Hardware & Technical Radeon VII teardown, courtesy of Gamers Nexus

[Morbad]

[video=youtube;0b9_c4oJRTE]https://www.youtube.com/watch?v=0b9_c4oJRTE[/video]

Was pretty meh about the Radeon VII, until I saw how easy it would be to put underwater with just a GPU block and the stock VRM cooling plate (no need for a full-cover block). Could probably attach it to an existing custom loop for 50 dollars or just slap any half-way decent AIO on it and be good to go.

Will need to see final benchmarks to see how it really stacks up against the competition (mostly the RTX 2080 non Ti), but being fairly simple to dismantle and cool is extremely promising with regards to being able to overclock the card.

 

[Caramel Clown]

Hah, I'm not the only one on this forum who goes to GN for some competent content Those guys are on top of the game when it comes to hardware testing.
Honestly, I'm very intrigued by what AMD are doing now. I've been with team green for many years now. Mainly because AMD had very little to offer in high-mid to high performance market. Let's not mention Vega...But now, it might change. I've not seen benchmarks yet, but it looks interesting so far.

Edit. Missed it the first time - they mention there's still embargo on performance benchmarks.

 

[Morbad]

Gamers Nexus is one of the few review sites that does teardowns instead of 'unboxings' for the initial reveal. Techpowerup posted some images as well: https://www.techpowerup.com/reviews/AMD/Radeon_VII_Unboxing/3.html. Unfortunately, I can't read the info on the MOSFETs and Burke sorta mumbled his way through their part numbers, so I'm not 100% sure of the VRM info. That said, it looks solid and I'd be very surprised if it didn't have significant headroom.

There have been some leaked benchmarks that look plausible (trades blows with the 2080 in 3DMark, but falls behind in GameWorks heavy NVIDIA optimized FFXV), but the embargo probably doesn't lift until the day before launch (the 6th). The Radeon VII still looks like a stop-gap so far...a quick consumer conversion of the Instinct MI50 so that AMD has some product, any product, in the high-end segment.

Anyway, I've been expecting the Radeon VII to not look particularly impressive compared to the 2080...slightly slower, higher noise and power consumption, similar price. However, the fact that it can be taken apart (the RTX FE editions are comically difficult to completely disassemble) and cooled relatively easily could give it enough of a bump to be a convincing enthusiast part.

I guess we'll see for certain at the end of the week.

 

[Caramel Clown]

I was surprised to see so much bare copper in there. Often it's either nickel plated, or not copper at all. Spared no expense on the materials. Can't wait to see if heatsink design is up for the job.
Also, graphite pad instead of a paste. Never seen that from a factory. In fact, only recently found out graphite pads are a thing. Only seen normal thermal pads on VRMs and memory on certain cards.

 

[Morbad]

I was surprised to see so much bare copper in there. Often it's either nickel plated, or not copper at all. Spared no expense on the materials.

AMD has been using big vapor chamber reference coolers on the higher-end parts since the Radeon HD 6000 series. They omit the nickel plating to save money as tarnish generally isn't visible on the reference parts.

The most notable thing about the VII cooler is the heatpipe extensions coming off the vapor chamber and the recessed fan placement.

Can't wait to see if heatsink design is up for the job.

I think it will be a pretty solid cooler in absolute terms, but I expect the VII to run quite warm simply because any part with that sort of TDP will be limited by a dual-slot air cooler. At least axial fans will be quieter than the radial blowers they've used on reference cards until this point.

Also, graphite pad instead of a paste. Never seen that from a factory. In fact, only recently found out graphite pads are a thing. Only seen normal thermal pads on VRMs and memory on certain cards.

Yeah, that was a surprise. I've been aware of some high-performance graphite pads for a while, but it does seems strange to use them on the GPU die. As GN mentioned, bondline thickness is generally more important than rated thermal conductivity and these pads weren't terribly thin. This seems to suggest there is an issue with the consistency of the vapor chamber base, or that their may be GPU/HBM packaging differences that could require a gap filler pad to account for differences in contact...in which case graphite was probably the only viable option at such thicknesses.

I'm sure it will be tested with other TIMs so we can see how the graphite pad does in comparison.

 

[Caramel Clown]

I don't think there's a gap to fill between GPU and memory in terms of height. It's filled with resin. So the contact surface between heatsink and CPU/memory chip is totally flat.

Edit. I see what you are saying. The heatsink surface. That might be a reason...perhaps? Which would mean the pad is just a compensation for a shoddy manufacturing standards. I hope that's not it

 

[Asp Explorer]

I'm sure it will be tested with other TIMs so we can see how the graphite pad does in comparison.

Interesting. Is the graphite pad encapsulated or is it matrixed with something to give it structure?

Graphite has quite decent thermal conductivity, but also electrical conductivity, and is prone to grinding teeny bits off itself if you give it a funny look - and every schoolboy on the planet has experimented with pencil leads and 9v batteries

 

[Morbad]

I don't think there's a gap to fill between GPU and memory in terms of height. It's filled with resin. So the contact surface between heatsink and CPU/memory chip is totally flat.

On this particular package variant, yes.

However, Vega had at least three different sources for the complete GPU/HBM/interposer package and not all of them were encapsulated...they still used identical coolers and some samples had thermal issues with the HBM (which was lower on some variants and required extremely thick TIM) as a result.

GN speculated that something similar may be the case with the Radeon VII, which could explain the graphite pad.

Vega:

https://tweakers.net/nieuws/128409/...e-hoogte-en-afwerkkwaliteit-aan-partners.html

So far I've only seen two Radeon VII samples, both encapsulated, but that's hardly a large enough sample size to come to any conclusions.

Interesting. Is the graphite pad encapsulated or is it matrixed with something to give it structure?

Seems to be an acrylic rubber matrix: http://www.hitachi-chem.co.jp/english/report/057/57_tr05.pdf

 

[Morbad]

VRM analysis:

[video=youtube;zx9rKmsxk00]https://www.youtube.com/watch?v=zx9rKmsxk00[/video]

As expected, there shouldn't be any issues with the board's power delivery, even while overclocking.

Actual performance reviews should be available soon.

 

[Robert Maynard]

Actual performance reviews should be available soon.

I'll be interested to see them.

.... not that I'm likely to be replacing my Vega 64 - I'm waiting to see what Navi brings, whenever that may be.

Could have scored a Sapphire Radeon VII for £649....

 

[Jon flint]

[video=youtube_share;6jP3tetYnVI]https://youtu.be/6jP3tetYnVI[/video]

Rough summary, at 1440p its 2% slower than a 1080ti and 4% slower than a 2080.

 

[Stewart_]

Another disappointing AMD card.

Doesn't output perform a RTX 2080, yet there isn't much price difference.

AMD really needs to improve its GPU performance with Nvidia then maybe we'll see price drops.

 

[Morbad]

Some more reviews:

https://www.gamersnexus.net/hwreviews/3437-amd-radeon-vii-review-not-ready-for-launch

https://www.tomshardware.com/reviews/amd-radeon-vii-vega-20-7nm,5977.html

Performance wise it trades blows with the 2080 and 1080 Ti, while using power similar to the former.

Choice of the graphite TIM does seem to be dictated by poor contact pressure and overclocking is essentially non-existent due to driver issues and lack of information at release.

 

[Caramel Clown]

It kind of makes little sense for that price. Would it be just ever so slightly cheaper. Considering they probably won't make that many of them price drop isn't happening. Unless demand is very low.

 

[GJ51]

Meh on the card

BUT

That Mod Mat !!!!!!!!!!!!

Gotta have one

Unfortunately already sold out

 

[Caramel Clown]

Meh on the card

BUT

That Mod Mat !!!!!!!!!!!!

Gotta have one

Unfortunately already sold out

They restock most of their products regularly. Mod mats have been quite a hit

 

[Caramel Clown]

Well. You were right about the thermal pad. It's just there to compensate for wonky design and manufacturing imperfections. That's a shame really.

 

[Morbad]

Well. You were right about the thermal pad. It's just there to compensate for wonky design and manufacturing imperfections. That's a shame really.

Yeah, cooler base seems to be erratic and concave. Probably found it more economical to use a high-end gap filler pad than get whoever they have assembling the cooler ensure a consistently flat base.