AMD Claims Zen 2 Has 29% Higher IPC than Zen 1 in Certain Workloads

Megalith

Megalith

24-bit/48kHz
Staff member
Joined
Aug 20, 2006
Messages
13,000
Notebookcheck is pointing out that AMD had quietly disclosed some information on Zen 2's IPC deep within the footnotes of their Rome announcement: the company is claiming a 29% increase in IPC when comparing Zen 2 to Zen 1 in "combined floating point and integer benchmarks." The author calls this "decently high," as “AMD achieved a 52% IPC uplift when comparing Zen to Piledriver.”

Of course, don't take AMD's word for granted, the IPC increase in reality may be slightly lower in floating point and integer workloads and it could be much lower in other workloads that do not utilize the FPU. FLOPs are important for the data center, so it makes sense AMD would have improved the ability of Zen to push those FLOPs. This may not translate to additional performance for games, web browsing, video encoding, and other common usages for CPUs that exist within the professional and consumer spaces.
 
What Ryzen needs is higher clocks.
Hopefully now that these are fab'd by TSMC it will improve
 
  • Like
Reactions: Meeho
like this
sleepybp said:
What Ryzen needs is higher clocks.
Hopefully now that these are fab'd by TSMC it will improve
Click to expand...

What they need is higher per-core performance. Higher IPC would actually be preferable to higher clocks if you had to choose- higher clocks mean higher power usage and heat.

But that's asking for miracles ;)
 
If these projections are on par from AMD, they have increased IPC, increased clocks and (at least for server) increased core counts. Even if one of those three factors isn't going to live up to AMD's expectations, they still could take the performance crown.

Intel desperately needs their 10 nm production going and Ice Lake out the door now. There are still reasons to go with Intel but they are increasingly niche and come at premium (Optane DIMMs, on package fabric, quad/octo socket support, and some select RAS features). Intel could still pull off a few surprises (eDRAM on desktops/E series Xeons) but it looks like they are already in panic mode and running out of solutions. I will say that the unlocked Xeon 28 core LGA 3647 part looks very nice from a 'win at all costs' perspective.
 
KazeoHin said:
Estimates are between 10-20% clockspeed improvement. so you're looking at 4.5 to 5.0 as the 'typical' boost scenario.
Click to expand...

I heard this as well, but like anything I'll wait to see it before and I would prefer a more mature Mobo anyways, the 1700x was a bitch on the crosshair when it originally released.
 
Just hope X399 don't get obsolete anytime soon, i would like to be able to update one time on this platform.
Maybe a nifty little 24 / 48 core CPU down the line to double up what i just got and maintain the same wattage.

NOW if Gigabyte could just fix their RGB fusion cuz the B18.1019.1 version don't really work as one would like it to.
 
If I remember right from their earlier roadmap we are both good till atleast 2020.
 
This may not translate to additional performance for games, web browsing, video encoding, and other common usages for CPUs that exist within the professional and consumer spaces.​
So why not claim 75%. :facepalm:
#BothSidesDoIt
 
Kor said:
If they can pull a general 10%-ish IPC gain along with a competitive clock speed to Intel it should would be a legendary buy if the pricing remains consistent.
Click to expand...
4.2 * 1.1 = 4.62 Ghz * 1.1 ipc = ~5 Ghz equivalent compared to ryzen+. Still a tad slower than Intel's ipc but with the pricing should blow the doors off any Intel offering by a long shot.
 
29% isn't bad at all for best case. What was Zen+'s best case IPC over Zen?

Considering clock for clock, AMD is around (CMIIW) 5% behind Intel? clock vs clock, AMD and Intel seem very close, except that current Intel offerings are clocking above 10% higher than the 2700X/2950X at the moment, which account for the 15-20% single thread advantage of Intel.
 
Last edited:
Darkswordz said:
29% increase in IPC would take Intel about 15 years to achieve.

:whistle:
Click to expand...
Considering it took them 7+ years to achieve a 25% gain in IPC between the 2600K and 7700K, I would actually be inclined to believe this!
We can thank AMD for kicking them in the ass to start providing more cores and (hopefully) fixing their hardware security-faults.
 
Montu said:
"Certain Workloads" nuff said. Cherry picking isn't a good sign.
Click to expand...

Cherry picking for press releases and such is the rule. However, in most cases you don't hear anything about certain workloads. How many times now have we seen Intel claim 10-20% IPC improvement period. Once people get their hands on the CPUs we find out it's close to 2-5% overall and the 10-20% was only in certain workloads. The difference here is that it's actually being stated the increase of 29% is certain workloads only.

I have no issue with the statement about the IPC increase.
 
DigitalGriffin said:
4.2 * 1.1 = 4.62 Ghz * 1.1 ipc = ~5 Ghz equivalent compared to ryzen+. Still a tad slower than Intel's ipc but with the pricing should blow the doors off any Intel offering by a long shot.
Click to expand...
Ryzen perf may not directly translate from epyc if they use different i/o chips (likely). It may be a bit better than Epyc in certain workloads, and it may be worse in others. Still, it should have better clock speed and ipc than zen+, we'll just have to wait to see how much better.
 
Montu said:
"Certain Workloads" nuff said. Cherry picking isn't a good sign.
Click to expand...

They will be getting 10-15% IPC increase across the board in general. That is what's been widely reported for a while now. They are reporting up to 29% best case so not sure how that is a bad sign.
 
KazeoHin said:
Estimates are between 10-20% clockspeed improvement. so you're looking at 4.5 to 5.0 as the 'typical' boost scenario.
Click to expand...
I'd agree with ~10% IPC as a reasonable estimate overall. Maybe tad more in some common applications and higher IPC in niche applications that take advantage or are tuned for AMD CPUs (which hopefully will be a whole lot more).


DigitalGriffin said:
4.2 * 1.1 = 4.62 Ghz * 1.1 ipc = ~5 Ghz equivalent compared to ryzen+. Still a tad slower than Intel's ipc but with the pricing should blow the doors off any Intel offering by a long shot.
Click to expand...
There is a bench floating around that has Zen 5.0 vs Intel 5.0 and Zen already has higher IPC in it at the same clocks... very interesting.
TSMC says 20% speed for same power... so 1.15-1.2 is probably a closer number.
At 1.15x you're pushing ~5.3GHz equivalent after IPC jump.. if that is the case it's going to be an interesting year.

Also don't forget possible bumps to cache, most of the IO is off the chiplets now.
 
Montu said:
"Certain Workloads" nuff said. Cherry picking isn't a good sign.
Click to expand...
Montu said:
I hope that's not the case. I'm really looking forward to Zen 2, but this kind of thing doesn't smell right.
Click to expand...
They doubled the FPU width from 128 to 256bit. So it's expected it'd be considerable number like "29%". It's not exactly cherry picked though, since it could be 29% for any application that is a FP heavy workload.

Nobu said:
Ryzen perf may not directly translate from epyc if they use different i/o chips (likely). It may be a bit better than Epyc in certain workloads, and it may be worse in others. Still, it should have better clock speed and ipc than zen+, we'll just have to wait to see how much better.
Click to expand...
Being it isn't IO related, it shouldn't get crippled any (I would hope at least), but they still could hold back from adding the full width FP unit on desktop if there's no gain on that front. May save on thermal headroom for boost? I dunno...

EDIT: Just in case I'm misinterpreting something
https://techreport.com/news/34242/amd-teases-zen-2-epyc-cpus-with-up-to-64-cores-and-128-threads
 
Looks like Zen 2 will be a nice upgrade, beats the hell out the Intel's usual 5% or less.
 
Kor said:
Second coming of the Athlon XP right there.
Click to expand...

It's funny you mention that; I fired up a P4 today that I have some coding software on, and was amazed by the sheer heat it gives off. (Stock Clocks)

I remembered I replaced it with an Athlon system, so I dug it out, and swapped the HD I needed into it.

It's current processor is a Athlon 2500M, and at a 1GHz overclock is only drawing ~75W at idle, running WinXP32.

From the Wall.

That's mobo, 1HD, 2GB of DDR memory, and a processor overclocked from 1888MHz to 2800 (200fsb x 14 mul)


I replaced this system with a x58 based Core 2 duo I first built for Crysis, but it melted a long time ago (cooler failure), and was replaced with a i7-920 system.

The only mobo I now have that ever got chip upgrades was the Athlon, and the 920, which got replaced with a $40 Xeon last year.

Running some tests, the speed improvement over these generations scales directly with memory bandwidth; since we've hit a wall about 50GB/s on bandwidth, we'll see more channels having the most impact, IMHO.

Remember all the stuff in Intel runs thru one internal bus; Ryzen has multiples per chiplet.

It will get more interesting next year, I believe.

:)
 
I wonder if this is like their IPC claims in the past where it was just changed boost speeds for certain core loads.

I'll take a 10% IPC + 10% MHz increase, but lets not get over board with the marketing hype here...
 
Trimlock said:
I wonder if this is like their IPC claims in the past where it was just changed boost speeds for certain core loads.

I'll take a 10% IPC + 10% MHz increase, but lets not get over board with the marketing hype here...
Click to expand...

Well, IPC is instructions per cycle, so changing boost speeds wouldn't improve IPC at all. AMD is being specific about IPC improvements, and at least with Ryzen it was shown that they were truthful (40% improvement turned into 52% in fact). And given it was buried in the footnotes of their announcement, it is difficult to argue they are hyping it in any measureable way either.
 
Geni said:
Well, IPC is instructions per cycle, so changing boost speeds wouldn't improve IPC at all. AMD is being specific about IPC improvements, and at least with Ryzen it was shown that they were truthful (40% improvement turned into 52% in fact). And given it was buried in the footnotes of their announcement, it is difficult to argue they are hyping it in any measureable way either.
Click to expand...
They've done this before, many times, but to be fair it was during their dog days.
 
I just hope Zen 2 for the desktop isn't something stupid like 12nm instead of 7nm.

I have doubts AMD will be making Ryzen 3000 2 dies for mainstream parts.
 
You must log in or register to reply here.
Back
Top