[News] Overclocker der8auer delids AMD Ryzen CPU to reveal solder TIM

[StyM]

http://hexus.net/tech/news/cpu/10305...al-solder-tim/

One of the better known professional overclockers on the world circuit has published a video in which he delids a Ryzen 7 1700 CPU. Der8auer starts his video by warning that you shouldn't try this at home, which is sage advice for two reasons revealed later in the video: 1. It?s a nerve-wracking processes which could easily break your shiny new CPU, and 2. AMD's Thermal Interface Material (TIM) is an indium-based solder which is a great conductor so delidding is rather unnecessary.

[Ket]

How is this news? Everybody knows AMD still use really good TIM under the IHS.

[chew*]

How is this news? Everybody knows AMD still use really good TIM under the IHS.

It actually is news worthy and explains a lot as you will soon find out. very narrow and small die. I see concentration issues over a very small surface area.

[AliG]

It actually is news worthy and explains a lot as you will soon find out. very narrow and small die. I see concentration issues over a very small surface area.

Are you referencing the problems the GTX 1080 faces with super high concentration of heat?

[chew*]

Well first off usually the center of the die is the hotspot. Most heatsinks and waterblocks are tuned for that. It looks like there are actually 2 dies on the pcb.

It is a rather thin strip compared to say thuban which was a massive core making far more contact with the IHS.

This could explain why i'm seeing rather significant core temps on the higher leakage chips and the heat output from the heatsink does not jive with the actual temps.

I must have done at least 5 remounts before I came to terms with the fact that it was not a bad mount.

[Ket]

It actually is news worthy and explains a lot as you will soon find out. very narrow and small die. I see concentration issues over a very small surface area.

I would of thought people would expect the Zen die to be rather small anyway. I automatically thought it would be, but maybe I just have more foresight with that kind of thing.

[Sparky]

So a CPU water block that focuses its cooling power in the center just won't do as good of a job.

Hm.

I have an older XSPC Raystorm. Wonder how well it would do. I think with its design it would do fine, as long as I orient the block to match the die. Still, most of the cooling power is more centered on the block, so...

I'm sure it will be fine.

[NEOAethyr]

I think most blocks will be fine, they usually use a cheap micro fin strip design, that should cover it.
Unless you make you own blocks you should be using ek or aquacomputer anyways...
Ehh there's a few others but I don't care lol.

I think there's all sorts of hot spots to potentially worry for, depending on model..., intel/amd
For ex.
One of the caches could have issues at a limit.
Or one of the cores, for several reasons, one of them may be a certain way (1 core is always worse then others), or maybe the core was at the edge of the wafer in the beginning and get messed up during manufacturing.
There's also the fpu up in steps, simply sse and avx, fpu's have generally took more volts then there integer counterparts.
I just noticed yesterday the current asus intel mainstream lv board has a avx divider/multiplier.
Also pcie... remember dudes ati/amd cards don't clock well on avg for pcie, there board have in the past though, just using a amd vga card may limit your bclk (maybe...).

There's gonna be all sorts of things, even going to 1 stick of single sidded ram may help in the end.
That may be cache related.
That sorta thing...

It's always that way in some form or another, but with 2 dies, if that is what that is (sure looks like it)....
You can double that chance of 1 piece of the core being worse then the other pieces.
(edit: It looks like they are building these in modules, which means the server chips will have 4 instead of 2 chips, getting quad chan, in that sense maybe it could be interesting but I have little hope for a good server board)

I would expect the x6's to show up after a few months to collect bad batches.
I wonder if they're gonna let us unlock those?
4 cores looks like it would be the best bet for checking out clocks, maybe even for mem clocks too (edit: I'm starting to wonder however if the quad cores will be 64bit single chan mem :\, or if they just use bad x6 cores lol).
Depending on how bad the 8 cores actually were...


And what I actually wanted to post about...
Gaming wise...
Anyone try turning off core parking?
And not using windows 10 lmao...
Other then that I was think it could be a missing 3rd timing or something like that, something that's a little slack.

I would like to see compares with these games, the rest I don't really care about.
gta5
the forest
dying light

[Sparky]

Does seem odd to me that the CPU is a dual "glued" die (practically MCM isn't it?). I wasn't expecting that. Makes it easier to do 4c units though I guess?

[AliG]

Does seem odd to me that the CPU is a dual "glued" die (practically MCM isn't it?). I wasn't expecting that. Makes it easier to do 4c units though I guess?

My guess is that was a cost cutting measure for manufacturing yield. As long as the scheduler appropriately distributes work across all cores and the L3 cache enables quick communication between the 4c units, I don't really think there's too much to gain by going native.

In either case, they have bigger issues with the memory bandwidth and AVX performance.

[NEOAethyr]

My guess is that was a cost cutting measure for manufacturing yield.

That's what I was thinking.
Those little square chips will come out of the oven more consistent.

Though like I said, I think adding 2 together in effect takes the worst of both and adds them up...