New Zen microarchitecture details

[Maverick177]

And the official IPC figure (avg, by AMD) between Piledriver and Excavator is 15.5% (PD to SR = 10%, SR to XV = 5%). Also there are cases where Excavator has up to 12% lower IPC than Steamroller and several cases where the two architectures perform the same. In certain workloads Skylake can have up to 108% higher IPC than Excavator.

Where do you think ZEN IPC will fall in?

 

[Arachnotronic]

As per AT the IPC jump from SB to SKL is 25%.

AT's testing seems to have gimped SKL, I have seen game tests for example that show SKL performing much better in terms of perf/clock. Pairing SKL with faster memory seems to help the CPU reach its full potential, something that AT's testing methodology failed to capture, IMO.

 

[inf64]

AT's testing seems to have gimped SKL, I have seen game tests for example that show SKL performing much better in terms of perf/clock. Pairing SKL with faster memory seems to help the CPU reach its full potential, something that AT's testing methodology failed to capture, IMO.

Games are part of the average and even if SKL benefits that much that won't change the average that much. Other uarchitectures benefit from low latency high clocked DDR3 in games as well, Haswell is one example(pair it with DDR3-2400+Mhz RAM and see how it gets the similar boost as SKL has from higher end DDR4 memory).

 

[Arachnotronic]

Games are part of the average and even if SKL benefits that much that won't change the average that much. Other uarchitectures benefit from low latency high clocked DDR3 in games as well, Haswell is one example(pair it with DDR3-2400+Mhz RAM and see how it gets the similar boost as SKL has from higher end DDR4 memory).

Not just game tests though, Russian Sensation provided links that showed massive performance scaling with higher frequency memory. AnandTech dropped the ball with its SKL review, IMO.

 

[The Stilt]

Where do you think ZEN IPC will fall in?

I expect it to vary heavily depending on the workload, however the average should be pretty well matched with Sandy & Ivy Bridge.

 

[Maverick177]

I expect it to vary heavily depending on the workload, however the average should be pretty well matched with Sandy & Ivy Bridge.

Wow that's shitty.


Profanity is not allowed in the technical forums,
Markfw900
Anandtech Super moderator

 

[The Stilt]

Wow that's shitty.

Huh?
I don't think it has ever been realistic to expect AMD to be able to make up the ~ seven year lead Intel has, in a single generation?

The most important thing for AMD from now on is to get the clocks on Zen and it's successors to the proper levels. Otherwise much of the good work is wasted.

 

[Abwx]

As per AT the IPC jump from SB to SKL is 25%.



I think Zen will be from 5 to 10% slower than Haswell on average in common desktop workloads. That would make it 10-15% slower than Skylake. In specific optimized workloads like FMA or AVX the gap will certainly be bigger but that is a different story and different segment.

The improvements are concentrated in FP, in Integer, wich is the most relevant for the average consumer, progress is much slower than what could be deducted from the FP + INT aggregated numbers.

 

[Arachnotronic]

Huh?
I don't think it has ever been realistic to expect AMD to be able to make up the ~ seven year lead Intel has, in a single generation?

The most important thing for AMD from now on is to get the clocks on Zen and it's successors to the proper levels. Otherwise much of the good work is wasted.

Can you say anything publicly about the clocks that you expect from Zen?

 

[The Stilt]

Can you say anything publicly about the clocks that you expect from Zen?

My expectations from post #1259 still stand:

I expect Zen cores the have similar IPC as Ivy Bridge, on average. Faster in certain workloads but the average should be quite well matched.

Seeing how much additional effort AMD has taken in regards of Zen power and clock management and how strict the VRM requirement (based on the existing boards) are, I don't feel that Zen will be able to scale too well in any aspect.

Because of that I actually have lowered my expectations for Zen's shipping frequencies from my original estimations (for the "halo" 8C/16T desktop FX), which originally were 3000MHz (±200MHz) base and 3600MHz (±200MHz) maximum boost. At the moment I'd expect 2600MHz (±200MHz) base and 3200MHz (±200MHz) maximum boost. However I have no idea what the clocks will actually be, so as always anyone elses guess is just as good as mine.

 

[Arachnotronic]

My expectations from post #1259 still stand:

Thanks, your posts are always great, Stilt

 

[moonbogg]

2600/3200 MHz would land Zen squarely in the trash bin. Who would buy something like that with sandy/ivy IPC? It would seriously be Bulldozer all over again as far as desktop usage is concerned. That can't happen and I wouldn't expect it to. It simply must be clocked higher. There is just no way around it from a consumer standpoint. Servers? I don't know and couldn't possibly care any less about servers.
Nothing on this pearly blue planet could possibly be any more boring to me than a server. If Zen is only going to be good for servers, then they shouldn't even try to sell these chips to consumers. They will get laughed at and owned hard at 2600/3200. I got more faith. I expect clocks to be north of 3ghz for both base and boost with OC's past 4ghz as typical.

 

[The Stilt]

Currently the ES CPUs AMD has sent to partners / ODMs for evaluation are clocked 2.8GHz base, 3.05GHz ACB and 3.2GHz SCB.

ACB = all core boost, SCB = single core boost.

40% IPC increase over Excavator will put the IPC to Sandy / Ivy Bridge levels, depending on the workload. That's the figure AMD decided to give out and that's the figure I use in my estimations. Naturally the actual IPC will vary depending on the workload.

EDIT: Performance -> IPC.

 

[Insert_Nickname]

I expect it to vary heavily depending on the workload, however the average should be pretty well matched with Sandy & Ivy Bridge.

That sounds about right, and fits with my own expectations.

Who would buy something like that with sandy/ivy IPC?

Here's one. At the right price, 8 physical cores @ Ivy-level IPC would be nice. Its not like I expect Zen to beat SKL, let alone KBL, head-to-head.

 

[Arachnotronic]

2600/3200 MHz would land Zen squarely in the trash bin. Who would buy something like that with sandy/ivy IPC? It would seriously be Bulldozer all over again as far as desktop usage is concerned. That can't happen and I wouldn't expect it to. It simply must be clocked higher. There is just no way around it from a consumer standpoint. Servers? I don't know and couldn't possibly care any less about servers.
Nothing on this pearly blue planet could possibly be any more boring to me than a server. If Zen is only going to be good for servers, then they shouldn't even try to sell these chips to consumers. They will get laughed at and owned hard at 2600/3200. I got more faith. I expect clocks to be north of 3ghz for both base and boost with OC's past 4ghz as typical.

Dude, this is a server die being repurposed as a HEDT chip; it wasn't designed for enthusiasts at the get go. Set your expectations accordingly.

 

[moonbogg]

Dude, this is a server die being repurposed as a HEDT chip; it wasn't designed for enthusiasts at the get go. Set your expectations accordingly.

Consumers don't know that and don't care, including me. AMD has been saying, "We're back! Here comes Zen! Faster than Broadwell-E! Desktop enthusiast chip inbound!". If the thing gets stuck at 3ghz with Ivy IPC, its a fail, even with 8 cores. Someone who wants 8 slow Ivy cores could have already had 6 fast Ivy cores that perform the same or better for several years already. This would be too ridiculous to be true. It goes north of 4Ghz or bust.
Lets not pretend like AMD isn't positioning this as a high end, enthusiast desktop CPU, because they are. If its slower than an OC'd Sandy, they are finished in the eyes of enthusiasts and gamers and many more people as well. They can't blow it that big this time, not after Bulldozer. There's just no way. They won't recover from it ever. It will be the death blow and they know it. They'd be better off avoiding any sales on the desktop altogether.

 

[Arachnotronic]

Consumers don't know that and don't care, including me.

They will know soon enough.

AMD has been saying, "We're back! Here comes Zen! Faster than Broadwell-E! Desktop enthusiast chip inbound!".

That's the job of marketing/PR -- hype their products, pump up the stock. Don't be alarmed, but the sky is also blue

If the thing gets stuck at 3ghz with Ivy IPC, its a fail, even with 8 cores. Someone who wants 8 slow Ivy cores could have already had 6 fast Ivy cores that perform the same or better for several years already. This would be too ridiculous to be true. It goes north of 4Ghz or bust.

Intel is a giant that spends more in R&D in a year than AMD has in a decade. Why is it a surprise that Intel is really good at what it does?

Lets not pretend like AMD isn't positioning this as a high end, enthusiast desktop CPU, because they are. If its slower than an OC'd Sandy, they are finished in the eyes of enthusiasts and gamers and many more people as well. They can't blow it that big this time, not after Bulldozer. There's just no way. They won't recover from it ever. It will be the death blow and they know it. They'd be better off avoiding any sales on the desktop altogether.

There will be people who buy it because they hate Intel, think their few hundred bucks will save AMD and "promote competition," and those who simply want 8 cores on the cheap. Just like the FX chips.

 

[moonbogg]

There will be people who buy it because they hate Intel, think their few hundred bucks will save AMD and "promote competition," and those who simply want 8 cores on the cheap. Just like the FX chips.

I know people have hate for certain companies, but when push comes to shove, do you really think they will spend money on a crappy product instead of a good one? I hate Nvidia but I just bought a huge ass Gsync monitor. The reason is because there wasn't any other good option or choice. No way will someone part with a bunch of cash on a bad product just because they don't like a company. Speaking your mind is one thing, but screwing yourself over with a piece of expensive garbage is a whole other tin of Almond Roca.

 

[Arachnotronic]

I know people have hate for certain companies, but when push comes to shove, do you really think they will spend money on a crappy product instead of a good one?

Yes, absolutely.

I hate Nvidia but I just bought a huge ass Gsync monitor. The reason is because there wasn't any other good option or choice.

Congrats, PG348Q is an amazing monitor, you should start a thread in the displays subsection and tell us about your experience with it

No way will someone part with a bunch of cash on a bad product just because they don't like a company. Speaking your mind is one thing, but screwing yourself over with a piece of expensive garbage is a whole other tin of Almond Roca.

These people convince themselves that they aren't buying a bad product. That's why you had so many people back in 2011/2012 talking about how AMD FX chips were "smoother" than the comparable Intel chips even though those chips got absolutely wrecked in benchmarks. Or why you will find people who try to find these corner cases to justify why a chip from their preferred-but-objectively-worse-off vendor is "superior" to the chip from the vendor they don't like.

It's not rational, but it's human nature.

 

[moonbogg]

Well none of this matters anyway because Zen is faster than Broadwell-E and will be the choice of gamers and high end desktop enthusiasts.

 

[Sheep221]

Well none of this matters anyway because Zen is faster than Broadwell-E and will be the choice of gamers and high end desktop enthusiasts.

I really want this to happen so I can start buying i5s in bulk for $50 a piece.

 

[krumme]

In this cb test skylake is 11% faster than sb at 4.8Gzhz as i can tell?
http://www.bit-tech.net/hardware/2015/08/05/intel-skylake-review/8
It seems to me progress is always talked up whatever the arch is portrayed.
And again its fp heavy test.
Making an average across avx2 whatever and loads of heavy fpu test seems wrong imo as it skews the results to even most enthusiast desktop users. Its seems better to ask what kind of performance is needed. If the result is 8c zen is not the most obvious choice its hardly a surprice but making this to a fp compettition is just the wrong way.

 

[Arachnotronic]

In this cb test skylake is 11% faster than sb at 4.8Gzhz as i can tell?
http://www.bit-tech.net/hardware/2015/08/05/intel-skylake-review/8
It seems to me progress is always talked up whatever the arch is portrayed.
And again its fp heavy test.
Making an average across avx2 whatever and loads of heavy fpu test seems wrong imo as it skews the results to even most enthusiast desktop users. Its seems better to ask what kind of performance is needed. If the result is 8c zen is not the most obvipus choice its hardly a surprice but making this to a fp compettition is just the wrong way.

SKL at 4.8GHz is 23% faster than SNB at 4.8GHz in the test you linked.

 

[Arachnotronic]

Well none of this matters anyway because Zen is faster than Broadwell-E and will be the choice of gamers and high end desktop enthusiasts.

We'll see, but I doubt it.

 

[The Stilt]

In this cb test skylake is 11% faster than sb at 4.8Gzhz as i can tell?
http://www.bit-tech.net/hardware/2015/08/05/intel-skylake-review/8
It seems to me progress is always talked up whatever the arch is portrayed.
And again its fp heavy test.
Making an average across avx2 whatever and loads of heavy fpu test seems wrong imo as it skews the results to even most enthusiast desktop users. Its seems better to ask what kind of performance is needed. If the result is 8c zen is not the most obvipus choice its hardly a surprice but making this to a fp compettition is just the wrong way.

23.1% you mean?