a fan of AMD cpus, and I am currently using an AMD gpu. So if they make a product that meets my needs I will buy it.

Are you joking? I see a 1000x more vitriol pessimist predictions from intel fanboys, than I see overzealous fanboys of AMD. It's gotten to the point that people will flame other members for even mentioning AMD.

Instead of adding to the vitriol, what if you tried not clicking on a topic about zen, if you aren't interested in Zen?

 

[Maximilian]

Wonder if it'll actually be any good or it'll be bulldozer all over again where its only really good at one or two things. Then people claim itll be better in the future when multi threading takes off bigtime etc etc yada yada and it never really happens. :\

I hope by some miracle its actually good, hardware is pretty boring with no competition. All I see is these days is AMD lose money, AMD lose employees, AMD lose market share, its a shame tbh.

 

[cbn]

It certainly appears that AMD could have made a 4m SR chip sans L3 that would be smaller than the current Kaveri die. There are those in the know who insist such a beast could never exist within an reasonable power envelope, and they probably know exactly why . . .

4M SR on FM2+ would be 95W and 23.75W per module. So it seems possible, but maybe they didn't want to because of overlap with AM3+?

P.S. 4M XV would have been even smaller than 4M SR (although it probably wouldn't overclock as well due to limited frequency/voltage scaling past 25W per module and certainly by the same token the turbo on 4M XV would have been less than the turbo on 4M SR when comparing stock 95W configurations).

 

[bystander36]

Are you joking? I see a 1000x more vitriol pessimist predictions from intel fanboys, than I see overzealous fanboys of AMD. It's gotten to the point that people will flame other members for even mentioning AMD.

Instead of adding to the vitriol, what if you tried not clicking on a topic about zen, if you aren't interested in Zen?

The reality is, given their track record, being pessimistic is just living in reality. Being a vocal optimist, and shooting down the possibility of them not living up to the hype, is a bit naive.

It's always a good idea going into the AMD hype train with a bit of caution. As much as you want the opposite to be true, you'll be very disappointing a lot more than not. You do not have to be an Intel fanboy to doubt AMD living up to the hype.

Is it really pessimism to expect an outcome that has repeated itself several times before?

 

[TurtleCrusher]

Why would I be "afraid" of Zen being a good chip? I just get tired of the endless, ever inflating predictions from the AMD fans in these forums. And BTW, I own no Intel stock, and have absolutely no financial interest in Intel. Just because someone does not buy into endless hype does not mean they have a financial interest in a competitor. You might also be surprised to know that I used to be a fan of AMD cpus, and I am currently using an AMD gpu. So if they make a product that meets my needs I will buy it.

There has seemingly been nothing unreasonable said about Zen and it's performance. The IPC increase from Northwood to Conroe was what, 100%? a 40% claim isn't crazy.

The lessons learned from Stars, Kabini, and Bulldozer should culminate into something akin to a Nehalem type jump from Yorkfield, or Prescott to Conroe, or somewhere in between.

If you're an enthusiast, you know that this might be the last hurrah for PCs as we know it if Zen (or another product from AMD) doesn't compete. The market will disappear. AMD seemingly is righting the ship that had been on a crash course for years. The tone spoken of AMD in these forums (both CPU and GPU) is that they are ran by a bunch of maladroit engineers, which isn't the case. They struck gold in the Jaguar APUs for consoles, and things are finally looking up.

You wouldn't know that by listening to all of the shills on here. Some of the mods know who I worked for up until recently and I personally have every financial reason to want Intel to prosper further, but I know the demise of AMD would be the worst thing for Intel and consumers, and all points so far look like Zen might be the thing that the market has been looking for.

 

[TurtleCrusher]

The reality is, given their track record, being pessimistic is just living in reality. Being a vocal optimist, and shooting down the possibility of them not living up to the hype, is a bit naive.

It's always a good idea going into the AMD hype train with a bit of caution. As much as you want the opposite to be true, you'll be very disappointing a lot more than not. You do not have to be an Intel fanboy to doubt AMD living up to the hype.

Is it really pessimism to expect an outcome that has repeated itself several times before?

The claims made are completely reasonable. Nothing that has been said is out of the question. Unfounded pessimism is just that. Unfounded.

 

[bystander36]

The claims made are completely reasonable. Nothing that has been said is out of the question. Unfounded pessimism is just that. Unfounded.

But it is not unfounded pessimism.

 

[TurtleCrusher]

But it is not unfounded pessimism.

How is it founded? The change from CMT to SMT will show significant gains by itself in the areas that PD/XV lacked. The massive process shrink (32nm to 14nm?) is a game changer as well.

The only people that are despondent towards Zen is those wearing the blue-tinted glasses.

 

[bystander36]

How is it founded? The change from CMT to SMT will show significant gains by itself in the areas that PD/XV lacked. The massive process shrink (32nm to 14nm?) is a game changer as well.

The only people that are despondent towards Zen is those wearing the blue-tinted glasses.

The last 10ish years is why people are pessimistic. Many of if not most those who are, would happily buy an AMD system if it lives up to the hype. It's not like I don't buy AMD. I just haven't bought an AMD CPU in the past 10 years. I was quite happy with my Athlon x64, but Intel has been better since that purchase, they've just been behind anytime I've looked at a new PC. I've also bought 5870's, 6950, the 9800 Pro as well as other products I don't recall.

I'm not pessimistic because I'm an Intel fanboy. I'm pessimistic because of history.

 

[dark zero]

The last 10ish years is why people are pessimistic. Many of if not most those who are, would happily buy an AMD system if it lives up to the hype. It's not like I don't buy AMD. I just haven't bought an AMD CPU in the past 10 years. I was quite happy with my Athlon x64, but Intel has been better since that purchase, they've just been behind anytime I've looked at a new PC. I've also bought 5870's, 6950, the 9800 Pro as well as other products I don't recall.

I'm not pessimistic because I'm an Intel fanboy. I'm pessimistic because of history.

Currently Phenom II was a really good step. By far, one of the best than Bulldozer... so if they are returning to Phenom, then going SMT, improving latencies and having a really competent nm design.... definately won't reach Skylake levels, but they can reach Broadwell ones and in ST they can be in the middle of Haswell and BW. So I still have faith on AMD, despite the massive glomm.

 

[bystander36]

I do hope you are right.

 

[frozentundra123456]

But it is not unfounded pessimism.

No point arguing any further. If they say it often enough and click their heels together, it has to come true, right?

 

[Arachnotronic]

How is it founded? The change from CMT to SMT will show significant gains by itself in the areas that PD/XV lacked. The massive process shrink (32nm to 14nm?) is a game changer as well.

The only people that are despondent towards Zen is those wearing the blue-tinted glasses.

Yeah, that's it. It couldn't possibly be due to AMD's terrible track record, the fact that it's been laying off employees left and right, and the fact that it's going up against an incredibly strong competitor that has momentum on its side.

No, it's "blue tinted glasses," Intel shills, etc. Of course, judging from your prior posts, you will tell us that your livelihood does depend on Intel doing well but that, because you are so incredibly objective in a forum filled with a bunch of Intel-biased shills, you are reasonable in thinking that Zen and products based on it will be awesome.

Have I got it right?

 

[Tuna-Fish]

There has seemingly been nothing unreasonable said about Zen and it's performance. The IPC increase from Northwood to Conroe was what, 100%? a 40% claim isn't crazy.

40% IPC isn't crazy. As I've pointed repeatedly in this thread, IPC is a design point you get to pick freely, so long as you are willing to trade enough clock speed for it. What is crazy is taking that as "40% performance improvement". Zen is clearly aiming to be a wider, brainier core than BD. Those are also typically slower than narrow, fast cores. You should expect reductions in achievable clock speeds, just like with Northwood to Conroe.

The truth is, we right now know absolutely nothing about the performance of Zen. Performance = IPC * clock speed, and while we know the IPC, clock speeds are unknown. Since Zen is a whole new project, you absolutely cannot predict it's clock speed from the range BD happened to run at.

There is very little of any value in anything anyone here can say about Zen until we hear something concrete from AMD.

 

[LTC8K6]

We don't know the IPC or the clock speed, actually.

The 40% is almost certainly "up to 40%" in the best case scenarios for the chip.

 

[Abwx]

The 40% is almost certainly "up to 40%" in the best case scenarios for the chip.

AMD always talks of averages, they said 4% for EXV, was it up to 4%..?

It was 9-13%, so even people who talk of track record are blatlantly ignoring this record and are instead posting completely made up assumptions that are contradicted by the facts.

Yeah, that's it. It couldn't possibly be due to AMD's terrible track record, the fact that it's been laying off employees left and right, and the fact that it's going up against an incredibly strong competitor that has momentum on its side.

No, it's "blue tinted glasses," Intel shills, etc. Of course, judging from your prior posts, you will tell us that your livelihood does depend on Intel doing well but that, because you are so incredibly objective in a forum filled with a bunch of Intel-biased shills, you are reasonable in thinking that Zen and products based on it will be awesome.

Have I got it right?

You re wrong since you thought that a marketing discourse is worth a mathematical analysis of a process caracteristics when debating such issues, you call this objectivity..?..

Anyway, one number for your insight, GF s 14nm LPP LVT is 20% more efficient at 2.4GHz than the process used for SKL, and the LPP sLVT will extend this lead even further while allowing frequencies well over 3GHz.

 

[LTC8K6]

AMD always talks of averages, they said 4% for EXV, was it up to 4%..?

Marketing speak is marketing speak, whether AMD, Intel, or NV. They all do it.

See Fury.

See Skylake.

Etc.

 

[Abwx]

Marketing speak is marketing speak, whether AMD, Intel, or NV. They all do it.

See Fury.

That s a GPU

See Skylake.

Etc.

AMD is not Intel, so see Piledriver, Steamroller, Excavator and Kabini instead.

Besides, up to 40% wouldnt even worth it, an EXV based CPU would be much more relevant, even if it s just a die shrink of Carrizo but with 8 modules...

 

[LTC8K6]

No, Fury is a video card, Fiji is a GPU.

Just assume the pre-release hype of anything is marketing speak, and you will rarely be disappointed. Once in a while, you will be delighted.

 

[looncraz]

The reality is, given their track record, being pessimistic is just living in reality.

Track record is actually in their favor.

It's really helpful to realize something: Excavator is only 7~10% faster, per core, per clock, than Thuban (Phenom II). It basically ties Penryn (Core 2). Penryn and Thuban both had 3 ALUs. Excavator only has 2. Thuban has 4 AGUs, but they are paired with an ALU and really only two can be used at once, so it is effectively only two AGUs for all three of these CPUs.

So you're starting out with a 50% better ALU (including utilization improvements) compared to Penryn & Thuban. By itself, the improvements that allow that improved performance, placed in a 3+2 configuration akin to Penryn, would yield excellent results. But it would not be 40%, more like 15~20%. That would get Zen to slightly better than Nehalem performance.

Add in another ALU, with proper assignment considerations, and you can push out another 15%. That takes Zen to Sandy/Ivy Bridge territory, without any other improvements being made. That does not take into consideration any cache improvements nor the end of the module penalties, each worth more than 5% for full-time average performance. That, alone, has you scratching at Haswell's IPC.

There is no magic sauce, no unusual expectations, no major advancement AMD has to make, nothing unusual at all. It's just what happens when you take the easy road to performance, as AMD has obviously done with Zen.

In fact, is I see it, Zen should actually have marginally better ILP than Haswell. The pipeline assignments are just generally much better than Intel's usage of ports.

http://looncraz.net/ZenAssignments.html
http://looncraz.net/HswAssignments.htm

 

[NostaSeronx]

The change from CMT to SMT will show significant gains by itself in the areas that PD/XV lacked.

Except, the change to SMT is going backwards.

CMP is complete duplication of all resources.

SMT is replication of a few resources to allow for a second thread to operate. The reason to use SMT is majority of the app_base will not utilize the full resources of x amount of cores. In effect, SMT allows a single core to have execution capabilities of x amount of cores.

CMT is an extension of SMT, while an evolution of CMP.
- A replication of a few resources to allow for a second thread to operate.
- A complete duplication of core resources.

A core is defined by these resources;
- Control Unit
- Instruction Bus
- Datapath
- Memory Bus
Thus, the above is core resources.

Bulldozer 32nm PDSOI CMP 1C/1T -> ~26 mm² die area
Bulldozer 32nm PDSOI CMP 2C/2T -> ~52 mm² die area
Bulldozer 32nm PDSOI SMT 1C/2T -> ~28 mm² die area
Bulldozer 32nm PDSOI SMT 2C/4T -> ~54 mm² die area
Bulldozer 32nm PDSOI CMT 2C/2T -> ~31 mm² die area
Bulldozer 32nm PDSOI CMT 4C/4T -> ~62 mm² die area

The issue with the 15h architecture isn't that its CMT implementation sucked. It is that the L1 Instruction Bus is mismatched with its L0 datapaths.

[Bulldozer/Piledriver]
4 AMD64 ops become
-4 load ops
-4 computational ops
-up to 4 store ops

^-- This implies up to four FP128 FMACs for the L0 vector datapath, and duplicated L0 datapaths to have proper feed for 4 ALUs.
[Steamroller/Excavator]
4+4 AMD64 ops become
-4+4 load ops
-4+4 computational ops
-up to 4+4 store ops
^-- This implies up to eight FP128 FMACs for the L0 vector datapath, and duplicated L0 datapaths to have even more proper feed for 4 ALUs.

The 15h architecture is a high ILP/MLP design limited by low IPC datapaths.

 

[bystander36]

Track record is actually in their favor.

Nothing you said has anything to do with their track record. Perhaps you can explain how their track record has been so favorable.

 

[looncraz]

Nothing you said has anything to do with their track record. Perhaps you can explain how their track record has been so favorable.

Compare Bulldozer to Excavator, and each performance increase claimed vs that which was realized. They always claimed a range, or an average expected IPC gain, and their claims were always quite accurate. AMD never officially claimed that Bulldozer would be faster per clock per thread than Thuban, though one of their employees did repeatedly (JF-AMD)*.

AMD is using a third fewer resources to deliver more performance than they did with Thuban. Now, suddenly, they are doubling the most important of those resources. Put two and two together, and we can see how a 40% IPC increase may even be a conservative figure.

* I argued that Bulldozer could not possibly have higher IPC due to its cache design and latencies once the details came out. I was 100% certain when I received the optimization guide. I argued at length with JF-AMD, publicly, about this at the time. It became quite clear that the design was at a real deficit even before the first benchmarks came in. My performance estimates were within 10% for single threads (I had no idea how CMT would work) - much closer than I expected. I also had no idea how well it would or would not clock.

 

[AtenRa]

Lets keep it simple, lets keep ti technical,

Does the Intel Sales Team (IST) have any technical arguments to counter the 40% increase of IPC with the available technical details of ZEN Architecture we know today???
Because every time the IST talks in this topic is all about AMDs track record and financial BS.

 

[coercitiv]

Lets keep it simple, lets keep it technical

Does the Intel Sales Team (IST) have any technical arguments

This is what happens again and again in these threads: we pretend to talk specs while we call each other names. Both sides. It is incredibly tiresome for people wanting to learn more about any upcoming product to filter this noise constantly.