• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD 7nm GPU News and Rumours 2018/2019

People keep envisioning this future as being smaller versions of current monolithic cores tied together with an interconnect but that isn't the most effective and probably the eventual way MCM GPUs would be done - a proper MCM design that can scale effectively would to some degree separate out the systems that exist within a current single GPU core package making use of advances in substrate technology and specialised interconnects along with 7nm and smaller processes to make it possible.


That's literally what I stated in that post.

Although on size, there is no requirement for it to be 7nm or below. It just wasn't financially required at large processes. Likewise it's more complicated and process nodes changed every 2-3 years so there was little value in moving towards such complex (and thus more expensive) designs till around about now. As we've moving into a period of slowing process nodes, increasing costs per transistor and worsening yields on smaller processes (though that will be somewhat reversed in the shorter term with EUV), then trying to create chips with more transistors at lower costs naturally leads to chiplet designs. Also as process nodes slow down gains will be have to be sought out elsewhere. That will include optimising chips by making different parts of them on different process nodes. For an APU that could mean the cpu cores on an 7nm FDX chip from Glofo 5 years from now alongside a 5nm bulk denser process for the gpu.
 
If that were true then they wouldn't have quietly scrapped the Vega Nano.

Not sure why scrapping the Vega Nano has anything to do with Navi been a Polaris replacement?

I know it was but what I'm saying is that I'd put money on the reason they scrapped the plan for a Vega Nano being because Vega was not good for it, not that they scrapped the Nano because Vega was being replaced, they wouldn't of done that, It's much more likely they scrapped it because they weren't getting enough cherry picked chips capable of sustaining a line of Vega Nano's, The fact that Powercolor are using the Vega 56 for their Nano instead of the full chip supports the theory as well.

And the reason they got rid of the Nano was simply because of demand for Vega cards for mining. It would have been a waste of time and money. What AIB's do going forward is a different story as it might be worth their while making a Nano now to sell more Vega chips.
 
That's literally what I stated in that post.

Wasn't contradicting you - just that as per what you described a bit below what I quoted a lot of people's idea of MCM in terms of what would enable more efficient larger GPUs and come closer to realising the ideal of tying together multiple GPUs as if a larger logical unit is different to the reality.

7nm or below is kind of needed though in terms of the kind of interconnects that would be used related to latency, etc. and thermal/power properties with the amount of space stuff would be in.
 
Because if Navi was just a Polaris replacement then they wouldn't have scrapped the improved Vega. Hence we know Navi isn't just a Polaris replacement.

The Nano wouldn't have been an improved Vega though. What gave you that idea? Fury Nano wasn't an improved Fury. It was just the best Fury chips possible, the most power efficient Fiji cores they had to fit in such a small form factor.

And all the rumours point to Navi been a Polaris replacement.
 
Confirmed :eek:

AMD To Introduce World’s First, 7nm, High-Performance CPUs and GPUs at CES 2019 – Aimed To Catapult Computing and Gaming Technologies Forward https://wccftech.com/amd-intros-7nm-cpus-gpus-ces-2019/

We have known about the Radeon Instinct coming for ages now.

And she is giving a Keynote speech, not releasing new products. So, no, nothing for the gaming market confirmed at all.
 
That's what Infinity fabric is. It's not just about GPU's or CPU's, it's about different tech communicating seamlessly.

Well that's where my confusion came from, I missed the part about IF actually being a technology concept and implementation of communication. I thought it was just the fancy name for Zen's silicon interposer.

Well, you live and learn.
 
Yes it is a keynote - but I am definitely expecting her speech to be punctuated by either the launch or announcement of at least one product. CES is too good an opportunity to miss.

This is CES so I am not expecting it to be about enterprise kit.

There are two possibilities to me. The way the announcement is worded focusing on VR and Gaming... it might be possible that this will be about Console and not PC.

I think it is far more likely to be a glowing conversation about 7nm the continuing journey about Zen, the confirmed progress to 7nm GPU's but I think there is likely to be a 'surprise' product.
 
That's literally what I stated in that post.

Although on size, there is no requirement for it to be 7nm or below. It just wasn't financially required at large processes. Likewise it's more complicated and process nodes changed every 2-3 years so there was little value in moving towards such complex (and thus more expensive) designs till around about now. As we've moving into a period of slowing process nodes, increasing costs per transistor and worsening yields on smaller processes (though that will be somewhat reversed in the shorter term with EUV), then trying to create chips with more transistors at lower costs naturally leads to chiplet designs. Also as process nodes slow down gains will be have to be sought out elsewhere. That will include optimising chips by making different parts of them on different process nodes. For an APU that could mean the cpu cores on an 7nm FDX chip from Glofo 5 years from now alongside a 5nm bulk denser process for the gpu.

The requirments fo smaller ndoe sizes are related to bandwidth, latency, cahce sizes and coherency etc. It depends on the workload, Some of the papers I have read suggest for HPC 7nm would be sufficient, but gaming operates at a much higher frequency with smaller data sizes so real performance gains wouldn't come until 5nm or smaller.



Ironically, Realtime ray tracing is actually a perfect candidate for an MCM design sicne there is far easier parallelism and next to zero spatial coherency. Every ray is entirely independent and since you want to shoot mutliple rays per pixel then duplication of the scene BVH is acceptable.
Current rasterization technologies make MCM designs harder to truly scale
 
Because if Navi was just a Polaris replacement then they wouldn't have scrapped the improved Vega. Hence we know Navi isn't just a Polaris replacement.


Vega Nano was going to be the exact same Vega chip thoguh, just hand picked like with Fury Nano and a some downclock/voltage.

We do know that navi is not a new architecture, that git postponed when Raja left so everything has been renamed,
 
We have known about the Radeon Instinct coming for ages now.

And she is giving a Keynote speech, not releasing new products. So, no, nothing for the gaming market confirmed at all.


The Ryzen 7nm CPUs will be released and that will be the gaming reference. 7nm Vega GPU will be for HPC. At some point mid-2019 there will be a gaming 7nm GPU but all the rumours indicate something that replaces Polaris and not a high end GPU.
 
The Ryzen 7nm CPUs will be released and that will be the gaming reference. 7nm Vega GPU will be for HPC. At some point mid-2019 there will be a gaming 7nm GPU but all the rumours indicate something that replaces Polaris and not a high end GPU.

That would be a wise move given the current state of the market. Cheap GTX 1080 performance for the masses could sell very well. May help adoption of 1440p too.
 
Vega Nano was going to be the exact same Vega chip thoguh, just hand picked like with Fury Nano and a some downclock/voltage.
No it wasn't, that wouldn't even have been possible (No Vega 64 chip would work at stock clocks with that little power/cooling). It was going to be a 7nm Vega refresh launched around early 2019, but it's been quietly scrapped in the wake of Rajagate.
 
Back
Top Bottom