AMD - Zen chitchat

[watercooled]

Yeah, I know about the AGESA update - it was the first thing I did. Let me clarify - if I set the RAM to run at anything above the BIOS default, it reverts back to that default speed at the next reboot.

The ASUS board I had would do that with a set of Corsair memory - it was like the speed setting just wasn't saving. I swapped it with the Crucial/Micron memory from my other system and it worked fine at that speed.

[Spreadie]

The only one which might have had the AGESA update was released yesterday - the previous one is too old.

That's interesting - I was told the 7A37vA4 BIOS was the AGESA update.

I'll have a look at the BETA.

[CAT-THE-FIFTH]

That's interesting - I was told the 7A37vA4 BIOS was the AGESA update.

I'll have a look at the BETA.

I was looking at the file creation dates - one is early May and the other June and the update only started filtering in late May.

I think we need to clarify what RAM you have and if you try that application I suggested it will tell you the chips used.

[CAT-THE-FIFTH]

This is hilarious:

https://www.youtube.com/watch?v=j3ssad0qZuU

[Phage]

Hmmm looks like I need to hang on to my cash for a couple of months. But it will be a Ryzen bundle.

[CAT-THE-FIFTH]

Another great comparison by Hans De Vries.

[watercooled]

It's also a more complex comparison as the higher core count versions of Zen use multiple smaller dies, in theory leading to higher yields (though there's still packaging yield to consider). But still - that 698mm2 Skylake die!!!

[kompukare]

Seems the overhead (actual die size vs per core size) is all over the place for those Skylake-X core sizes:

Obviously since all the current Ryzen are based on the same die, they all have the same specs.

[watercooled]

That's not really unexpected though - some sections of the processor will be about the same regardless of the die e.g. the fairly big memory controllers, IO, etc - things that are the same size become less significant the more cores you add. The core area itself looks to be fairly consistent across the dies.

[kompukare]

That's not really unexpected though - some sections of the processor will be about the same regardless of the die e.g. the fairly big memory controllers, IO, etc - things that are the same size become less significant the more cores you add. The core area itself looks to be fairly consistent across the dies.

Yes, that makes sense. Anyway, looking at the shots the Skylake-X dies don't look like the actual counts. Looks like all have redundant cores for yields (Turbo 3.0 might play a part here as they might have to bin for their magic 'Top Two' cores as well now).
10 cores looks like 12 (3 x 4)
18 cores looks like 20 (5 x 4)
28 cores looks like 30 (5 x 6)
Which makes each core (excluding overheads) about 20-21mm², and the rest of the chip between 71mm² and 79mm²:

[watercooled]

Take another look at the dies, in particular, look carefully at the two outermost 'cores' on the second row from the top. They're not actually cores. I'm not sure what they actually are yet, maybe some sort of IO with a connection to the mesh.

[DanceswithUnix]

Take another look at the dies, in particular, look carefully at the two outermost 'cores' on the second row from the top. They're not actually cores. I'm not sure what they actually are yet, maybe some sort of IO with a connection to the mesh.

I would guess from the way there is one each side that there is one per pair of memory channels, so probably memory control, buffering, prefetch etc. If you look at a normal quad core skylake there is a block for memory control and display control which is about the size of two cores, so if these don't have display outputs...

[scaryjim]

I would guess from the way there is one each side that there is one per pair of memory channels, so probably memory control, buffering, prefetch etc. ...

Perhaps some of it is extra cache too? not sure how cache sizes/layout work...?

[CAT-THE-FIFTH]

Interesting comments from TTL:

https://forum.overclock3d.net/showthread.php?t=80970

[scaryjim]

Interesting comments from TTL ...

You mean

... other sites - mentioning no names, let's call them Gexus - broke NDA anyway ...

?

Fight, fight, fight!

[CAT-THE-FIFTH]

You mean



?

Fight, fight, fight!

Nope,the comments about Coffeelake.

Edit!!

There are rumours about another Skylake update coming later in the year,so it makes me wonder if Coffeelake has been delayed until next year,or even cancelled.