Attentiuon wendell AMD zen 6 bandwidth anomalies

Hardware Hub CPU
1

Wendell this is chew*

I’ve done countless exploration into “anomoilies” possible regression…

I have run many straps…6000/2000 6400/2133 6600/2200 8000/2000 8400/2100 8800/2200.

Even explored 2233 albeit with slight latency penalties…

2 things remains constant when compared to 7700x…I can’t max out the theoretical copy cap 70250 ( 2200 fclk ) and 71250 ( 2233 fclk )

Karhu output in mb/s is down…bestr so far is 258mb/s compared to 273mb/s.

Any thoughts and i can provide a multitude of screenshots buts lets start with the unicorn as that one definitely makes you say hmmmmm.

image
image1920×1080 261 KB

1 Like
2

First, the title should be zen 5, not zen 6.

Sadly, as you found, there is no point to run memory at 2:1 mode for single ccd ryzen. The infinite fabric bandwidth is no more than DDR5 6000MT/s. You have to use dual ccd to get full benefits of 8000MT/s.

3

Sorry about typo’s im on a wooting with rapid fire enabled.

Secondly I am an internal tester so while I appreciate your feedback i’m well aware and have 9950x and 9950x3d sitting here but this is just about anomalies where a single ccd is not matching a previous gen single ccd.

expected behavior at 2200 fclk is 70250 higher write speeds and 70250 copy.

expected behavior is being able to get about rougly 270 mb/s in karhu on a good sample 7700x/7800x3d.

9700x is not exhibiting expected behavior. Not even A unicorn synced 8800/2200.

Like i said i have tested all optimal straps so while i appreciate your concern of 1/2 even 6600/2200 is not better. nor is 6800/2233 or 7000/2233.

4

6600/2233

image
image1920×1080 230 KB

so anyway like i said I appreciate the suggestion but this is not my first rodeo.

7700x

100K write
100K write911×568 158 KB

5

9700x is 65W, while 7700x is 105W.
Make sure you set them equally to do the comparison.

6

I tested on phase change at -35c If you look closer at the bench mate program you will see I’m already wattage uncapped :wink:

The reason we tested at -35c was to free up the imc more to see if we could ever get the write/copy up…and still could not even at 8800/2200 synced. You can tell i’m synced as the latency boost is significant. Sync is working as expected however…copy/writes are still not expected.

1 Like
7

here’s another one.

7800x3d 6200/2233

64gb 2233
64gb 22331693×803 401 KB

8

Do you see any IMC improvements on Ryzen 9000?

9

I can only comment on what is retail currently.

1/1 either I’m lucky or 6600 may be more common. I’m either lucky or 2233 may be more common.

I’d say I’m probably not lucky as through entire 7000 series only 2 chips were found to do 2233 and so far I’ve seen 3 capable of 2233 on 9000 series.

On the other note concerning 1/2 mode 82/84/86/88 multipliers work for 1/2 mode also what I was testing and why i went to -35c

M dies seems like 8400 will be realistic on good imc cpu’s and 8000 is fairly simple on A die but higher than 8000 on A die is problematic.

This of course finding when using a 1 dpc motherboard and a 2dpc may be more challenging.

1 Like
10

How did you get 2233 fclk? The Ryzen 9000 uses the same I/O die and the same Infinity Fabric design as the Ryzen 7000, only CCD changed.

11

same way i got 2233 out of a 7800x3d if you look more closely and note it has stability testing to back it up…

silicon varies. Nothing is impossible only highly improbable.

Chew* ex internal pro AMD overclocker :wink:

1 Like
12

Chew??? :astonished: :astonished: :astonished:
Crazy to see such an OG around again.

13

crazy and og is my style :blush:

image
image1920×1446 189 KB

2 Likes