AMD Radeon Pro V340 - First Dual Vega GPU

[orbitalbuzzsaw]

1 minute ago, GoldenLag said:

And that AMD had to run the card out of its effiency range to remain competetive against the 1070 and 1080.

I never said it was a good card

[GoldenLag]

3 minutes ago, OrbitalBuzzsaw said:

I never said it was a good card

Its not a terrible card. Just not the right card for the job. 

Edited August 26, 2018 by GoldenLag

[orbitalbuzzsaw]

1 minute ago, GoldenLag said:

Its not a terrible card. Just not the right card for the job. 

Yep, holding out for Navi here, hoping it has OpenRT or something like that

[AlTech]

AyyyyyMD .

[GoldenLag]

7 minutes ago, OrbitalBuzzsaw said:

Yep, holding out for Navi here, hoping it has OpenRT or something like that

Im holding out aswell. Not because i want to, not because i have to, not because i cant afford the RTX 2080ti, no its because im going to be a cheapskate about it. 

[orbitalbuzzsaw]

Just now, GoldenLag said:

Im holding out aswell. Not because i want to, not because i have to, not because i cant afford the RTX 2080ti, no its because im going to be a cheapskate about it. 

lel I don't play any games that are gonna support Raytracing or DLSS so doesn't much affect me rn but I'll upgrade when 2nd or 3rd gen RT capable cards come out

[GoldenLag]

2 minutes ago, OrbitalBuzzsaw said:

lel I don't play any games that are gonna support Raytracing or DLSS so doesn't much affect me rn but I'll upgrade when 2nd or 3rd gen RT capable cards come out

RT is asm for car games and airplane sims and games like warthunder and much more. But untill it get wide implementation it is going to be a niche that is super cool and nice to bring the industry forwars (because very little has changed over the past few years)

[orbitalbuzzsaw]

2 minutes ago, GoldenLag said:

RT is asm for car games and airplane sims and games like warthunder and much more. But untill it get wide implementation it is going to be a niche that is super cool and nice to bring the industry forwars (because very little has changed over the past few years)

Yep

[MMKing]

 

https://www.amd.com/en/products/specifications/professional-graphics/9196

 

As pointed out earlier in regards to cooling. The server market generally handle cooling themselves with the entire room the server is housed in, is dedicated to the task of cooling.

[goodtofufriday]

Still waiting on the dual vega 64 from asus...

 

[rawrdaysgoby]

Ehhhh Nahhh just 2 chips on a PCB... I thought it was infinity fabric  

[schwellmo92]

9 hours ago, PopsicleHustler said:

So how do you cool this thing?

It is housed in a rack server with fans that are blowing through the rack at over 10,000rpm

[leadeater]

11 hours ago, S w a t s o n said:

The big deal to me is the SR-IOV support. This effectively allows to passthrough GPUs to VM's in manner more useful than the current methods, like the ones Linus has used with unraid. As shown in the videos you could pass 1 card (2 GPUs) through to 32 users, with each user having a 1GB frame buffer that is kept secure from the other 31 users using the MxGPU technology.

AMD actually had SR-IOV support in the previous generation of GPUs, FirePro S9150 for example. In fact MxGPU is not a new name at all, that's what they have always called it, gotta love marketing and making something old new lol.

https://www.amd.com/Documents/FirePro-S-Series-Datasheet.pdf

http://www.tomsitpro.com/articles/amd-firepro-s1750-firepro-s7150x2-hardware-virtualization,1-3129.html

[S w a t s o n]

2 minutes ago, leadeater said:

AMD actually had SR-IOV support in the previous generation of GPUs, FirePro S9150 for example. In fact MxGPU is not a new name at all, that's what they have always called it, gotta love marketing and making something old new lol.

https://www.amd.com/Documents/FirePro-S-Series-Datasheet.pdf

http://www.tomsitpro.com/articles/amd-firepro-s1750-firepro-s7150x2-hardware-virtualization,1-3129.html

Yes and Nvidia also has SR-IOV which is how they run geforce now apparently but it's still good to see it on a crazy ass card like a dual vega. The real hope is we get it on consumer cards one day.

[leadeater]

2 minutes ago, S w a t s o n said:

Yes and Nvidia also has SR-IOV which is how they run geforce now apparently but it's still good to see it on a crazy ass card like a dual vega. The real hope is we get it on consumer cards one day.

Nvidia doesn't actually use SR-IOV for their multi user/multi session GPU solutions, their's is software and driver based. AMD is indeed the first and only SR-IOV GPU which is actually really important because Nvidia's solution doesn't support all graphics APIs, older ones usually found in CAD and medical applications. For those Nvidia use dedicated passthrough where AMD can use SR-IOV with full and proper API support.

[S w a t s o n]

3 minutes ago, leadeater said:

Nvidia doesn't actually use SR-IOV for their multi user/multi session GPU solutions, their's is software and driver based. AMD is indeed the first and only SR-IOV GPU which is actually really important because Nvidia's solution doesn't support all graphics APIs, older ones usually found in CAD and medical applications. For those Nvidia use dedicated passthrough where AMD can use SR-IOV with full and proper API support.

Huh, i was under the impression the P100 and V100 did but you're right. GG Nvidia

[laminutederire]

15 hours ago, S w a t s o n said:

Navi has been rumored to be both MCM and Monolithic, could be both if RTG is crazy enough to try it.

As you can see they have no fan either. These are meant to be used in temp controlled datacenters/server rooms. Additionally, i dont believe the expectation would be that they are running at 100% for extended perioids of time as these would be passed through to thin clients. So unless you have enough users using the card at once it wont be hitting 100% usage

It could just be something akin to what Nvidia does, just by splitting ray tracing hardware and regular hardware onto two different chips for instance.

[cj09beira]

16 hours ago, S w a t s o n said:

MCM is definitely coming for Navi in this type of card. If you look at the die shot we can see it's definitely just crossfire or crossfire on steroids.

 

The problem with MCM is it's basically crossfire at the die level. Not a problem for these workloads but a problem for gaming for sure

you sure about that, ?

16 hours ago, GoldenLag said:

And that AMD had to run the card out of its effiency range to remain competetive against the 1070 and 1080.

and the best dies went to apple which means we got stuck with the bad ones 

15 hours ago, goodtofufriday said:

Still waiting on the dual vega 64 from asus...

 

maybe with navi asus might gives something sweet like that 

[GoldenLag]

1 minute ago, cj09beira said:

and the best dies went to apple which means we got stuck with the bad ones 

Vega was far out of efficiency regardless of good or bad dies. Tbh, good on AMD making some profit selling Vega to Apple.

[Doobeedoo]

I wonder which 7nm GPU will hit consumers though. There were sayings about Vega but that's not an actual flagship successor and Navi is supposed to be like MCM with infinity fabric hm. 

[cj09beira]

1 minute ago, Doobeedoo said:

I wonder which 7nm GPU will hit consumers though. There were sayings about Vega but that's not an actual flagship successor and Navi is supposed to be like MCM with infinity fabric hm. 

navi is confirmed to not be an mcm, and vega 20 is as big as gcn gets so yes its a flagship, like it has 1TB/s of bandwidth 

[Doobeedoo]

6 minutes ago, cj09beira said:

navi is confirmed to not be an mcm, and vega 20 is as big as gcn gets so yes its a flagship, like it has 1TB/s of bandwidth 

So it's just a refined Vega on better node, refined flagship really, not generational jump. Navi is last of GCN no. 

[GoldenLag]

3 minutes ago, Doobeedoo said:

So it's just a refined Vega on better node, refined flagship really, not generational jump. Navi is last of GCN no. 

Navi is the end of CGN at least in gaming. It might continue in the enterprise as its a very good compute architecture and might work well in a multi-die scaling enviroment

[cj09beira]

1 minute ago, Doobeedoo said:

So it's just a refined Vega on better node, refined flagship really, not generational jump. Navi is last of GCN no. 

its vega with 2 times the bandwidth (or more if they use 2.4Gbps hbm), 1/2 Double precision, some new instructions (compute and encryption related), it should also have pcie 4 and xgmi

[Pixel5]

17 hours ago, PopsicleHustler said:

So how do you cool this thing?

crank the cooler up to 100% permanently and have a good ventilated Server chassi.

 

its a card for datacenters basically where noise does not matter.