Info Big Navi - Radeon 5950 XT specs leak.

[Hans de Vries]

Seems this might be a valid leak from inside partner SK Hynix..

Twitter

Big Navi - Radeon 5950XT: Twice the compute units as Navi 10.

Shading units: 5120
TMUs: 320
Compute units: 80
ROPs: 96
L2 cache: 12MB
Memory: 24GB (4 x HBM2e, 3 die)
Memory bus: 4096 bits
Band Width: 2048 GB/s

All these Big Navi numbers are perfectly consistent. The 96 ROPs are tiny pieces of logic at the edge of the memory tiles of the 12MB L2 cache which explanes the factor 3 in these numbers. An old example here https://bjorn3d.com/2010/01/nvidia-gf100-fermi-gpu/

SK Hynix will make 6GB HBM2e stacks with 3 dies per stack for consumer applications on request. SK Hynix recently announced HBM2e at 512GB/s at ISSCC-2020. Samsung went a step further with 640GB/s HBM2e (5 Gb/s/pin)

 

[Hitman928]

AMD Financial Analyst Day shows a projected 50% perf/watt gain for RDNA2:

View attachment 17795

Gimme 5700 XT performance at half power and you can have my money.

I believe they are moving to 7+ for RDNA2 which means they'll get a bit of an efficiency bump there but most of that efficiency improvement should be coming from the architecture. Interested to see what did with it.

 

[exquisitechar]

I believe they are moving to 7+ for RDNA2 which means they'll get a bit of an efficiency bump there but most of that efficiency improvement should be coming from the architecture. Interested to see what did with it.

Yeah, considering that Navi 10 is already on N7P, moving to 7+ will not get them much.

 

[soresu]

Sadly, those of us who aren’t interested in high end and very high end cards are being left out with Navi 2x. Oh well, RDNA3 is next year.

Maybe not - rumours put 3 different 2x chips on the cards 21,22 and 23.

That has to be at least 6 SKU's.

 

[maddie]

AMD Financial Analyst Day shows a projected 50% perf/watt gain for RDNA2:

View attachment 17795

Gimme 5700 XT performance at half power and you can have my money.

2/3 rds the power. With 7+ a claimed 15% more efficient, then they are doing better from the design viewpoint that the GCN > RDNA jump.

 

[JasonLD]

I believe they are moving to 7+ for RDNA2 which means they'll get a bit of an efficiency bump there but most of that efficiency improvement should be coming from the architecture. Interested to see what did with it.

Doesn't look like RDNA2 is using 7nm+

AMD's 2020-2022 Client GPU Roadmap: RDNA 3 & Navi 3X On the Horizon With More Perf & Efficiency

www.anandtech.com

And given TSMC’s roadmaps, it’s more or less inevitable that this will be the point where AMD begins using an EUV-based process for their GPUs, as AMD has indicated that this year’s RDNA 2 will not be using TSMC’s EUV-based 7nm+ process.

 

[soresu]

2/3 rds the power. With 7+ a claimed 15% more efficient, then they are doing better from the design viewpoint that the GCN > RDNA jump.

Especially considering it didn't look so hot going from Vega 2 to RDNA - that alone had me holding back.

 

[soresu]

Doesn't look like RDNA2 is using 7nm+

AMD's 2020-2022 Client GPU Roadmap: RDNA 3 & Navi 3X On the Horizon With More Perf & Efficiency

www.anandtech.com

They also don't mention Zen+ directly (only as 12nm Zen, nothing about its enhanced IPC/cache) - don't read into it too much.

More important is that they announced Zen4 as 5nm, but did not do the same for RDNA3, makes me think either RDNA3 is delayed, or it's far enough out that they are considering waiting for the next process, either N5P or N3.

 

[JasonLD]

They also don't mention Zen+ directly, don't read into it too much - more important is that they announced Zen4 as 5nm, but did not do the same for RDNA3.

That isn't that surprising. I expect GPU will stay 7nm for a while. AFAIK, 5nm is bit slower than 7nm in terms of yield progress.

 

[NostaSeronx]

AFAIK, 5nm is bit slower than 7nm in terms of yield progress.

TSMC's 5nm is beyond 7nm in terms of yield, and far beyond 7nm+/6nm in yield progress.

Expect AMD to bum rush with everyone else to the 5nm node. Unlike the 7nm node, AMD has its own joint-developed node from TSMC.

 

[soresu]

That isn't that surprising. I expect GPU will stay 7nm for a while. AFAIK, 5nm is bit slower than 7nm in terms of yield progress.

They do say "Advanced Node", so I expect it to be beyond 7nm at least:

 

[JasonLD]

They do say "Advanced Node", so I expect it to be beyond 7nm at least:

View attachment 17804

They can also use Samsung "5nm", which would be somewhere between TSMC's 7nm and 5nm. They do have partnership going on with GPU architecture.

 

[DXDiag]

Late 2020 release, which means after consoles probably!

 

[uzzi38]

Late 2020 release, which means after consoles probably!

Wrong. Little bit beforehand.

 

[sandorski]

AMD Financial Analyst Day shows a projected 50% perf/watt gain for RDNA2:

View attachment 17795

Gimme 5700 XT performance at half power and you can have my money.

Use 1080p/60hz and set Framerate Target to 60fps with Auto Voltage on, and you have achieved your goal.

 

[extide]

AMD Financial Analyst Day shows a projected 50% perf/watt gain for RDNA2:

Gimme 5700 XT performance at half power and you can have my money.

It would be 2/3 of the power but yeah

 

[DownTheSky]

AMD Financial Analyst Day shows a projected 50% perf/watt gain for RDNA2:

This would put a 300W TDP card @125% faster than 5700 XT.

 

[Qwertilot]

Need to be a bit careful with this though - I believe they've got previous of quoting perf/watt gains as being from what are effectively factory overclocked cards vs ones clocked to sensible levels?

We'll see how well it turns out

 

[exquisitechar]

Need to be a bit careful with this though - I believe they've got previous of quoting perf/watt gains as being from what are effectively factory overclocked cards vs ones clocked to sensible levels?

We'll see how well it turns out

Edit: Remembered wrong.

 

[EXCellR8]

In.

 

[Mopetar]

Exciting news no matter how you split it, but it does seem to Osbourne their existing products. Why would I want to buy a 5700XT now when I can wait half a year and get something 50% better?

 

[Leadbox]

Need to be a bit careful with this though - I believe they've got previous of quoting perf/watt gains as being from what are effectively factory overclocked cards vs ones clocked to sensible levels?

We'll see how well it turns out

Well they didn't get their Zen power management ppl over to RTG just to count transistors.

 

[Qwertilot]

Hopefully not anyway Once they can use the dividends from Zen in the GPU division they should get much closer to NV.

It is, being more serious, really quite hard to do a 'basic' perf/watt for GPU's. Look how much that varies between NV's various ones from the crazy OEM chips, to the saner clocked references and all the way down to the MaxQ laptop ones.

Marketing departments aren't precisely known for their adherence to strict honesty.

 

[Hitman928]

Doesn't look like RDNA2 is using 7nm+

AMD's 2020-2022 Client GPU Roadmap: RDNA 3 & Navi 3X On the Horizon With More Perf & Efficiency

www.anandtech.com

There was somewhat of a clarification on this by AMD.

Because AMD labeled those as 7nm+, when TSMC called its version of 7nm with EUV to be N7+, one of the obvious assumptions that people have made is that where AMD wrote 7nm+, it was to be on the N7+ process. We have since learned that this is not entirely correct.

In order to avoid confusion, AMD is dropping the ‘+’ from its roadmaps. In speaking with AMD, the company confirmed that its next generations of 7nm products are likely to use process enhancements and the best high-performance libraries for the target market, however it is not explicitly stating whether this would be N7P or N7+, just that it will be ‘better’ than the base N7 used in its first 7nm line.

 

[extide]

It might, it might not, they just aren't willing to share yet.

 

[Saylick]

This would put a 300W TDP card @125% faster than 5700 XT.

Wouldn't that be 100% faster (2x the performance)?

5700 XT is 225W TDP. If RDNA2 is 50% better performance per watt than RDNA 1, then that implies RDNA 2 can achieve the same performance at 225W / 1.5 = 150W. With a 300W TDP, that's 2x the TDP and assuming linear scaling, it's 2x the performance over a 5700 XT, which is inline with the rumored performance for Big Navi.