Overclock.net › Forums › Industry News › Hardware News › [OC3D] AMD's Zen will have a "greater than 40%" IPC improvement over Excavator, says Lisa Su
New Posts  All Forums:Forum Nav:

[OC3D] AMD's Zen will have a "greater than 40%" IPC improvement over Excavator, says Lisa Su - Page 52

post #511 of 841
Quote:
Originally Posted by Cyrious View Post

Wouldn't a ~52% IPC increase over Excavator put it right up there next to Broadwell if not playing chicken with Skylake?

It would, yes. And that's still before process improvements. There's so much that can prevent attaining idealized performance, though, that the only safe bet is 40%. Which is probably why AMD chose that number and has been so adamant about sticking with it.

For example, AMD has only stated ">40%" for the server CPUs, when we all know it should be >60%, since Piledriver is the reference - which is two full generations behind Excavator, which is an easy 17% ahead in IPC... They're definitely playing it safe with the numbers. And we should all do so with our expectations.
post #512 of 841
Any word if the Zen+ will also be AM4? So one could just pop and swap a Zen for a Zen+ at a future date?
Intel Build
(13 items)
 
AMD Build
(13 items)
 
 
CPUMotherboardGraphicsRAM
Intel Q9450 at 3.55Ghz Gigabyte GA-X48-DQ6 Sapphire HD 5770 Vapor-X G.SKILL 8GB 1066 DDR2 
Hard DriveOSPowerCase
WD 640GBs 2-Blue 6-Black Windows 7 Ultimate 64-Bit ABS Tagan BZ800 Antec 900 
CPUMotherboardGraphicsRAM
AMD PII X4 720 BE 3.0Ghz BioStar TA790GX 128M Sapphire HD 5770 Crossfire G.SKILL 8GB 1066 DDR2 
Hard DriveOSPowerCase
Twin WD Caviar Blue 640GB Windows 7 Ultimate 64-Bit Antec EA650 650W Antec 300 
  hide details  
Reply
Intel Build
(13 items)
 
AMD Build
(13 items)
 
 
CPUMotherboardGraphicsRAM
Intel Q9450 at 3.55Ghz Gigabyte GA-X48-DQ6 Sapphire HD 5770 Vapor-X G.SKILL 8GB 1066 DDR2 
Hard DriveOSPowerCase
WD 640GBs 2-Blue 6-Black Windows 7 Ultimate 64-Bit ABS Tagan BZ800 Antec 900 
CPUMotherboardGraphicsRAM
AMD PII X4 720 BE 3.0Ghz BioStar TA790GX 128M Sapphire HD 5770 Crossfire G.SKILL 8GB 1066 DDR2 
Hard DriveOSPowerCase
Twin WD Caviar Blue 640GB Windows 7 Ultimate 64-Bit Antec EA650 650W Antec 300 
  hide details  
Reply
post #513 of 841
Quote:
Originally Posted by Seraphic View Post

Any word if the Zen+ will also be AM4? So one could just pop and swap a Zen for a Zen+ at a future date?

I don't think we're that far ahead, but I would imagine so.
Big Timmah
(13 items)
 
  
CPUMotherboardGraphicsRAM
Ryzen 5 1600 Asrock x370 Killer SLI/AC Sapphire Radeon Nitro Fury CORSAIR Vengeance LPX 16GB 3200mhz 
Hard DriveCoolingOSMonitor
PNY 480GB SSD PH-TC12DX Black Windows 10 Pro LG 29inch Ultrawide 
KeyboardPowerCaseMouse
Corsair K70 Thermaltake SMART M Series 850W NZXT S340 White Steel ATX Mid Tower Case Wireless Logitech thing 
Mouse Pad
With a supple pad  
  hide details  
Reply
Big Timmah
(13 items)
 
  
CPUMotherboardGraphicsRAM
Ryzen 5 1600 Asrock x370 Killer SLI/AC Sapphire Radeon Nitro Fury CORSAIR Vengeance LPX 16GB 3200mhz 
Hard DriveCoolingOSMonitor
PNY 480GB SSD PH-TC12DX Black Windows 10 Pro LG 29inch Ultrawide 
KeyboardPowerCaseMouse
Corsair K70 Thermaltake SMART M Series 850W NZXT S340 White Steel ATX Mid Tower Case Wireless Logitech thing 
Mouse Pad
With a supple pad  
  hide details  
Reply
post #514 of 841
Quote:
Originally Posted by looncraz View Post

It would, yes. And that's still before process improvements. There's so much that can prevent attaining idealized performance, though, that the only safe bet is 40%. Which is probably why AMD chose that number and has been so adamant about sticking with it.

For example, AMD has only stated ">40%" for the server CPUs, when we all know it should be >60%, since Piledriver is the reference - which is two full generations behind Excavator, which is an easy 17% ahead in IPC... They're definitely playing it safe with the numbers. And we should all do so with our expectations.
True true. Still, 40% over Excavator sticks it in the Haswell region of performance, and an unlocked 8c/16t at that performance level for sub $700 would make for a good buy for me.

I do have a question for you though, since you seem to actually know your stuff and have the ability to simulate what a possible Zen design would look like in terms of performance. The question being: Assuming the theoretical 52% core design you said as an example earlier, how much of a hit would Zen take from the loss of the L3 cache(s)? I ask because the Exascale Zen APU is rumored to not have its L3 active, and it begs the question on whether it would be a big issue or not.
Leviathan
(17 items)
 
Charred
(10 items)
 
 
CPUMotherboardGraphicsGraphics
Xeon E5-2690 Biostar TPower X79 PNY GTX 660 2GB HP GT 440 OEM 
RAMHard DriveHard DriveHard Drive
Gskill Ripjaws 4x2GB 1600mhz Seagate Barracuda 500GB Seagate Barracuda 1.5TB Western Digital Caviar Blue 640GB 
Hard DriveCoolingOSMonitor
Patriot Pyro 60GB Xigmatek Gaia Windows 7 Ultimate Acer S230HL 
MonitorKeyboardPowerCase
Princeton 1280x1024 19" Logitech K120 Seasonic G550 Xclio Nighthawk 
Mouse
Logitech MX310 
  hide details  
Reply
Leviathan
(17 items)
 
Charred
(10 items)
 
 
CPUMotherboardGraphicsGraphics
Xeon E5-2690 Biostar TPower X79 PNY GTX 660 2GB HP GT 440 OEM 
RAMHard DriveHard DriveHard Drive
Gskill Ripjaws 4x2GB 1600mhz Seagate Barracuda 500GB Seagate Barracuda 1.5TB Western Digital Caviar Blue 640GB 
Hard DriveCoolingOSMonitor
Patriot Pyro 60GB Xigmatek Gaia Windows 7 Ultimate Acer S230HL 
MonitorKeyboardPowerCase
Princeton 1280x1024 19" Logitech K120 Seasonic G550 Xclio Nighthawk 
Mouse
Logitech MX310 
  hide details  
Reply
post #515 of 841
Skylake can be > 80% faster (IPC) over Excavator in pure FP workloads utilizing modern instruction sets, so 40% will just not cut it (match Haswell / Broadwell / Skylake). A realistic expectation would be Ivy Bridge like IPC in FP workloads, which I think will be quite accurate. I still wouldn´t expect Zen to match Ivy Bridge in single threaded performance, since I expect Ivy Bridge to have a significant frequency advantage over Zen. The relative integer performance is somewhat irrelevant in Zen, since it has been more than sufficient even in the 15h series. The inferior FP performance is the ultimate cause for AMDs suffering.
post #516 of 841
The ipc alone tells nothing. We dont know anything about the frequency

Even if it is 50% IPC improve with 3ghz stock clocks means nothing
Workstation
(4 items)
 
  
CPUMotherboardGraphicsMonitor
Xeon E5-2690 Supermicro 2011 Nvidia GP100/ Vega FE Dell ultrasharp 4k 
  hide details  
Reply
Workstation
(4 items)
 
  
CPUMotherboardGraphicsMonitor
Xeon E5-2690 Supermicro 2011 Nvidia GP100/ Vega FE Dell ultrasharp 4k 
  hide details  
Reply
post #517 of 841
IPC tells everything about the relative performance, but nothing about the absolute performance biggrin.gif
post #518 of 841
Quote:
Originally Posted by Cyrious View Post

True true. Still, 40% over Excavator sticks it in the Haswell region of performance, and an unlocked 8c/16t at that performance level for sub $700 would make for a good buy for me.

I do have a question for you though, since you seem to actually know your stuff and have the ability to simulate what a possible Zen design would look like in terms of performance. The question being: Assuming the theoretical 52% core design you said as an example earlier, how much of a hit would Zen take from the loss of the L3 cache(s)? I ask because the Exascale Zen APU is rumored to not have its L3 active, and it begs the question on whether it would be a big issue or not.

It is really hard to say what the impact of L3 removal would be without actual platform benchmarks, but usually it was only 3~5% on average for the construction cores - but there are specific cases (mostly in server/big data) where removing the L3 can cost 15% or more.

If that APU has HBM memory that is used to cache the same tasks, it may be a much lesser loss - or even a gain given the immense increase in capacity. The L3, in that case, may even act as a source of unwanted latency, which may be why removal would make more sense than inclusion.
post #519 of 841
Quote:
Originally Posted by The Stilt View Post

IPC tells everything about the relative performance, but nothing about the absolute performance biggrin.gif

Amen to that. I don't expect Zen to exceed 4GHz easily, though AMD certainly has the building blocks to make CPUs that can do so, they have shown no prowess for creating caches that can do so. However, the K10 cache, on 14nm LPP, would be able to hit 4.5GHz (transistor switching speed improvements are great). Caches and memory are one area where 14nm FinFet should really excel... but I'm talking pushing the voltages to achieve that.

So long as I can get ~4.5GHz Sandy Bridge level of ST performance with 6+ cores w/ SMT (& 50%+ MT improvement), Zen is viable for my development/VM machine. Any less than that, and it is only viable to upgrade my wife's machine (Phenom II X4 955) or my HTPC (A8-7600 w/ 7870XT).

Zen will need to be 24/7 reliably clocked to 4GHz+ with six cores for me if it has Haswell IPC. The better the IPC, the lower that clockrate needs to be.
post #520 of 841
Quote:
Originally Posted by The Stilt View Post

Skylake can be > 80% faster (IPC) over Excavator in pure FP workloads utilizing modern instruction sets, so 40% will just not cut it (match Haswell / Broadwell / Skylake). A realistic expectation would be Ivy Bridge like IPC in FP workloads, which I think will be quite accurate. I still wouldn´t expect Zen to match Ivy Bridge in single threaded performance, since I expect Ivy Bridge to have a significant frequency advantage over Zen. The relative integer performance is somewhat irrelevant in Zen, since it has been more than sufficient even in the 15h series. The inferior FP performance is the ultimate cause for AMDs suffering.

Yes, floating point is something of a wildcard. I think Zen's FPU should be quite a jump, though. Floating point comparisons with the construction cores are very difficult and the control code is usually in the ALU, so there is a constant dependence on the ALU/AGU setup for extracting full performance. The FPU could be anywhere on the spectrum for performance, even in the negative territory, from what we know of it. In theory, though, it should match or exceed Haswell in non AVX code. Instruction sets always make things ugly, which is why I tend to exclude them from my comparisons thumb.gif
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Hardware News
Overclock.net › Forums › Industry News › Hardware News › [OC3D] AMD's Zen will have a "greater than 40%" IPC improvement over Excavator, says Lisa Su