AMD Ryzen Threadripper Installation and Unboxing

Subject: Processors | August 3, 2017 - 09:00 AM |
Tagged: Threadripper, ryzen, amd

Though we are a full week away from the release of the Ryzen Threadripper reviews, AMD is letting us share the installation process of the new TR4 socket, as well as an unboxing of the awesome kit that AMD put together for media and reviewers. Enjoy!


August 3, 2017 | 10:05 AM - Posted by Martin Trautvetter

AMD are sure getting maximum mileage out of being stuck with a socket and CPU designed primarily for server installation!

My favourite part of the whole thing is the safety-orange retention bracket, but then, I'm a sucker for that colour! :P

August 3, 2017 | 10:19 AM - Posted by Benjamins (not verified)

I saw a picture that shows that the Epyc version is blue.

August 3, 2017 | 12:36 PM - Posted by NotTrautSomthingsFUDagain (not verified)

Threadripper is using the same parts as the server SKUs, but what's so bad about that. The Threadripper platform CPU SKUs and MBs offers 64 PCIe lanes for their consumer TR X399 motherboard SKUs, a bit more than Intel is offering! AND AMD's Epyc SKUs offer 128 PCIe lanes and 8 memory channels, so there are implications for maybe in the future Threadripper/MB variants from AMD being able to offer more than 64 lanes and more than 4 memory channels on any Threadripper future SKUs, that's not a bad option for AMD's consumer customers.

That Threadripper can is huge to conduct plenty of heat away from those to active Threadripper dies and the inactive dies/spacers are just there for mechanical stress relief reasons, but they can also act as heat sinks for any thermal dissipation from the MCM. One can see that all the 4 dies/spacers, 2 active dies and the 2 inactive filler dies/spacers, are not using any Intel Brand toothpaste but are using indium solder. So we have in that large Threadripper/Epyc sized can plenty of extra thermal conductive mass to draw heat away from Threadripper's 2 active Zeppelin dies.

Depending on the 2 active Threadripper dies placement, catty-cornered would be best to isolate the 2 active dies more physically/thermally from each other, each Threadripper/Zeppelin die will represent an actual physically separated thermal domain under that big metal can! So each of the active Zeppelin dies can be overclocked more in the same overclocking regimens of a single core Ryzen 7 series SKU. As far as thermal conductance is concerned each of the physically separate Zeppelin dies are most definitely for thermal reasons dies unto themselves that do not have to worry about any excess thermal silicon conduction sideways across any large monolithic die such as Intel is using with much hotter(The BAD thermal kind of hotter) results.

Those Modular Zeppelin Dies exist in their own isolated thermal domains under that giant Threadripper sized can, and are tied together logically buy AMD’s infinity fabric such that they act logically as a single logical unit of many CPU cores.

The Futrure is Modular Baby, for both CPUs and GPUs from AMD. Just look at those Zeppelin Die/Wafer yields coming in at 80%+ on those Zeppelin die lines, that’s the way of the Future at a very affordable price!

August 3, 2017 | 04:41 PM - Posted by Martin Trautvetter

Aren't we all glad you got to vomit that can of incoherent thermal conductivity nonsense as a reply to a completely unrelated post rather than starting a new thread of your own. We are, right?

August 3, 2017 | 06:17 PM - Posted by MartinTrautpooperNot (not verified)

Look at all the little instances of your FUD-O-Babble GEMs that add up to a mountain of high quality guano when one looks at them as a whole!

Your overpriced overlord's bean counters are currently tripping over each others gravy stained ties scurrying about on all fours trying to find new and unique methods of crazy product segementation in order for Chipzilla to not bleed billions from having to lower those madly inflated margin markups on their overpriced kit!

August 4, 2017 | 01:21 AM - Posted by James

They generally call them a "lid" rather than "heat spreader" because they are there to physically protect the die, not spread heat. They actually reduce cooling efficiency, which is why some people remove them. Very little heat can be transferred sideways through the lid. Heat transfer is proportional to cross sectional area, and the lid is very thin metal. Many existing heat sinks will be far from optimal since they are made to contact a die right in middle of the package. There is no die in middle of the package on Threadripper or Epyc. They are not going to work well if they aren't in contact with the full die area, which is off center for Threadripper. We might have to wait a little while for more optimal heat sinks. The large lid doesn't do anything for cooling. The larger die area and having the chips farther apart could, if you can get a heat sink to actually be in contact with the entire die area.

August 4, 2017 | 12:27 PM - Posted by Clmentoz (not verified)

There are 4 dies, 2 Zeppelin dies and 2 filler/spacer dies in a relatively closer proximity around the center of the IHS/lid than its outer edges and the metal on the lid is also going to be a little thicker owing to the Threadripper MCM's larger size/die placment. There will also be the indium solder between the dies and the IHS, flowed/heated on to thermally connect/interface those 4 dies to the IHS.

So whatever the cooler's coldplate measurments are they will at least have to be big enough to cover the entire area over the 4 dies(active and filler) including the "Cross" void area created between the 4 dies by their separated placment. It's probably better considering the indium solder's flowed/heated on nature to also extend any cooler's coldplate/thicker block protion a little farther out for a little larger area coverage.

But I'd pay more to have that coldplate extended out to the edge of the IHS/lid and have no heat be conducted into the side walls of the lid or trapped heat inside any internal void spaces in and around any attatched to the MCM/processor die's surrounding CAPs/etc. Any metal will conduct heat, including traces and whatever capacitors are under/around the die/s. And is not a total matter of conducted heat there will be radiated heat soaking into every nook and cranny in close proximity to the active dies/circuits/traces and into the lid.

Looking at the Der8auer delidding images there appears to be a line of epoxy down the longer center axis under that lid. And looking at the underside of the lid there is a large metal ridge in that metal lid that is quite thick for structural support along that centerline axis so that will soak up and conduct heat. That's no thin lid by any standard with that big centerline ridge of extra matal will soaking up any heat. So the larger the contact area of the coldplate to any Threadripper IHS/Lid the better. The side walls on the Threadripper IHS/Lid are fat also as is that center matal ridge structure that runs about 80% of the entire large axis length of the lid.

August 3, 2017 | 10:10 AM - Posted by PJ (not verified)

I wish heatsink manufacturers would provide torque specs for tightening their retention systems. I'm never quite sure if I've over-tightened or under-tightened them.

August 3, 2017 | 10:15 AM - Posted by John H (not verified)

This has a very premium feel to it - nice job AMD!

Does the torque wrench actually click or break or something when you hit the torque limit? It looked very basic so I was curious how it worked :)

August 3, 2017 | 10:56 AM - Posted by Benjamins (not verified)

You see it break in the video

August 3, 2017 | 10:22 AM - Posted by Sunain (not verified)

Very impressed with this Threadripper platform. Hopefully the benchmarks show it to be good as well!

August 3, 2017 | 12:23 PM - Posted by James

I thought that they would have two different sockets, one for 4 channel memory and the one for 8 channel memory. It seems like a waste to use the 8 channel socket for both.

August 3, 2017 | 12:49 PM - Posted by icebug

Is DIMM.2 for Optane memory? Haven't seen anything about this socket before.

August 3, 2017 | 01:03 PM - Posted by NotTrautSomthingsFUDagain (not verified)

That an DIMM.2 slot is for The DIMM.2 card riser designed to decongest the crowded motherboard that has no room for M.2 slots.

"Non Volatile Memory Invades the Memory Bus: Performance and Versatility is the Result

A discussion on why and how Non Volatile Memory is moving
onto the DDR bus and what it means for the industry, as well
as the performance gains and versatility it brings with it.

Adam Roberts
Chief Solutions Architect
SanDisk Enterprise Solutions Division"

https://www.flashmemorysummit.com/English/Collaterals/Proceedings/2014/2...

August 3, 2017 | 03:26 PM - Posted by icebug

This explains it very well, thank you. Seems pretty cool, can't wait to see this used by more vendors (if that happens.)

August 3, 2017 | 01:47 PM - Posted by SomeGoodStuf (not verified)

Everybody go and read this article(1) and watch the video over at GamersNexus, Steve Burke is actually taking apart coolers and looking at the actual thermal compound spread. With the disassembled cooler parts(Cold plates) he is finding that because of the placement of the Zeppelin dies that the current cold plate designs are NOT optimal for Threadripper's die placement under the IHS, and the cooler makers will have to come up with better solutions, even if their current designs are adequate. This article us very interesting as is the video!

(1)

"Threadripper Cooler, Thermal Paste Coverage vs. Die Area & IHS"

GamersNexus link triggers Spam filter! so just go to GN website for 8/3/2017.

Damn PCPER that spam filter is runing any proper forum discourse by flagging legit refrences! And both Your site's and GamersNexus' site have serious long running script problems. Do not let your ad partners' ads ruin your websites! It's gitting damn near imprssible to view content for the damn long running scripts.

August 3, 2017 | 01:52 PM - Posted by SomeGoodStuf (not verified)

Edit: imprssible
to: impossible

August 3, 2017 | 06:25 PM - Posted by James

This was pretty obvious. Most coolers designs assume the die is small and in the middle of the socket. That will not work optimally for TR since it has two separate die that are not in the middle. They should still work well if the cooler can cover the die area and can be mounted. It would be bad to have part of the base plate/heat pipes not in contact with the cooler. The lid cannot transfer much heat sideways. Also, for coolers with pre-applied thermal compound, the compound may not be applied to the full area. Some research should be done before buying a cooler. Unfortunately, most designs probably did take such a large socket into account. I am going to wait a bit before buying anything, but that is usually good advice anyway.

August 3, 2017 | 02:19 PM - Posted by Mr.Gold (not verified)

Video start with a zinger ?
The cooler is not indicative of the power the CPU uses...
The CPU is 180w rated. (same for the power supply, the 1950x doesn't require 1200w, but this will let you run 2 overclocked Vega FE + and overclock 1950x)

So the cooler and PSU is to do what review should do.
Eplore the UNLOCK features of a product.
So if you want to push the 16 core past 4ghz, with this cooler you can.

Cant beleive I had to stop 20 second in the unboxing to write this :(

Ryan... do you own stock ?

August 3, 2017 | 09:18 PM - Posted by Paul A. Mitchell (not verified)

Ryan, Many thanks. For our information,
what sorts of mistakes can occur
during installation? I noticed that
you had to slide that orange "carrier frame"
into its channel twice during your demo video.
Can you predict any other possible mistakes
that we should know about in advance?

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Lines and paragraphs break automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd> <blockquote><p><br>
  • Web page addresses and e-mail addresses turn into links automatically.

More information about formatting options

By submitting this form, you accept the Mollom privacy policy.