Vega 7nm early benches

Following the anandtech coverage and seems like there will be a vega 7nm for gaming after the ai/pro variant... The slides after 11.08 seems to hint at it.

https://www.anandtech.com/show/12909/amd-computex-2018-press-event-a-live-blog-10am-taiwan-2am-utc

IMG_7715_575px.JPG


And yea the 32 core TR is going to be a godless monster and yet run on current mobos. We kinda knew it was doable being that Epyc and Threadripper used the same socket. Quad channel ram only I suppose. Price? 2k at least imo... maybe $1500 on sale.
 
But is not for us gamers.

Yep, however Lisa Su confirmed the 7nm GPU will be coming to gaming which means it'll be Navi for sure. The 7nm Vega is only meant for HPC and datacenters since it is unlocked FP64 and others. It is just a beast compute card anyway.
 
It's still a big chip even at 7nm, so it's obvious that they added more hardware than Vega at 14nm, while reducing power consumption in the process....An easy hint is that the picture shown above has 4 stacks of HBM, while the 14nm Vega only has 2, so regardless of the hardware inside the GPU die, this thing has a huge increase in memory bandwidth if each stack has a 1024 bit interface to the GPU die.....It would allow the chip to have over 1 TB of memory bandwidth...



As for me, it'll get the 32 core thread ripper ASAP since it seems it operates very close to the current 16 core version clock speed wise, while offering twice the processing power and still being compatible with existing X399 boards, so it's a no brainer for me.
 
Fury was 600mm^ ( 25mm x 24 mm ), while Vega is 484 mm^ ( 22mm x 22 mm ).....Disregarding fab process changes of course ( 28nm for the first and 14nm for Vega ).


Third picture is also interesting because the GPU is more rectangular shaped rather than nearly a perfect square in the 2 top examples, but overall it doesn't seem that small just the same....So it's a given there's more hardware and a lot more bandwidth supplying the GPU too.


Edit: the wider part of the 7nm chip is almost as wide as 2 HBM stacks, so assuming those stacks are the same as the current 14nm vega in terms of physical size, the GPU could be at least 20mm in one direction, but it's hard to say on the other....
 
Quick and dirty measuring says that the narrower part is 12.5 millimeters, so when multiplied by the widest part ( 20 millimeters ), we get a 250mm^ die.
 


Supposedly, Nvidia will release a refresh part in august, as it's features will be discussed in a developers conference that I can't even remember the name of it, but it could be anything from a Pascal made at 12nm and clocked even faster while packing GDDR6 memory, or an actual new architecture.


2019 will be interesting in that AMD bypassed the 12nm step altogether and will go for 7nm when it comes to GPU's as shown with Vega, so the question is if Nvidia ( specifically TSMC ) will be ready for high volume 7nm production the way Global foundries seems to be gearing up for.


This could an interesting situation in a year's time....:)
 
I like the talk about finally using infinity fabric with 2 or more gpus. But will it allow seemless crossfire support in games whether coded for it or not...

We'll have to wait and see...
 
I like the talk about finally using infinity fabric with 2 or more gpus. But will it allow seemless crossfire support in games whether coded for it or not...

We'll have to wait and see...

depends on how they do it but I do think that is the goal and why they are using on chip HBM

if say two or more GPU's working on one 32gb+ memory pool all at once I don't see why not
windows would probably need to know about it thru a driver but not the games hopefully
 
Supposedly, Nvidia will release a refresh part in august, as it's features will be discussed in a developers conference that I can't even remember the name of it, but it could be anything from a Pascal made at 12nm and clocked even faster while packing GDDR6 memory, or an actual new architecture.


2019 will be interesting in that AMD bypassed the 12nm step altogether and will go for 7nm when it comes to GPU's as shown with Vega, so the question is if Nvidia ( specifically TSMC ) will be ready for high volume 7nm production the way Global foundries seems to be gearing up for.


This could an interesting situation in a year's time....:)

NVIDIA’s CEO, Jensen Huang, Says Next Generation GeForce Graphics Cards Are Still A Long Way Away

https://wccftech.com/nvidia-ceo-jensen-next-generation-geforce-gpus/


they got wind of AMD skipping 12nm and taking longer and they are going to milk 1080 & 1080 ti's till AMD puts out a gaming card .
 
https://wccftech.com/nvidia-ceo-jensen-next-generation-geforce-gpus/


they got wind of AMD skipping 12nm and taking longer and they are going to milk 1080 & 1080 ti's till AMD puts out a gaming card .




The disadvantage of being with just 2 major brands to choose from unfortunately.....It's one of those situations where you miss the old days with more brands to chose from, be it Nvidia, ATI, Matrox, 3dfx, S3, power VR.....The majority are all gone and we're down to just 2.
 
The disadvantage of being with just 2 major brands to choose from unfortunately.....It's one of those situations where you miss the old days with more brands to chose from, be it Nvidia, ATI, Matrox, 3dfx, S3, power VR.....The majority are all gone and we're down to just 2.

for a year and a half maybe

https://www.forbes.com/sites/jasone...-gaming-gpu-to-fight-nvidia-amd/#528b3dbb3578

if Raja does nof **** up another one :lol:

and intel does not end up with more patent lawsuits from both NVidia and AMD than video cards :lol:

and then AMD will go after Raja and Intel if he uses anything that even looks like AMD stuf
 
and then AMD will go after Raja and Intel if he uses anything that even looks like AMD stuf


They already are with some Intel chips having a Vega die in the same package, which for 1080p gaming is not exactly crappy by any standard.


Will they aim higher and if so, just how much higher is the question.....Intel and AMD have so many cross licensing agreements at this point, that it's hard to even know who developed what first.


It's no secret that Intel is for instance going to use the same multi die strategy that tread ripper does as a short and medium term solution, given how hard it is becoming to keep on shrinking the process node, and their latest joke with their 28 core CPU being aimed at the HEDT market, where said chip is normally clocked at 2.5 Ghz for a 200 watt TDP, but they got it to 5Ghz using a chilled water that cools the chip to -10*C and can dissipate 1700 watts of heat.....Picture follows :

https://imgur.com/CEeFky7


Seems that the chip was pushing nearly 1000 watts of power at 5Ghz.....They're scared stupid of AMD's 32 core thread ripper cooled with nothing more than this:


https://imgur.com/kjppWKw


Still running an all core turbo of 3.4Ghz, with a single core maximum of 3.7 Ghz at a 250 watt TDP......I can't wait to get my hands on this puppy in early august and double my current CPU power, while still using the same motherboard..... :drool:
 
Back
Top