Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)

Page 388 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Sort by date Sort by votes
tamz_msc

tamz_msc

Diamond Member
Jan 5, 2017
3,865
3,729
136
As usual, I return to see garbage analyses being posted and regurgitated for dozens of pages.

These are the two results on Geekbench 5 so far:

[1]

[2]

Now, anybody interested in what "Xen HVM domU" is could do a Google search to find out that it is the Xen hypervisor.

Here's the thing - Virtualized environments are WAY MORE unreliable when it comes to Geekbench detecting the CPU specs using the usual method of querying MSRs.
So that automatically discards the first result as it is running a hypervisor.

So that leaves result [2].

Doing the usual appending ".gb5" to the URL shows that it is running 2000 MHz frequency.

And here is my 11370H @2000 MHz fixed frequency.

N.B. It is very easy to fix a frequency in Linux - just type
Code: 
cpupower frequency-set -u xxxxmhz
and
Code: 
cpupower frequency-set -d xxxxmhz
as root in the terminal.

So here I am focusing on the result for my 11370H

1713970854844.png

And here is the Strix ES

1713970907659.png

So compared to Tiger Lake at 2.0 GHz - the frequency being the same as the leak - Zen 5 is ~31% faster clock-for-clock.

Intel has this covered with Lion Cove.
 
  • Haha
  • Like
Reactions: Tlh97, Det0x, Glo. and 4 others
Upvote 1 Downvote
H

Hitman928

Diamond Member
Apr 15, 2012
6,375
11,352
136
tamz_msc said:
As usual, I return to see garbage analyses being posted and regurgitated for dozens of pages.

These are the two results on Geekbench 5 so far:

[1]

[2]

Now, anybody interested in what "Xen HVM domU" is could do a Google search to find out that it is the Xen hypervisor.

Here's the thing - Virtualized environments are WAY MORE unreliable when it comes to Geekbench detecting the CPU specs using the usual method of querying MSRs.
So that automatically discards the first result as it is running a hypervisor.

So that leaves result [2].

Doing the usual appending ".gb5" to the URL shows that it is running 2000 MHz frequency.

And here is my 11370H @2000 MHz fixed frequency.

N.B. It is very easy to fix a frequency in Linux - just type
Code: 
cpupower frequency-set -u xxxxmhz
and
Code: 
cpupower frequency-set -d xxxxmhz
as root in the terminal.

So here I am focusing on the result for my 11370H

View attachment 97732

And here is the Strix ES

View attachment 97733

So compared to Tiger Lake at 2.0 GHz - the frequency being the same as the leak - Zen 5 is ~31% faster clock-for-clock.

Intel has this covered with Lion Cove.
Click to expand...

Any chance you could run it on the same sub-version as the Zen 5 run?
 
Upvote 0 Downvote
H

Hitman928

Diamond Member
Apr 15, 2012
6,375
11,352
136
Mahboi said:
Zen 5 already confirmed worse than Zen 4.
Click to expand...

Yeah, the 1.41 GHz result seems too good to be true (probably not reading the frequency correctly) because that would be ~2X increase. However, the 2 GHz result also seems too low to be true (probably bad result with immature platform/BIOS), otherwise it's like a <5% IPC increase gen on gen.
 
  • Like
Reactions: Tlh97, lightmanek and spursindonesia
Upvote 0 Downvote
Mahboi

Mahboi

Golden Member
Apr 4, 2024
1,035
1,899
96
Hitman928 said:
Yeah, the 1.41 GHz result seems too good to be true (probably not reading the frequency correctly) because that would be ~2X increase. However, the 2 GHz result also seems too low to be true (probably bad result with immature platform/BIOS), otherwise it's like a <5% IPC increase gen on gen.
Click to expand...
I'm not good enough technically to throw my dart in, but comparing it at 2Ghz to cores that are years old on other nodes seem a bit silly, when the expected clocks should be around 6Ghz, which none of the older CPUs can reach. Scaling isn't fully linear across frequency IIUC.
I'm also finding very funny that an 8 wide front end would somehow have near or worse IPC than Zen 4 after Z2->Z4 has been all on 6 wide.
 
Upvote 0 Downvote
G

Geddagod

Golden Member
Dec 28, 2021
1,296
1,368
106
Hitman928 said:
Yeah, the 1.41 GHz result seems too good to be true (probably not reading the frequency correctly) because that would be ~2X increase. However, the 2 GHz result also seems too low to be true (probably bad result with immature platform/BIOS), otherwise it's like a <5% IPC increase gen on gen.
Click to expand...
How? Thought this would still be a ~15-20% increase over Zen 4 still?
 
Upvote 0 Downvote
G

Geddagod

Golden Member
Dec 28, 2021
1,296
1,368
106
Mahboi said:
I'm not good enough technically to throw my dart in, but comparing it at 2Ghz to cores that are years old on other nodes seem a bit silly, when the expected clocks should be around 6Ghz, which none of the older CPUs can reach. Scaling isn't fully linear across frequency IIUC.
I'm also finding very funny that an 8 wide front end would somehow have near or worse IPC than Zen 4 after Z2->Z4 has been all on 6 wide.
Click to expand...
I don't think the score is worse on a zen 4 cpu at 2ghz
I also think GB SC/IPC scores pretty linearly tbh...
The node doesn't matter, and GLC/RPC/Zen 4 are like 2 years older than Zen 5, not decades. That shouldn't matter either.
 
Upvote 0 Downvote
T

Timmah!

Golden Member
Jul 24, 2010
1,524
856
136
  • Like
Reactions: Tlh97 and Joe NYC
Upvote 0 Downvote
H

Hitman928

Diamond Member
Apr 15, 2012
6,375
11,352
136
Geddagod said:
How? Thought this would still be a ~15-20% increase over Zen 4 still?
Click to expand...

I’m going off of @tamz_msc ’s results and rough IPC math from Anandtech reviews.

Zen 3 over Icelake was about 15% integer IPC and ICL ~ TGL while Zen 4 is roughly 10% over Zen 3. That leaves Zen 4 being roughly 26% higher IPC than TGL while @tamz_msc showed 31% higher for Zen 5 which would mean Zen 5 would be less than a 5% IPC increase over Zen 4. It’s really rough math but close enough to show that something is off in the score.
 
  • Like
Reactions: Tlh97, spursindonesia and Mahboi
Upvote 0 Downvote
tamz_msc

tamz_msc

Diamond Member
Jan 5, 2017
3,865
3,729
136
Hitman928 said:
I’m going off of @tamz_msc ’s results and rough IPC math from Anandtech reviews.

Zen 3 over Icelake was about 15% integer IPC and ICL ~ TGL while Zen 4 is roughly 10% over Zen 3. That leaves Zen 4 being roughly 26% higher IPC than TGL while @tamz_msc showed 31% higher for Zen 5 which would mean Zen 5 would be less than a 5% IPC increase over Zen 4. It’s really rough math but close enough to show that something is off in the score.
Click to expand...
...and 12 percent higher IPC than Golden Cove at 2000 MHz, according to @gdansk posted score.
 
Upvote 0 Downvote
H

Hitman928

Diamond Member
Apr 15, 2012
6,375
11,352
136
tamz_msc said:
...and 12 percent higher IPC than Golden Cove at 2000 MHz, according to @gdansk posted score.
Click to expand...

Zen 4 has slightly higher IPC than Golden Cove so the numbers track with the above calculation, roughly speaking at least. There’s basically no way Zen 5 gets that low of an increase unless AMD completely botched something. The far more plausible answer is that it’s just that the score is an ES on an immature platform that is limiting performance.
 
  • Like
Reactions: Tlh97 and Mahboi
Upvote 0 Downvote
adroc_thurston

adroc_thurston

Diamond Member
Jul 2, 2023
3,722
5,437
96
Hitman928 said:
Zen 4 has slightly higher IPC than Golden Cove so the numbers track with the above calculation, roughly speaking at least. There’s basically no way Zen 5 gets that low of an increase unless AMD completely botched something. The far more plausible answer is that it’s just that the score is an ES on an immature platform that is limiting performance.
Click to expand...
it's not 2GHz.
what are you on.
clk readouts are in .gb5. like, wat?
 
Upvote 0 Downvote
You must log in or register to reply here.

TRENDING THREADS

COMPANY
RESOURCES
FOLLOW
Top Bottom