Vega 12 Engineering board picures

[Madgemade]

This topic has appeared on the LTT forums. It seems to show an engineering card for the Vega 12 GPU. It's not clear why it's been posted but it does say "first look at Vega12 GPU and board" which implys that maybe AMD are going to release a desktop version of the Vega 12 GPU.
The performace shown is not very high and I don't see how this could be competitive when the RX 560 is only $105 and this is using HBM2.
Here is the main pic (other pictures and firestike scores at link):

I think this is not a leak at all and is just somebody posting an engineering board used for developing and testing the Radeon Pro Vega 20. Still interesting though. Would be nice to have if only for that overbuilt VRM.

 

[Valantar]

Yeah, definitely no way AMD are launching a consumer (or even enterprise, really) board based on this chip at this time. Vega (and derivatives) might live on in the datacenter, but definitely not in this form. The layout and design of the board isn't suited to being cut down/not populated for a consumer version, so this is for engineering purposes only.

I'm betting this is a testing/debugging board either for internal AMD use or for Apple (considering they're the only ones using these GPUs).

 

[FordGT90Concept]

That looks like the GPU paired with Intel CPUs. Maybe it is what they used to test it.

 

[Jism]

Such a overbuild VRM could only come in place when your LN2'ing the card to pieces. Otherwise the most stock VRM's do suite any video card or CPU for that matter 'right'. Even under extremer conditions.

Apart from that: engineering boards are always way bigger then consumer / final product(s). It's to make sure to get a proper baseline, readings and all that before it's being published.

That looks like the GPU paired with Intel CPUs. Maybe it is what they used to test it.

So yeah how do you think the CPU will get it's communication from? All over the PCI-E bus? It's a vega 12 chip with 1 stack of 4GB HBM2 memory.

 

[FordGT90Concept]

So yeah how do you think the CPU will get it's communication from? All over the PCI-E bus? It's a vega 12 chip with 1 stack of 4GB HBM2 memory.

Yes

The Intel NUC8i7HVK (Hades Canyon) Review: Kaby Lake-G Benchmarked

www.anandtech.com

Vega M was connected to the i7-8809G CPU via PCIe x8 lanes. When testing the GPU, they could have easily traced those PCIe x8 to the pictured PCI Express slot.

Test PCBs are always overbuilt so they don't damage the prototype GPU (which is far more expensive than the PCB). If power spikes, the VRMs can handle it, they can catch it, and investigate why.



Edit: Maybe it is a smidge bigger, which could mean it was designed for a Mac product instead of an Intel NUC:

Either way, it's nothing to get excited about. The picture is most definitely old.

 

[Jism]

Perhaps you should click the link in the article:

Same chip, but now on a actual graphics board.

 

[FordGT90Concept]

Look at the date: November 30, 2017. These smaller boards were probably for software engineers to code on/test after AMD got FCC approval for production on the engineering sample board.

I'm still not convinced this is ever going to see retail channels because performance is close to RX 560:

AMD Radeon RX Vega M GL Graphics Review

AMD Radeon RX Vega M GL Graphics Graphics Card review with benchmark scores. See how it compares with other popular models.

benchmarks.ul.com

Edit: noticed the stickers are covering where some PCIe connection tabs would be. Definitely x8.

 

[Madgemade]

Yeah, this looks like more like someone who got hold of an engineering board and want to post so pictures. They would be better off selling it to a collector. I'm sure some on here would want it. I sure would if only for the ablity to get top place for overclocking it...

 

[El Cinico]

Back when the hype for Vega was just starting, I expected them to release mainstream Vega card too, full top to bottom architecture. My guess was that's why they developed (and talked a lot about) HBCC, so they could ship cheaper midrange Vega's with 4 gigs of HBM2 without it being bottlenecked by framebuffer capacity.

Maybe we never got this card because Vega didn't perform as well as AMD expected it to, so they axed the project and used Polaris for 3 generations up to navi. This could be a prototype for that mythical midrange desktop Vega that never came.

 

[FordGT90Concept]

HBM2 is expensive and packaging for it is expensive too. Discreet Vega cards were never meant to be mainstream. It did get used in mainstream by way of 2200G, 2400G, and Hades Canyon. Compare Hades Canyon prices to 2200G and 2400G though: much more expensive. I think that reflects how much HBM2 is compared to GDDR/DDR.

In other words, if you don't need crazy high memory bandwidth or a really small package size, GDDR is the economic way to go. Maybe at one point AMD was hoping HBM2 pricing would get competitive with GDDR but it never happened.

 

[El Cinico]

Well in the case of supposed mainstream Vega, economics of scale would play a big role, ordering massive amounts of HBM2 would drop the price per stack and packaging would also be cheaper I expect.
But even with cheap HBM2 it's all pointless if the architecture in question is actually slower then Polaris (in gaming).