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

[MageTank]

No, Zen is 40% improvement over Excavator. It's been said at least a dozen times in this thread alone. Excavator is around 25% faster ipc than bulldozer, making Zen around 65% faster clock for clock with bulldozer. This brings it theoretically either neck and neck with, or slightly faster than Haskell ipc. However, the New architecture changes, and clock speeds need to be known before we can say what real world perfmance might rival.

You didn't read his post, did you? He said Excavator is 40% slower IPC than Zen. That is the same as saying Zen IPC is 40% faster than Excavator. 

[DocSwag]

You didn't read his post, did you? He said Excavator is 40% slower IPC than Zen. That is the same as saying Zen IPC is 40% faster than Excavator. 

Excavator having 40% slower IPC than Zen and Zen having 40% more IPC than Excavator are different. If Zen has 40% more IPC, then Excavator has about 28% less IPC than Zen. If Excavator has 40% slower IPC than Zen then Zen has about 66% higher IPC than Excavator. It's basic algebra.

 

No, Zen is 40% improvement over Excavator. It's been said at least a dozen times in this thread alone. Excavator is around 25% faster ipc than bulldozer, making Zen around 65% faster clock for clock with bulldozer. This brings it theoretically either neck and neck with, or slightly faster than Haskell ipc. However, the New architecture changes, and clock speeds need to be known before we can say what real world perfmance might rival.

Also you can't add IPC like this. If Excavator=Bulldozer times 125% and Zen = Excavator times 140% then Zen = Bulldozer  times 125% times 140% which is actually 180%, more than 165% (which is even better)! Check the math. Percents don't always add like this.

[MageTank]

Excavator having 40% slower IPC than Zen and Zen having 40% more IPC than Excavator are different. If Zen has 40% more IPC, then Excavator has about 28% less IPC than Zen. If Excavator has 40% slower IPC than Zen then Zen has about 66% higher IPC than Excavator. It's basic algebra.

 

Also you can't add IPC like this. If Excavator=Bulldozer times 125% and Zen = Excavator times 140% then Zen = Bulldozer  times 125% times 140% which is actually 180%, more than 165% (which is even better)! Check the math. Percents don't always add like this.

Yeah, i goofed up. MageTank and his mortal nemesis... Grade School Mathematics!

 

Excavator = 1. Zen = 1.4. 1.4 -40%(0.56) = 0.84. 66% difference. If Zen = 1.4, and Excavator = 1, then 1.4 - 28%(0.392) = 1. Also, please ignore the order of operations, i used parenthesis to tell people exactly what the % values were out of those numbers. Again, math is my nemesis. Ironic, given what I do on a daily basis.

 

Thank you for taking me back to school, DocSwag!

[DocSwag]

Yeah, i goofed up. MageTank and his mortal nemesis... Grade School Mathmatics!

 

Excavator = 1. Zen = 1.4. 1.4 -40%(0.56) = 0.84. 66% difference. If Zen = 1.4, and Excavator = 1, then 1.4 - 28%(0.392) = 1. Also, please ignore the order of operations, i used parenthesis to tell people exactly what the % values were out of those numbers. Again, math is my nemesis. Ironic, given what I do on a daily basis.

 

Thank you for taking me back to school, DocSwag!

lol. Yeah it's easy to take the urge to skip da math! I've done it before too.

[MageTank]

lol. Yeah it's easy to take the urge to skip da math! I've done it before too.

Yeah, now I just need to apologize to @ace_cheaply for calling him out with my terrible math skills. Sorry Ace. Now, back to overclocking ram with terrible math skills. 

[Humbug]

No, Zen is 40% improvement over Excavator

Ya, thanks. Corrected.

Stupid mistake.

[DocSwag]

hi llt community. if anyone has the equipment and time to do this please do: could someone benchmark a fx-8350 and i5-6600k with only one core enabled and underclock both processors to 3.5GHz. this would allow us to see relative ipc (instructions per clock) and see if a 40% increase in performance will make the new zen architecture cpu's performance competitive with the skylake cpu's. please do a variety of benchmarks for variety!edit: a further breakdown of single threaded performance http://www.cpu-world.com/Compare/441/AMD_FX-Series_FX-8350_vs_Intel_Core_i5_i5-6600K.html#bench 

The fx 8350 has identical clock speeds as the 6700k. If you look at the fx 8350 vs 6700k on geekbench the 6700k is rougly twice the 8350. With everyones' caluclations though we are seeing that Zen will have probably at least 1.8 times higher IPC, if not more, than the fx 8350(please correct me if I am wrong), which would put Zen pretty high up there. As well there are other improvements including better L2 Cache as well as getting rid of the module system that will affect it as well. The module system shouldn't make a difference but you never know. L2 Cache could affect it though. As well, the FX 8350 is more of a budget part while the 6700k is a top end one, so the 6700k will probably be paired with better ram and a better motherboard which also would affect the scores. I believe that Zen will probably have similar if not better IPC than Skylake, and will be an amazing competitor against Kaby Lake as well as even Skylake-E. I really hope Zen will bring AMD back, and potentially even help them become what they were in the mid-2000s, when they almost had more market share than Intel.

[looncraz]

hi llt community. if anyone has the equipment and time to do this please do: could someone benchmark a fx-8350 and i5-6600k with only one core enabled and underclock both processors to 3.5GHz. this would allow us to see relative ipc (instructions per clock) and see if a 40% increase in performance will make the new zen architecture cpu's performance competitive with the skylake cpu's. please do a variety of benchmarks for variety!edit: a further breakdown of single threaded performance http://www.cpu-world.com/Compare/441/AMD_FX-Series_FX-8350_vs_Intel_Core_i5_i5-6600K.html#bench 

 

Zen should have ~64% higher IPC than the FX-8350.  That should leave the 6600k (Skylake) about 10~15% ahead.  Floating point, however, remains a major wildcard - as does clock speed.

[MrDynamicMan]

wow, I left this thread alone for a while, and good job LTT. we managed to keep it factual, concise, very well informed and without flaming -too- hard. +1.

[Dabombinable]

wow, I left this thread alone for a while, and good job LTT. we managed to keep it factual, concise, very well informed and without flaming -too- hard. +1.

Kind of hard when your looking to replace a laptop like this:

And you don't actually have many options in regards to the hardware ATM.

[MrDynamicMan]

Kind of hard when your looking to replace a laptop like this:

finally off the 2010 drivers.jpg

And you don't actually have many options in regards to the hardware ATM.

not on the AMD side.

[AlTech]

40% from bulldozer?   :lol:

 

40+ % from bulldozer (so realistically probably 40-50% better than bulldozer).

 

If this is anywhere close to Haswell's IPC then I'm in.

[Dabombinable]

not on the AMD side.

I'm not on any side like some of the people here-we just want some competition from AMD not "herp a derp, lets just add more cores and GHz".

[MrDynamicMan]

I'm not on any side like some of the people here-we just want some competition from AMD not "herp a derp, lets just add more cores and GHz".

very much agreed.

But then why the problems with replacing the laptop?

I dont follow.

[Dabombinable]

very much agreed.

But then why the problems with replacing the laptop?

I dont follow.

Zen=possibly cheaper laptops. And hopefully I might be seeing more Tonga/Fiji laptops.

[MrDynamicMan]

Zen=possibly cheaper laptops. And hopefully I might be seeing more Tonga/Fiji laptops.

aaaah, I see.

[patrickjp93]

Also remember this article doesn't say 40%, it says MORE than 40%, so this means it will be the same as if not better than Skylake. Also keep in mind AMD has generally been better at hitting higher clock speeds on CPUs which means even higher performance. They developed a 4ghz octa core 125 watt CPU on 32 nm (FX 8350). Granted higher IPC could mean more heat but this still holds weight, so AMD could match Intel in IPC while having higher clocked CPUs.

Higher clocks than Intel on a low-power node and working with FinFET for the first time? Keep dreaming.

[Humbug]

Higher clocks than Intel on a low-power node and working with FinFET for the first time? Keep dreaming.

Doesn't it depend on the architecture as well? I.e. Some architectures inherently clock high, others don't. I guess we have no info on which to speculate re Zen clocks.

[msx68k]

40% from bulldozer?   :lol:

Nop. 40% from Excavator ~= 60% from Bulldozer.

[DocSwag]

Doesn't it depend on the architecture as well? I.e. Some architectures inherently clock high, others don't. I guess we have no info on which to speculate re Zen clocks.

Yeah it does but AMD has typically been better with clock speeds. Just look at how well they optimized 28 nm, even if their architecture wasn't the best. Chances are they have been able to achieve relatively high clock speeds.

[patrickjp93]

Doesn't it depend on the architecture as well? I.e. Some architectures inherently clock high, others don't. I guess we have no info on which to speculate re Zen clocks.

Yes, but there's a reason Vishera could clock so high, and most of it had to do with being on a high-leakage bulk process, and that era is decidedly over. The other part was having only 2 ALUs per core and a shared FPU per module. The control logic wasn't nearly as complex as Intel's for Sandy Bridge which had 3 ALUs per core, one of which was a dedicated FPU.

[Prysin]

Kind of hard when your looking to replace a laptop like this:

finally off the 2010 drivers.jpg

And you don't actually have many options in regards to the hardware ATM.

meh, why bother with laptops. Get a real PC....

 

or hell, just build a mITX rig, you can carry it around and if you are really clever adn get something like a 295x2 or Fury X2 (yes, these GPUs work perfectly fine in mITX cases aslong as you have clearance for GPU length + radiator) and a 6700k, and really, you get the most bad-ass system possible

 

One of my best friends bought a 295x2 at the same sale i did. He has it installed in a Corsair 380T alongside a 4790k.... Yeah, the 380T IS on the large side of ITX, but still. Try find a more potent setup that you can effortlessly pick up and move about

[looncraz]

40+ % from bulldozer (so realistically probably 40-50% better than bulldozer).

 

If this is anywhere close to Haswell's IPC then I'm in.

 

40% from Excavator - which should be >70% above Bulldozer.  And, in theory, that is speaking gently at Haswell IPC.

 

Of course, some apps will probably be faster on Zen than Haswell (or even Skylake due to some of Zen's instruction assignment advantages)... while certain others will probably struggle to keep up with Sandy Bridge.

[looncraz]

Doesn't it depend on the architecture as well? I.e. Some architectures inherently clock high, others don't. I guess we have no info on which to speculate re Zen clocks.

 

 

Architecture matters a lot, but going twice as wide, with faster caches, on a first-time use of a barely-tested process?  I think 4GHz stock speeds on a quad core would be an astonishing feat.  AMD will certainly be aiming for that, of course, but accomplishing a new CPU design, on a vastly new process, bringing it to market, and reaching those types of speeds at the claimed level of performance would probably be the single largest jump in CPU performance in at least a decade.

 

Of course, they actually have to make that jump to stand a chance of catching Intel's high-end client CPUs.

[TechGod]

40+ % from bulldozer (so realistically probably 40-50% better than bulldozer).

If this is anywhere close to Haswell's IPC then I'm in.

It's from excavator! Not bulldozer... Jeez. When will you people stop repeating that?