• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Question Zen 6 Speculation Thread

Page 411 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
adroc_thurston said:
Also "maxing out fabric and memory speeds" is not how those cores are designed to run.
Click to expand...
Well yeah, but we were comparing over at overclock.net 😉

Here is a other higher clocked arrow lake with full specs:

p60
e52
r44
d2d 38
ngu36
ddr5 9200cl42 g2

1776545895552.png

1776545945890.png

I ran the following on my X3D

5.8 static
2200mhz FCLK
8800MT/s GDM disable "synced with FCLK"
1776546074126.png
 
Last edited:
Det0x said:
Now we get to the more interesting part in my opinion, IPC comparison.
Click to expand...
I am not familiar with the benchmark you have run, but looking at the screen output it looks like it's measuring the instruction throughput per specific operation (compare the scores against the instruction tables, they mostly align) what could be technically called instructions per clock but it deviates from what other people measure as such.

Of course I might be nitpicking, not knowing the code I am just speculating based on the output screen😉. But if I am right, then Skymont should average higher IPC there than both Zen5 (x3D shouldn't matter for these tests) and Lion Cove on specific tests, simply because it has 4 fma units.
 
CouncilorIrissa said:
Too high.
The official figure for ARL vs RPL is 9% in workloads™.
View attachment 141988
Click to expand...
Yeah maybe, but that comparison is vs stock legacy Arrow Lake which ran pretty slow internal fabric no ?
ARL gain lot from clocking the uncore, especially when your running something like r44, d2d 38 and ngu36 together with above 9000MT/s memory speeds in gear2

285k stock numbers are r38, d3d 21 and ngu32

Just thinking out loud for a reason 🤷‍♀️
 
Last edited:
Det0x said:
Yeah maybe, but that comparison is vs stock legacy Arrow Lake which ran pretty slow internal fabric no ?
ARL gain lot from clocking the uncore, especially when your running something like r44, d2d 38 and ngu36

Just thinking out loud for a reason 🤷‍♀️
Click to expand...
Yep, 270K should score 2-3% higher iso-clock I think.
 
adroc_thurston said:
We'z talking 1t stuff.
Click to expand...
CPU temp is controled by skipping a given number of cycles or/and by very short frequency dips,
so if you have less of thoses at high ST core power you can grab 2% even if the intrinsical IPC is the same.

A relevant comparison should be done at fixed and reasonable frequency such that the core is at a distance
from its max temp.
 
Abwx said:
CPU temp is controled by skipping a given number of cycles or/and by very short frequency dips,
so if you have less of thoses at high ST core power you can grab 2% even if the intrinsical IPC is the same.

A relevant comparison should be done at fixed and reasonable frequency such that the core is at a distance
from its max temp.
Click to expand...
I don't think any of those parts jump to TJmax in 1t under good enough cooling.
This ain't a laptop.
 
Det0x said:
Not sure anyone are interested, but we did a little IPC comparison over over at the overclock.net forum with a tuned Arrow Lake refresh system and tuned Zen5X3D system.

Turns out the IPC between these two architectures was alot closer than what i expecting when pushed with maxed out fabric and memoryspeeds.

I also think this matter when we are trying to predict the future battle between Nova Lake and Zen6, as the going forum mentality is that Arrow Lake already have ~10% higher IPC than Zen5 (?)

We used MicroBenchX_v1.0.5 to compare our systems, github page can be found here:

First we compared the core to core latency between our systems, but not really anything unexpecting there.
View attachment 141978

View attachment 141979

Now we get to the more interesting part in my opinion, IPC comparison.

tuned 270k @ 5600mhz
View attachment 141980

tuned Zen5X3D @ 5800mhz
View attachment 141981

When compared side to side, it turns out Arrow Lake average IPC only was ~1.2% higher than Zen5XD
View attachment 141982

If Nova lake only gets lets say +15% IPC and no clock speed increase, and Zen6 gets +10% IPC together with a +20% speed bump Intel will be in really hot water for anything not embarrassing multithreaded. (atleast when comparing overclocked systems)

But then again, have no idea how trust worthy this MicroBenchX_v1.0.5 benchmark from seemingly capframex really are.
Click to expand...

This is great work. No matter what you do people will pick it apart. You isolated various routines and calculated IPC and people will complain it's not actual applications. If you use actual applications they will claim you should have isolated specific routines, etc... You can't win with the armchair quarterbacks.

This is an EXCELLENT analysis and lines up well with my post regarding IPC where I find ARL and Zen 5 are pretty much even. They are stastically on average a draw.
 
Last edited:
dangerman1337 said:
Wonder if RZL will get a significant improvement
Click to expand...
10% on top of NVL.
dangerman1337 said:
if Intel's teams know what problems with NVL can be relatively easier to rectify?
Click to expand...
NVL doesn't have 'problems'.
Their CPU IP sucks! That's literally all there is to it.

NVL has shiny 18AP cIOD, shiny updated d2d, it's on N2p and whatnot; all to deliver ~10% 1t.
 
poke01 said:
Should be the fastest CPU core by the looks of it.
Click to expand...
AMD and Apple are gonna be playing leapfrog for a good while.
adroc_thurston said:
No, M6 should probably outdo it (at half the power).
But still, solid effort.
Click to expand...
Z6 does come out before M6 though, so a short lived win.
Leapfrogging M5 is gonna shock a lot of people though, will be quite fun.
Z7 is where things could get more interesting iso-pp.
 
Last edited:
You must log in or register to reply here.

TRENDING THREADS

Back
Top