Full Skylake reveal result? Waiting for Zen.

[Enigmoid]

Interestingly, I think AMD can get 33% IPC improvement almost for free.

The module design is suppose to be about 85~90% efficient (each core performing about 85~90% of its optimum). The cache system is horrendous, and cost performance from phenom II's cache system (about 5~6% it seems, load dependent). Bulldozer integer cores are massively narrow, with only two ALUs and two AGUs.

The module design is 85-90% efficient when fully loaded. With only 1 core active, that core gets full access to the front end, its own integer execution units, and the FPU. Steamroller and excavator have improved this to some extent with each SR core getting its own 4-wide decoder (way more than it needs). ST efficiency will not improve in single thread workloads (per module) by removing CMT. If you took off the other integer core, L1 instruction cache, and decoder you would see the same performance as running 1 thread in a module (your core is no longer CMT).

That is a 30~42% improvement right there, with no other changes. Of course, these aren't all fully cumulative, so AMD will have had to make more changes to fully flesh out a 40% improvement. I just think many people underestimate just how much of a burden the front end in a Bulldozer CPU has to process the load for two cores. Sure, it can do it, which is amazing, but it can't do it without adding retirement latency, which reduces performance in MANY different types of workloads.

That is 2 core performance, NOT single thread IPC.

http://www.anandtech.com/show/5057/the-bulldozer-aftermath-delving-even-deeper

Nice explanation.

 

[IntelUser2000]

Intel:
350nm: November 1995 (Pentium Pro)
250nm: January 1998 (Pentium II Deschutes)
180nm: October 1999 (Pentium III Coppermine)
130nm: July 2001 (Pentium III Tualatin)
90nm: Feb 2004 (Pentium IV Prescott)
65nm: January 2006 (Pentium IV Cedar Mill)
45nm: September 2007 (Core 2 Wolfdale)
32nm: January 2010 (Clarkdale)
22nm: April 2012 (Ivy Bridge)
14nm: September 2014 (Broadwell)
10nm: 2017 (Cannonlake)

(Some corrections)

So I think the scenario is more the opposite of what you're saying. In 1995 there wasn't much gap and at times they were neck and neck. Now Zen will be 2 years behind in a process that's the same in name only, in reality being in between Intel's 22nm and 14nm. So more like 3 years behind.

I am not counting it just in terms of timelines. They've had process to process lead that can translate to additional time alone.

I've looked at transistor performance for quite some time, and even during the 0.18u SOI vs 0.18u the difference was quite large. For quite some time though, they've been driven by other factors that hid the true advantage. The artificial crippling thanks to EPIC, the everything-be-damned clockspeed approach.

While it seems it doesn't always translate into clockspeeds, the faster transistor could have been the culprit for their large perf/clock lead. Now, you see with their 14nm and 22nm transistors that peak performance gain essentially stops. Intel had their "Zen" back with the Core uArch. I find it very interesting even AMD shows that after Zen, the gain starts to flatten. Everyone reaches the same destination regardless of the way they went.

 

[frozentundra123456]

I think you got that wrong. Intel had their "Conroe" moment, all we have seen from AMD is their "Bulldozer" moment. AMD's "Zen" moment has not happened yet, so nobody realistically can say what it will bring.

But I do find it humorous that people assume because Bulldozer was so bad, it will be easy to correct all the mistakes and get a huge ipc gain. If it was that easy, why did they not do it with Bulldozer? And one cannot use the process node as an excuse, because Intel "Conroe'd" AMD on pretty much the same node.

 

[Arachnotronic]

I think you got that wrong. Intel had their "Conroe" moment, all we have seen from AMD is their "Bulldozer" moment. AMD's "Zen" moment has not happened yet, so nobody realistically can say what it will bring.

But I do find it humorous that people assume because Bulldozer was so bad, it will be easy to correct all the mistakes and get a huge ipc gain. If it was that easy, why did they not do it with Bulldozer? And one cannot use the process node as an excuse, because Intel "Conroe'd" AMD on pretty much the same node.

What's even more interesting is that Intel had their "Conroe moment" in 2006 and has been aggressively building on that success while AMD has been circling the drain since then.

 

[2is]

But I do find it humorous that people assume because Bulldozer was so bad, it will be easy to correct all the mistakes and get a huge ipc gain. If it was that easy, why did they not do it with Bulldozer? And one cannot use the process node as an excuse, because Intel "Conroe'd" AMD on pretty much the same node.

I find it humorous that you think bulldozer has been the only disappointing CPU from AMD since Conroe's introduction.

 

[TheProgrammer]

I think you need to stop posting entirely. Your entire argument revolves around insults and ignoring facts.

You are no developer, if you were you wouldn't resort to insults when someone challenges your argument based on feelings with facts.

Again Intel has offered better performance within my budget (and looking at the signatures on the forum, a good many others) than AMD for a decade.

LOL. Ok kid. See ya.

Mind share a scenario where TCO for Bulldozer server make sense?

It was a few years back when I was contracted to work the backend services platform for a game everyone here has heard of or plays. I can't say what they ended up doing after our POC, but it was early days for this work and we were looking for a single server. A systems team would scale out the hardware later on but my job was to ensure the backend services were proven to work and scale. For the server we were running and the way it was designed, our tests showed we achieved higher concurrency (roughly 900 connections a second IIRC) and AMD was the better choice.

When you scale out, the power efficiency probably really starts to hurt, but that wasn't relevant to us. We just needed as much concurrency as possible at the time. Some games I worked on, a single server like this would be enough other times (this case) it was scaled out for sure. The game is too big at this point.

There is no 1 size fits all answer like people desperately want to be the case when it comes to the interaction between hardware and software. It has to be profiled, period. I tune software everyday whether it's low level bit mashing or algorithmic.

I'm building things that aren't assisted by Anand's arranged benchmarks. Basically if you're looking at real work, you look at it more seriously. If you're just posting on Anandtech and watching Youtube, it really makes no difference if you're on AMD, Intel or VIA.

 

[AtenRa]

Wow did those goalposts move fast.

Really ??? Have you made a DX-12 gaming conclusion out of a single Alpha benchmark ??

 

[witeken]

I am not counting it just in terms of timelines. They've had process to process lead that can translate to additional time alone.

I've looked at transistor performance for quite some time, and even during the 0.18u SOI vs 0.18u the difference was quite large. For quite some time though, they've been driven by other factors that hid the true advantage. The artificial crippling thanks to EPIC, the everything-be-damned clockspeed approach.

While it seems it doesn't always translate into clockspeeds, the faster transistor could have been the culprit for their large perf/clock lead. Now, you see with their 14nm and 22nm transistors that peak performance gain essentially stops. Intel had their "Zen" back with the Core uArch. I find it very interesting even AMD shows that after Zen, the gain starts to flatten. Everyone reaches the same destination regardless of the way they went.

It's always that talking about the transistor, but the interconnect is at least just as significant. And Intel is also leading there since 14nm, with air gaps.

 

[AtenRa]

But I do find it humorous that people assume because Bulldozer was so bad, it will be easy to correct all the mistakes and get a huge ipc gain. If it was that easy, why did they not do it with Bulldozer? And one cannot use the process node as an excuse, because Intel "Conroe'd" AMD on pretty much the same node.

ZEN and Bulldozer architectures are completely different, different design goals, different architectures, different characteristics.
Bulldozer was not aiming at high IPC, the goal was to have the highest throughput at the lowest die size and lower power consumption possible.
ZEN design goals are different because they now include a higher IPC as well . Because they will not use a CMT design, they can beef up each core (wider, more execution units per core) and that will increase IPC. They also changed the Cache design making it faster with lower latency. All those changes will bring a tremendous increase in IPC over the Bulldozer (excavator) designs.

 

[mohit9206]

I have a feeling when Zen APUs arrive and the first reviews start trickling in, once everyone knows that it does not beat current gen Intel, they will be disappointed and choose Intel. But they don't realize that AMD's goal with Zen is not to beat Intel but just to bridge the gap as much as they can.
Also ideally Zen should have come out this fall not next year's fall. AMD wasted too much time with the current architecture. Right now if i want an AMD, my only options are A8-7600 and the Fx6300 and while both are pretty good value for money and offer decent performance, the upcoming Skylake i3 will be overall better for gaming.
So if Zen came out this year, we could have had good competition between Zen and Skylake i3.

 

[ShintaiDK]

Really ??? Have you made a DX-12 gaming conclusion out of a single Alpha benchmark ??

You already made the conclusion that 8 cores will be more or less minimum for DX12 games and that AMD will be king with Zen. yet the burden is on your shoulders. First DX12 benchmarks show no such need.

 

[StrangerGuy]

Again Intel has offered better performance within my budget (and looking at the signatures on the forum, a good many others) than AMD for a decade.

Remember when AMD had the gall to charge a higher asking price than a E6600 for their X2 6000+ when its practically worse in every single aspect except for raw clock speed? Reviewers were being overly kind for calling it "competitive"...Competitive my royal butt.

 

[mohit9206]

Remember when AMD had the gall to charge a higher asking price than a E6600 for their X2 6000+ when its practically worse in every single aspect except for raw clock speed? Reviewers were being overly kind for calling it "competitive"...Competitive my royal butt.

Only then most tech websites were neutral. Today most tech websites have been bought out by Intel and Nvidia.
Skylake is 2.3% faster than Broadwell and Anandtech among other sites still gives it a rave review. Never in a million years would respected Mr. Anand Lal Shimpi would give a new processor generation a recommended rating if it was only 2%faster than previous generation.

 

[Fjodor2001]

I have a feeling when Zen APUs arrive and the first reviews start trickling in, once everyone knows that it does not beat current gen Intel, they will be disappointed and choose Intel.

I don't think anyone expects Zen to beat Intel in ST performance, so they will not be disappointed if it doesn't either.

Most people are just hoping it'll come close to Intel IB/Haswell ST performance level, but much better in MT performance since it's 8 cores compared to 4 cores for Intel mainstream desktop CPUs. If AMD succeeds on that, I think Zen will sell well. And also a future APU with Zen and possibly HBM memory will likely also sell very well.

 

[mrmt]

When you scale out, the power efficiency probably really starts to hurt, but that wasn't relevant to us. We just needed as much concurrency as possible at the time. Some games I worked on, a single server like this would be enough other times (this case) it was scaled out for sure. The game is too big at this point.

In our experience the extra power consumption of Bulldozer processors, once you factor in datacenter costs tends to overwhelm whatever advantages you get from the smaller acquisition costs, and that has been true since SNB-EP for most cases. Since IVB-EP there are 0 cases where AMD makes sense. I know that AMD still had some server sales since 2011, but I wouldn't be able to point out which cases AMD chips made sense for servers. I would be rather curious to see one of these cases.

I'm building things that aren't assisted by Anand's arranged benchmarks. Basically if you're looking at real work, you look at it more seriously. If you're just posting on Anandtech and watching Youtube, it really makes no difference if you're on AMD, Intel or VIA.

Seconded, but in terms of our server environments AMD processors fare even worse than in anandtech reviews. While in Anand's reviews we see a gap in performance and power consumption, the results we get is that even if AMD *paid* us to use their processors it wouldn't be worth it.

 

[Burpo]

Knights Landing by year end..
http://www.theplatform.net/2015/07/13/momentum-building-for-knights-landing-xeon-phi/

" a whopping 72 cores across its over 8 billion transistors"
"and out of the Knights Landing’s cores, which have two vector math units each, and the near HBM memory will deliver more than 400 GB/sec of memory bandwidth."
"when implemented as a standalone processor, will come with two ports of Intel’s Omni-Path interconnect on the package"

 

[PliotronX]

I don't think anyone expects Zen to beat Intel in ST performance, so they will not be disappointed if it doesn't either.

Most people are just hoping it'll come close to Intel IB/Haswell ST performance level, but much better in MT performance since it's 8 cores compared to 4 cores for Intel mainstream desktop CPUs. If AMD succeeds on that, I think Zen will sell well. And also a future APU with Zen and possibly HBM memory will likely also sell very well.

This! Practically all of my friends are already diehard AMD fans, not even caring about ST performance or power efficiency. They're going to go apes*** no matter how Zen does.

 

[AtenRa]

You already made the conclusion that 8 cores will be more or less minimum for DX12 games and that AMD will be king with Zen. yet the burden is on your shoulders. First DX12 benchmarks show no such need.

No, i said that you will need more than 4 cores if you want to play games with thousands of units and people with quads may have trouble playing with maximum settings. As far as the 8-core FX, i said it could be close to Quad Core Core i7 at DX-12 games.

On the other hand you said that DX-12 will not bring anything more than DX-11 for the CPUs. Lets wait and see what will turn out to be next year with 16nm GPUs and onward.

 

[Phynaz]

Have you ever developed a AAA DX11 game? I have.
DX11 is not the future BTW.

My son works in the industry as a designer. Please let us know what AAA title you were a developer on and I'll have him provide confirmation.

 

[AtenRa]

My son works in the industry as a designer. Please let us know what AAA title you were a developer on and I'll have him provide confirmation.

Confirmation of what ???

 

[cytg111]

My son works in the industry as a designer. Please let us know what AAA title you were a developer on and I'll have him provide confirmation.

You should end your sentences with something akin to jk or /joking or something similar, otherwise people wont know when you're kidding around.

 

[Phynaz]

Confirmation of what ???

I'm calling shens on him being a AAA Windows game dev.

If TheProgrammer provides us with details (he won't) my son could reach out to the PM for the game with a sample of his posts and ask if the writing style is anything like somebody that was on that particular project.

Barring that, he's just trolling. It's time for him to put-up or shut-up.

 

[Phynaz]

You should end your sentences with something akin to jk or /joking or something similar, otherwise people wont know when you're kidding around.

I would if I were kidding.

 

[mahoshojo]

I PRAY Zen changes that because I don't want a 1 player market, but I'm realistic since AMD has slashed theri R&D budget a LOT and is competing against 2 companies that are capable of outspending them in Nvidia/Intel.

The R&D reduction is kind of always misleading. AMD sold >80% of their analog/IO group to Synopsys, and outsourced their south bridge to ASmedia. It's about 20% or more of AMD's R&D workforce.

 

[Fjodor2001]

I'm calling shens on him being a AAA Windows game dev.

If TheProgrammer provides us with details (he won't) my son could reach out to the PM for the game with a sample of his posts and ask if the writing style is anything like somebody that was on that particular project.

Barring that, he's just trolling. It's time for him to put-up or shut-up.

"writing style"? You mean as in source code writing? Or forum post writing?

How would it be possible to "confirm" your suspicion by that?

And BTW, does your son know all PMs in all Windows AAA game projects?

And regardless, are you really willing to go through all the trouble of contacting your son, and him in turn contacting all the PMs of that project, to what... ? Prove some suspicion you have about a poster on some Internet forum?

You should consider using the /joking tag anyway... or