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

[humbug]

I tend to agree with Mores Law is Dead.

AMD don't have infinite amount of 7nm wafer suply and they are selling more CPU's than they can produce, they could make four or even six 8 core CPU's with the die space one Big Navi will take up.

Clearly its more profitable for them to use that die space to make those CPU's which they know will have been sold before they are even made.

If AMD are to make a large 500mm2 GPU they will be doing it to regain some marketshare and mindshare in the GPU space, it will be a cost investment to them. I hope AMD are of the mind to make that investment, taking Nvidia on like this has not worked out well for them in the past so here's hoping AMD feel like with the good will and respect people have for their Ryzen CPU's some of it will rub off on the GPU side and people start buying into the brand.

AMD. Please just do it, we want to see you back battling with Nvidia.

Although I concurred with MLID's viewpoint on that, I did raise an eyebrow when he mentioned 7NP as the node for Big Navi. Zen 2 and Zen 3 aren't 7NP. I know the nodes are design-compatible with each other, but I was under the impression TSMC had physically different machines for each of their 7nm variants. If that's the case then AMD don't have to choose between CPUs or GPUs.

Of course, which node is TSMC using for the consoles?

Zen 3 is also coming... Zen 2 is end of life.

 

[Phixsator]

Yes its a bit of a problem, AMD will prob sell every Zen chip they make while huge area GPUs wont - then again they could make a total whopper and market it like the titan - limited edition crap ect... if its a monster folk will have to have it.

They could certainly do a very limited run just to say they have the best and then stick to a 3070 contender GPU which allow for much more profit.

 

[Nerusy]

It makes perfect sense if you want to make an informed choice and spend your money wisely.

OK, I get that historical precedent is a thing, but what exactly is wrong with waiting? You're interested in product X, product Y is coming to offer competition and choice, product Y ultimately proves to be unatttractive so you buy product X. You've lost a bit of time, but it's hardly the end of the world. Or product Y proves to be the better choice so you benefitted from waiting.

lol OK

Well this is the main point, how long to wait. I thought there was no announcenet for big navi and going by the precedent it can take several months. If we know that big navi announcement will follow soon then sure, waiting a couple of months is very reasoanble.

Edit: apperently amd's cfo said end of year, aiming to release big navi before next-gen consoles. So looking at November - December.

 

[Twinz]

If I were AMD and I actually had the parts to do it, I would post my own Doom video where our GPU is beating the 3080. End that video with a reasonably close date and price of $699.

-Release said Doom video on September 17th and drop the mic.

 

[LePhuronn]

I thought there was no announcenet for big navi

Amd have said they're launching the cards before the consoles. The consoles are scheduled for November (I believe). So that's 2 months max. When exactly nobody has said (rumoured September announcement and October 7th release, but it's a rumour nobody's cited a source for). Beyond the initial rush of launch-day sales, you won't be able to buy Ampere anyway, so you'll be stuck waiting regardless.

Arguably AMD should say something a lot sooner, arguably they don't need to, arguably they don't actually care, but these are different matters.

 

[LePhuronn]

-Release said Doom video on September 17th and drop the mic.

16th, no point dropping it on the actual day Ampere goes on sale. People will be too busy spamming F5 to notice anything else

 

[humbug]

AMD could have made a larger RDNA1 card, the 2080TI is a whole 35% faster than the 5700XT, big woop....... a 56 CPU 5800XT with a 384Bit bus would have been a 2080TI competitor, that's 40% more GPU on a 250mm2 40 CU 200 Watt GPU, 280 Watts and 350mm2 napkin maths 2080TI.

They didn't, IMO because they couldn't or didn't want to spare the wafer space.

That is what worries me.

 

[Phage]

Well this is the main point, how long to wait. I thought there was no announcenet for big navi and going by the precedent it can take several months. If we know that big navi announcement will follow soon then sure, waiting a couple of months is very reasoanble.

Edit: apperently amd's cfo said end of year, aiming to release big navi before next-gen consoles. So looking at November - December.

Agreed. November will be crucial. That's when I'll be buying

 

[Twinz]

16th, no point dropping it on the actual day Ampere goes on sale. People will be too busy spamming F5 to notice anything else

-just to avoid giving Nvidia any time to make a face-saving adjustment.

 

[LePhuronn]

AMD could have made a larger RDNA1 card...they didn't, IMO because they couldn't or didn't want to spare the wafer space.

Or it was scaling issues and power draw, making over 40 CUs unfeasible.

 

[BigBANGtheory]

3080 lands, 6900X beats it, 3080 Ti beats it back, 6950XT comes back again, Nvidia waves the 3090 as the untouchable, 6990XTX 400W HBM2e AIO-cooled monster gets within punching distance enough for people to take AMD seriously, even if they didn't take the actual crown. Hell, if this mythic unicorn 6990XTX beats the 3090, Nvidia will just drop an actual Titan when Samsung's yields improve.

Oh how I can dream

Its actually tougher than that in in practice though because Nvidia has normalised the 3080 price/perf back to where is should have been. So as soon as AMD go over $699 people will want to compare it with a 3090 and less so with a 3080 imho.... Whilst I'm in no way brand loyal to Nvidia you have to admit they've been quite cunning this time there is not looking like any easy win for AMD because even if they have the goods its not going to be cheap.

 

[LePhuronn]

-just to avoid giving Nvidia any time to make a face-saving adjustment.

It's a fine line indeed.

 

[Twinz]

AMD could have made a larger RDNA1 card, the 2080TI is a whole 35% faster than the 5700XT, big woop....... a 56 CPU 5800XT with a 384Bit bus would have been a 2080TI competitor, that's 40% more GPU on a 250mm2 40 CU 200 Watt GPU, 280 Watts and 350mm2 napkin maths 2080TI.

What about the VR benchmark where an AMD card is beating the fastest *overclocked* 2080Ti in the data base?

They already have that card.

 

[humbug]

Or it was scaling issues and power draw, making over 40 CUs unfeasible.

It really isn't, some 5700XT just have bad coolers on them, they are 225 Watts total board power, the GPU alone about 190 Watt's, tweaked about 170 watts with more performance.

40% on top of a 190 Watt GPU (the rest, GDDR6, scales relative) that's 265 Watt's. a 2080TI is 250 Watt's, yes that's the GPU Nvidia don't include the VRam in their TDP figures. AMD have made GPU's which have as high or higher power requirement than that, including on the same 7nm, the Radeon VII.

 

[LePhuronn]

Its actually tougher than that in in practice though because Nvidia has normalised the 3080 price/perf back to where is should have been. So as soon as AMD go over $699 people will want to compare it with a 3090 and less so with a 3080 imho....

Perhaps, but it depends on mow much over $699 they'll go if they do at all. If AMD bring competition, the game is all about Nvidia countering AMD, not the other way around. And we saw that with the Supers last year. 5700 XT came out, beat the 2070 and a slightly lower price point. 5700 came out, destroyed the 2060 but priced slightly over. Nvidia dropped the Supers at a higher price point to counter.

So repeat that. AMD beat the 3080 at the same or slightly lower price point. Nvidia release the 3080 Ti to take back the performance not the price, and obviously charge Ti money. The comparison is still doesn't bring in the 3090. AMD can then take on the Ti at the same, or slightly lower, price point. Again the 3090 only comes into the comparison with the mythic unicorn 6990XTX I mentioned to get to 3090 performance.

Ultimately this is all based on AMD coming in at the same price point and same tier. If we get a repeat of Vega where they charge 1080 money for 1070 performance then it's a joke. It'd be amazing if AMD come in with performance 1 tier above the price point (e.g. 3080 performance for 3070 money) but that's not going to happen.

 

[Nutella33]

It sounds like you are assuming AMD can't undercut Nvidia without "killing" their own margins.

Not sure where you picked that from. They can undercut Nvidia (like the 10%~ suggested in my post, or the recent pricing alteration rumour for the 6000 series) but a significant cut whilst offering more performance, like the post I quoted suggested, is significantly different to not being able to undercut at similar performance levels at all. Performance generally has a cost somewhere after all, be it better/bigger RAM, Die size, Processes etc.

Selling the same or similar performance for smaller margins than your competitor is a good way to destroy yourself financially down the line when people still dont buy the cards. As we saw in AMD's recent past.

 

[LePhuronn]

It really isn't, some 5700XT just have bad coolers on them, they are 225 Watts total board power

It's not about that, it's about an architectural flaw. Remember when we were expecting Navi at CES but got the Radeon VII instead? There was talk of an issue with Navi in the preceding September that required some reworking and ultimately delayed it. There's been suggestions that Navi had an issue moving past 40 CUs, it just wouldn't scale. So the 40 CUs that did work got the **** clocked off them, the RX690 branding was dropped and 5700 XT was born, moving the entry-level die up to the midrange, and the long-cancelled gaming 7nm Vega project hurriedly dug back out and pushed out as a PR stunt.

Although I do agree with the idea that AMD would prioritise wafers to their CPUs, I don't buy it as the reason Navi 1 never went past 40 CUs. The tech was there with RDNA 1 for AMD to regain some performance competitveness, regain some mindshare, regain some marketshare and ultimately have RTG make its own money instead of being subbed by Ryzen. The simple fact that they bothered with Radeon VII at all showed they did care about making a showing against Turing, so I just don't see AMD going "nah, we'll just make CPUs instead of Navi". And I would say this doubly applies to RDNA 2 and Big Navi today.

But it's all speculation and we'll never know.

 

[humbug]

It's not about that, it's about an architectural flaw. Remember when we were expecting Navi at CES but got the Radeon VII instead? There was talk of an issue with Navi in the preceding September that required some reworking and ultimately delayed it. There's been suggestions that Navi had an issue moving past 40 CUs, it just wouldn't scale. So the 40 CUs that did work got the **** clocked off them, the RX690 branding was dropped and 5700 XT was born, moving the entry-level die up to the midrange.

Although I do agree with the idea that AMD would prioritise wafers to their CPUs, I don't buy it as the reason Navi 1 never went past 40 CUs. The tech was there with RDNA 1 for AMD to regain some performance competitveness, regain some mindshare, regain some marketshare and ultimately have RTG make its own money instead of being subbed by Ryzen. I just don't see AMD going "nah, we'll just make CPUs instead". And I would say this doubly applies to RDNA 2 and Big Navi today.

But it's all speculation and we'll never know.

I hope your right

 

[LePhuronn]

I hope your right

I do too, because as we've both said RDNA 1 had the potential to match and beat the 2080 Ti in rasterisation, and a lot of this mindshare tomfoolery would've been addressed a year ago. And if RDNA 1 would've taken AMD back up the fighting stack, there'd be much less naysaying about RDNA 2.

 

[humbug]

I do too, because as we've both said RDNA 1 had the potential to match and beat the 2080 Ti in rasterisation, and a lot of this mindshare tomfoolery would've been addressed a year ago.

True.