AMD Zen to Bring Greater than 40% IPC Increase, New Reports Suggests

[marldorthegreat]

Yep. But AMD is AMD and since around 2011 its as if they've gone "fuck logic".

 

Same as "8 cores" when its really 4 modules. A lot of people get sucked in.

They are 8 cores. 4 modules, each with 2 cores.

[Trixanity]

While most people will be expecting a 40% single-thread performance increase, which will not happen if it is below 4.0GHz, fortunately most people who don't understand the statement also think it will be 40% over the FX-8350, not over Excavator. While the IPC difference isn't huge it does give a little bit of leeway for AMD, even if it's clocked at ~3.5GHz it will probably be at least 40% stronger than the FX-8350, and the less informed people will be satisfied

Excavator has a 20% higher IPC than Piledriver. So in theory we're approaching 60% higher IPC than current FX processors. Of course there's the clock speed difference to account for, as you mention. But these two parameters are still not everything. There's much more to account for, so everything is pure speculation.

People should also keep in mind that Summit Ridge is a high end desktop platform without IGP. There is also Raven Ridge APUs that will compete with Intels mainstream offerings such as Pentium, i3 and i5. While APUs have received a bad name, they are essentially similar to many of Intels offering. So that's what they're competing with.

[Humbug]

Around 2011, Lisa Su was not in charge.

 

Look at what she has done for the company in a few years... they've been hundreds of millions in the hole every year (still are), yet now they are approaching (albeit slowly) profitability.

I think her predecessor Rory Read also did a good job. He joined just as the bulldozer debacle was happening. And in the next few years he was able to cut costs and hire good people like Jim Keller and Raja Koduri.

 

[kuddlesworth9419]

It matters quite a bit. (instructions per clock)

 

Hypothetically if a 4Ghz FX-8350 is modified and given a 40% IPC boost then that means it would only need a 2.4Ghz clock-speed to achieve the same performance. So then you have a CPU which is running much quieter and producing less heat. Or alternatively if you clock it back upto 4Ghz and get 40% faster performance for the same TDP. And AMD isn't even talking about 40% IPC boost above piledriver, they are talking about 40% over excavator.

 

However there are other factors which come into play. Particularly how well does Zen clock. Hypothetically if it cannot exceed 3Ghz then it will be well faster than AMD's current line-up but nothing earth shattering. On the other hand if it can do 4Ghz + more overclock headroom combined with the IPC gain then AMD is back in the game... But we just don't know right now.

Do we have an estimated date for when Zen comes out. Or at least any reviews from the big sites and You Tuber's. I am quite interested to see what happens. I think if Zen fails that's pretty much it for AMD in the CPU market.

[Trixanity]

Do we have an estimated date for when Zen comes out. Or at least any reviews from the big sites and You Tuber's. I am quite interested to see what happens. I think if Zen fails that's pretty much it for AMD in the CPU market.

Q4. Anything else is speculation.

I think we'll see it in November. Early December the latest but that would cut it a bit too close, so they'll try to get it rolling as soon as possible but testing and tweaking takes time and they need to be able sell products from day one (and lots of them), so they need a buffer to avoid being out of stock. So I think they'll have a large enough stock to ship by November.

[SolarNova]

40% ...maybe a bit more, improved IPC over AMD 'current' CPU's. hmm that isn't going to be enough. That 'might' bring them in line with Intel's current line up or previous line up, but not Intel upcoming line up. So AMD will still be slower. Not unless they clock to very high frequencies, like 5ghz +.

[Trixanity]

40% ...maybe a bit more, improved IPC over AMD 'current' CPU's. hmm that isn't going to be enough. That 'might' bring them in line with Intel's current line up or previous line up, but not Intel upcoming line up. So AMD will still be slower. Not unless they clock to very high frequencies, like 5ghz +.

They don't have to beat Intel and at this point it's hard to do, particularly in the long term.

They just have to deliver a product that gives customers an incentive to buy it. For example that could mean slightly lower performance but at a better price meaning that your performance per dollar is better.

I could imagine a Zen/Polaris APU would be a much better purchase than any offering Intel can come up with particularly if you need graphical horsepower or if you need lots of cores but you don't wanna pay 999 dollars you can save some hundred dollars by going AMD.

I'm simply saying they don't necessarily need to have a faster product and all projections say they will be at least a few % behind some of Intel's current offerings but are you willing to pay the premium for a few %?

I also doubt we'll see 5 ghz or even 4.5 ghz products. The trade off in efficiency for the average customer isn't worth it.

I can imagine some SKUs boosting past 4 ghz but that's boost. And those SKUs won't have 8 cores.

[Humbug]

^^To add to above most workloads are not extremely CPU limited. The problem right now is that AMD is so much slower that they cannot sell at profitable margins or in decent quantities.

 

Which means that if they achieve this 40% IPC boost combined with good clockspeeds few people will be able to tell the difference between Intel and AMD Zen outside of synthetic benchmarks. Let's take the example of AAA games; they are going to be mostly GPU limited so even if Zen is a bit slower than Intel it will probably be just as fast in games or at most 2-3fps slower. Even more the case with DX12 and Vulkan coming into the picture further optimizing CPU workloads... So AMD with a solid product and competitive pricing can start making money on CPUs again. They don't have to outsell Intel juggernaut but they can certainly have a chance to obtain and maintain a profitable market share (which they did in the old days).

[flipped_bit]

Ya I was wondering the same thing. But I mean 40% is still quite a bit.

 

In the past they have said it's from Excavator, which itself is a big improvement on Bulldozer.  Excavator is the core used with Carrizo which may be coming to the new AM4 socket as early as this quarter.

[flipped_bit]

Give us an 8-core CPU with 36-40 PCI-E CPU lanes with around Haswell IPC for $500-600 and they'll sell like hotcakes.

[kuddlesworth9419]

Q4. Anything else is speculation.

I think we'll see it in November. Early December the latest but that would cut it a bit too close, so they'll try to get it rolling as soon as possible but testing and tweaking takes time and they need to be able sell products from day one (and lots of them), so they need a buffer to avoid being out of stock. So I think they'll have a large enough stock to ship by November.

Unless they know the performance isn't good enough in which case they will have limited stock I hope anyway. We wouldn't want them to make too much and n one bothers buying the things. I'm not sure weather or not Zen will be good or not I just hope it will surpass Intel because it needs to surpass Intel in all price points not just the low end. If they don't give us a processor that is much better then what Intel offerer's then what is the point of upgrading. 

[looncraz]

 

Same as "8 cores" when its really 4 modules. A lot of people get sucked in.

 

If you have a problem with AMD calling 4 dual core modules an 8 core CPU, then did you have a problem when Intel called the Pentium D a dual core CPU?  Or when they called Core 2 Quad a quad core CPU?

In both of those examples the CPUs had overhead not appropriate for a true dual or quad core CPU.

 

A core is the hardware required to independently, and concurrently, execute disparate threads in hardware.  In all cases, except for AVX, the FX-8xxx can do exactly that.

[looncraz]

Excavator has a 20% higher IPC than Piledriver. So in theory we're approaching 60% higher IPC than current FX processors. Of course there's the clock speed difference to account for, as you mention. But these two parameters are still not everything. There's much more to account for, so everything is pure speculation.

People should also keep in mind that Summit Ridge is a high end desktop platform without IGP. There is also Raven Ridge APUs that will compete with Intels mainstream offerings such as Pentium, i3 and i5. While APUs have received a bad name, they are essentially similar to many of Intels offering. So that's what they're competing with.

 

You're not too far off, but your math is wrong :angry:

 

If you have 20% improvement, that is 120% of the original performance.  To find out how much more a 40% improvement will bring, you multiply by 1.4.  Which gives 168% of the original performance.  Which is a 68% increase - not 60%

 

In this case, though, my numbers have Excavator at 17.21% higher IPC than Piledriver, which leads to 64.09% higher IPC than Piledriver.

[ManWithBeard1990]

I'll settle for Ivy Bridge IPC and no microcode bugs.

AMD's Excavator-based Bristol Ridge APUs coming out within a few months should already come close.

[Humbug]

Do we have an estimated date for when Zen comes out. Or at least any reviews from the big sites and You Tuber's. I am quite interested to see what happens. I think if Zen fails that's pretty much it for AMD in the CPU market.

Estimated date I have heard is October.

 

Unless they know the performance isn't good enough in which case they will have limited stock I hope anyway. We wouldn't want them to make too much and n one bothers buying the things. I'm not sure weather or not Zen will be good or not I just hope it will surpass Intel because it needs to surpass Intel in all price points not just the low end. If they don't give us a processor that is much better then what Intel offerer's then what is the point of upgrading. 

They are unlikely to surpass Intel in all price points. And people who already have haswell or other modern intel systems will not change anyway (for now).

All AMD needs to do is be in the same performance ballpark. Then people who need to upgrade anyway (people who have old or low end systems) will start considering AMD again as a viable choice. This allows them to sell enough CPUs at decent margins to start making profits again. Also the new more power efficient APUs allow them to get back into the laptop market.

[Dabombinable]

Around 2011, Lisa Su was not in charge.

 

Look at what she has done for the company in a few years... they've been hundreds of millions in the hole every year (still are), yet now they are approaching (albeit slowly) profitability.

 

Saying they went "Fuck logic" is far from right. They  have been playing their cards extremely carefully, with plans that were YEARS ahead of their competitors (APUs, whilst seemingly a failure, they are great for OEM systems. Low cost and high power for the price. GCN anyone? DX12 ready in 2012, 4 years before the first DX12 game demo was out)...

 

The only people who said "Fuck logic" was the ones in charge pre Bulldozer. The ones who though bulldozer was ACTUALLY a good idea... those people said "fuck logic"

 

 

Also, depending on the CPU workload it IS 8 "cores"... Although some workloads rely more on float, and as such it would be only 4 modules.

ALU are not cores. They are ALU and only part of any CPU. I've told you before the correct terminology for CMT is * Modules * Threads. With SMT you actually get processing cores.

 

 

They are 8 cores. 4 modules, each with 2 cores.

See above, people not knowing about the parts that make up a processing core let AMD market CMT the way they did.

[xiej]

ALU are not cores. They are ALU and only part of any CPU. I've told you before the correct terminology for CMT is * Modules * Threads. With SMT you actually get processing cores.

? You get hyperthreading with SMT, not cores.

[Dabombinable]

? You get hyperthreading with SMT, not cores.

My i5 4440 is SMT just like my 4790K for example.

[xiej]

The Intel Pentium 4 was the first modern desktop processor to implement simultaneous multithreading, starting from the 3.06 GHz model released in 2002, and since introduced into a number of their processors. Intel calls the functionality Hyper-threading, and provides a basic two-thread SMT engine. Intel claims up to a 30% speed improvement ^[2] compared against an otherwise identical, non-SMT Pentium 4.

https://en.wikipedia.org/wiki/Simultaneous_multithreading

[Prysin]

The Intel Pentium 4 was the first modern desktop processor to implement simultaneous multithreading, starting from the 3.06 GHz model released in 2002, and since introduced into a number of their processors. Intel calls the functionality Hyper-threading, and provides a basic two-thread SMT engine. Intel claims up to a 30% speed improvement ^[2] compared against an otherwise identical, non-SMT Pentium 4.

https://en.wikipedia.org/wiki/Simultaneous_multithreading

The concept of SMT however was invented back in the 70s though

[Glenwing]

My i5 4440 is SMT just like my 4790K for example.

No... SMT is not the opposite of AMD modules, SMT does not mean "normal cores". It means Simultaneous Multi-Threading, which means a core that can run multiple threads simultaneously on each core. Hyperthreading is an implementation of SMT. IBM Power 8 CPUs also have an 8-way implementation of SMT. i5 desktop chips do not have any form of SMT.

[Coaxialgamer]

If? Expect an 8-core Zen CPU to cost less than a 5820K.

 

AMD is not Intel.

i hope so ...

[Trixanity]

You're not too far off, but your math is wrong :angry:

If you have 20% improvement, that is 120% of the original performance. To find out how much more a 40% improvement will bring, you multiply by 1.4. Which gives 168% of the original performance. Which is a 68% increase - not 60%

In this case, though, my numbers have Excavator at 17.21% higher IPC than Piledriver, which leads to 64.09% higher IPC than Piledriver.

I have dyscalculia. My math is excused :-)

However I was under the impression, if we assume Bulldozer is 100%, that Piledriver was roughly 108-110%, Steamroller 115% and Excavator 128% or so.

I'm not doing math to get my numbers for existing architectures. Getting them from AMDs estimates as well as the performance charts various reviewers have made. The math fail regarding the jump from the existing to Zen is all mine.

However, I'm sure your math is correct. :-)

Regarding CMT, I don't think there is any conclusive proof that it's bad.

Sure, the CMT implementation we've seen from AMD wasn't too great but that doesn't mean that the concept is bad.

Now is SMT better than CMT? So far it would seem so since most of the big players go down that route including AMD, Intel and IBM.

Intels Saltwell Atom SoC was SMT and it sucked. Does that mean SMT sucks? No.

AMDs CMT adventure was a colossal failure but I doubt it was based solely on it being CMT.

[niofalpha]

If? Expect an 8-core Zen CPU to cost less than a 5820K.

 

AMD is not Intel.

Then you should expect AMD to lose money, fast... Money for R&D is probably at least 1 Billion, plus costs for distribution, manufacturing, retailers cut, e.t.c... 

They're going to need to maximize their profits to stay afloat, meaning they can't afford to have small profit margins . I'm honestly going to be surprised if an 8 core sells for less than 800$... They listed the 9590 for 1000$ at release... So don't be surprised when it's over 800$.

[DocSwag]

There are 2 things this could mean. The first is a quad core zen chip at 4ghz is about 1 1/2 times better than a 4ghz quad core (two module) Kaveri or Carrizzo chip including the new system getting rid of modules and SMT.

The other thing it could mean is a 4ghz amd chip has about 1.4-1.5 times better single core performance than a 4 ghz Carrizzo or Kaveri chip. Add that to the fact that they are getting rid of the module system (which should result in at least a 30% multi core increase) as well as smt (another 20-30%) means single core performance will get a 40-50% jump and multi core score will basically double. Currently the x4 860k, a 4ghz chip, gets about a 2700 on geek bench single core, with Carrizzo probably capable of 10-20% that (around a 3000). Basically I am getting that a 4 ghz zen chip will get a 4500 single core score (on par with the 6700k), and chances are you can get an 8 core version of that (the fx 8350 has around same tdp as as extreme edition chips) that will deliver a multi score of over 35k, probably around 40k, that could be more efficient than Intel ones (amd seems to have had to squeeze as much as they could out of 28 nm, so 14 nm finfet will help power consumption a lot). Basically a 4 ghz 5960x with around a 100 watt tdp that costs around the price of a 6700k or 5820k. Wow.

Sign me up! I'll sell my 6700k if this is true!