AMD Navi 23 ‘NVIDIA Killer’ GPU Rumored to Support Hardware Ray Tracing, Coming Next Year

[Sargatanas2511]

thats ur assumption, I have an idea what the performance is you can’t tell me what I do and don’t know what sort of non-sense is that. Anyone can easily estimate the performance of RDNA2 in a best case scenario and work back there is enough data out there already for that. All we lack really are the die sizes and CU counts at this point

I am assuming nothing except that only people inside AMD know how good an RDNA2 CU is. Are you telling me you know how good it is? A best case scenario could be anything from RDNA2 having double tht FP/Int throughput in a single CU to massively increased IPC and nothing else.

At this stage we have been given little information on what has changed, so how could you know if a 60CU part would be good enough or not?

 

[SmokSmog]

You are aware the watts of a PSU are not the same as TDP right? The PSU will be rated way above the actual power draw needed. So TDP for the systems is way lower than the actual PSU power rating.

Yes I'm aware of this, this is why I said that XSX looks promising for RDNA 2

XSX has 255W (12V/21.25A) power rail for APU and memory. We know that it won't pull that much because you need some buffer, let's say 200W max for APU, remove CPU power and you have 150-160W for 52CU GPU at 1825Mhz.

AMD can double XSX GPU

112CUs 128ROPs 350W monster with HBM2E clocked at 1,8-1,9GHZ or fast 80CU + 2GHZ

 

[BigBANGtheory]

I am assuming nothing except that only people inside AMD know how good an RDNA2 CU is. Are you telling me you know how good it is? A best case scenario could be anything from RDNA2 having double tht FP/Int throughput in a single CU to massively increased IPC and nothing else.

At this stage we have been given little information on what has changed, so how could you know if a 60CU part would be good enough or not?

Im not telling you anything, make ur own mind up you asked me a question I answered it.

For me its 80 CU plus or bust at this point because I believe that has enough grunt to walk past a 3080 and challenge a 3090, others may be happy with less good for them.

 

[Sargatanas2511]

Im not telling you anything, make ur own mind up you asked me a question I answered it.

For me its 80 CU plus or bust at this point because I believe that has enough grunt to walk past a 3080 and challenge a 3090, others may be happy with less good for them.

So if a 60CU part does come along and beats the 3080 (not saying it will) you won't buy it because it doesn't have enough CUs?

 

[Lokken86]

If I was a betting man (I'm not ) I'd say there's an 80% chance AMD's best RDNA2 SKU is slower than a 3080.

It may have been possible to build a faster GPU, with some rumours AMD is holding back and simply wants something to rival the 3070.

This is what I think aswell. AMD know that whatever they make Nvidia are likely just to release a TI version or something and best it. Then they are left having to make big expensive gpus and in no better situation.
Maybe in a few more generations if they close the gap a little more each time, we can start seeing them get a bit more ballsy like what we've seen with Ryzen, and start gunning for the top end.

 

[BigBANGtheory]

So if a 60CU part does come along and beats the 3080 (not saying it will) you won't buy it because it doesn't have enough CUs?

Correct cos having seen its only 60 CUs I will have made a judgement call that its a 3070 competitor and proceed with a 3080 order. Then in possession of a 3080 I would be looking at my next GPU upgrade which for me tends to be +100% performance which could be 2 generations away but I'd happily consider RDNA 3 or 4 if the goods are there. So no I can't see any circumstances what so ever that I would buy a 60 CU Big Navi.

 

[g67575]

Did AMD ever confirm how much the performance per CU was improved between Vega and Navi GPUs?

The rx 5700 XT has 50% less CUs than the Radeon VII, for nearly the same performance, so perhaps each Navi CU has upto 50% greater performance than Vega?

If we got a similar improvement (+ 50%) with RDNA 2, the PS5 GPU (same CU count as the RX 5700) would be almost as fast a RTX 2080 TI. Maybe faster if you take into account the increased clock rate.

Maybe AMD claiming +50% performance per watt per gen is just another way of saying more efficient CUs?

 

[Sargatanas2511]

Correct cos having seen its only 60 CUs I will have made a judgement call that its a 3070 competitor and proceed with a 3080 order. Then in possession of a 3080 I would be looking at my next GPU upgrade which for me tends to be +100% performance which could be 2 generations away but I'd happily consider RDNA 3 or 4 if the goods are there. So no I can't see any circumstances what so ever that I would buy a 60 CU Big Navi.

I think you need to read what I said again.

 

[gpuerrilla]

I think you could be in for just as much of a shock as Primoz Roglic!

 

[g67575]

I think a significant increase in the performance of each CU is probably the only way AMD will compete with the RTX 3080. IPC perf. increase seems unlikely cause they are still using the same fab. process.

 

[Nablaoperator]

IPC is dependent on architecture and not process, it's a theoretical measure determined through workload simulations. Don't think the process influences that a lot.. Since, there are rumors regarding a 6900, I believe the chip will clock higher compared to the 3090

 

[elroberto]

I think a significant increase in the performance of each CU is probably the only way AMD will compete with the RTX 3080. IPC perf. increase seems unlikely cause they are still using the same fab. process.

Actually we're not sure which variant of TSMC 7nm it is using. The more advanced nodes will improve that if used

 

[g67575]

No reason to think they are using an improved 7nm process. They removed the '7nm+' from the marketing slides for RDNA 2. Probably, because 7nm EUV isnt ready yet.

 

[EastCoastHandle]

IPC is dependent on architecture and not process, it's a theoretical measure determined through workload simulations. Don't think the process influences that a lot.. Since, there are rumors regarding a 6900, I believe the chip will clock higher compared to the 3090

Pretty much. Node shrinks have very little to do with IPC improvements. People believe that RDNA 2 is the exact same Uarch as RDNA 1. Lol.

 

[g67575]

meh its unknowable, probably even once its released.

I guess youve probably all seen this engineering sample photo by now:

https://cdn.videocardz.com/1/2020/09/AMD-Engineering-Sample-Photo-1200x472.jpg

I think it says 56 cores on the label, maybe an unlocked version of the series X GPU, which has 4 CUs disabled. Lots of people assuming that it therefore must be a cut down GPU (navi 22).

One guy estimated the die size to be 320-360mm2, so not massive.

Or, maybe just 'secure' lol, AMD dont want the core count to be known by anyone. Or this it a complete red herring.

 

[gpuerrilla]

**** evidence to go on theres bound to be more leaks coming that will help gauge.

 

[Jacky60]

I'm betting big Navi (the 80CU part) is 15% faster than 3080 at 4K 60. Level with 3090 sometimes, losing other times but ahead of 3080 without DLSS or RT. Priced at £599. That would make me smile too. Watch this space....

 

[gpuerrilla]

I cant see something that trades blows with a 3090 being half the price, but I wouldnt say no if they felt like being generous!

 

[Besty]

I'm betting big Navi (the 80CU part) is 15% faster than 3080 at 4K 60. Level with 3090 sometimes, losing other times but ahead of 3080 without DLSS or RT. Priced at £599. That would make me smile too. Watch this space....

its highly unlikely, based on the way the cards appear to be falling, the single, only reason, for the 3090s existence appears to be in order to beat the 6900XT by 'a margin', whatever that margin turns out to be.

It could be only 15% but that margin will almost certainly be there, nvidia need to protect their cash cow, even if it turns out to be a bit of a 450w turkey.

AMD have to have a big amount of stock and so i will expect them to go £50 more expensive than a 3080 with ample supply to get everyone a card. Nvidias rushed launch has given them that opportunity.

 

[Shiari]

AMD have to have a big amount of stock and so i will expect them to go £50 more expensive than a 3080 with ample supply to get everyone a card. Nvidias rushed launch has given them that opportunity.

They would have had to be making cards for months to be able to do this, surely we would've had more leaks by now if cards were sitting in warehouses or were being shipped in any quantity? Yeah I'd like it to be true, but I fear any availability will be low for a while.