Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Quick Links
AMD says Navi is all new "RDNA" architecture, launching in July as Radeon RX 5000 series
https://techreport.com/news/34609/amd-kicks-off-computex-with-a-12-core-ryzen-9-cpu-and-some-navi-news
Not to be outdone by Nvidia going 900 series, 1000 series, 2000 series, and then 1600 series (which is actually part of the 2000 series); AMD decided to go 400 series, 500 series, Vega series, VII not a series, and then 5000 series. That's five thousand series, not five hundred series, which launched in 2017. And also different from the Radeon HD 5000 series that launched about a decade ago.
Besides the goofy name, AMD is claiming that RDNA offers a 25% improvement in performance per clock and 50% improvement in performance per watt as compared to Vega. If the latter is in comparison to 14 nm Vega cards, than that's less than you'd hope for from just the die shrink.
AMD also announced that there will be a future announcement of more stuff about Navi on June 10, even though that's still well ahead of the launch. Because nothing says newsworthy quite like announcing that there will be a future announcement without announcing what that future announcement will actually say. Or something.
Apparently it is claimed that the 5000 series was chosen because it had 50 in it, to celebrate this being the 50th anniversary of AMD existing. And because the 500 series didn't already have 50 in it. Or something. And apparently someone at AMD actually gets paid to come up with these names.
Not to be outdone by Nvidia going 900 series, 1000 series, 2000 series, and then 1600 series (which is actually part of the 2000 series); AMD decided to go 400 series, 500 series, Vega series, VII not a series, and then 5000 series. That's five thousand series, not five hundred series, which launched in 2017. And also different from the Radeon HD 5000 series that launched about a decade ago.
Besides the goofy name, AMD is claiming that RDNA offers a 25% improvement in performance per clock and 50% improvement in performance per watt as compared to Vega. If the latter is in comparison to 14 nm Vega cards, than that's less than you'd hope for from just the die shrink.
AMD also announced that there will be a future announcement of more stuff about Navi on June 10, even though that's still well ahead of the launch. Because nothing says newsworthy quite like announcing that there will be a future announcement without announcing what that future announcement will actually say. Or something.
Apparently it is claimed that the 5000 series was chosen because it had 50 in it, to celebrate this being the 50th anniversary of AMD existing. And because the 500 series didn't already have 50 in it. Or something. And apparently someone at AMD actually gets paid to come up with these names.
Comments
It does seem to be landing in a somewhat odd spot. The top end Navi has roughly 2070 performance - ok, I can get that, and that's something, a 2070 isn't exactly slow.
Except AMD already has that, with current gen Vega64. So, not exactly an exciting product so far when your just replacing stuff you already had, unless you can also move the price so low that it makes a lot of waves that way.
Now, I can understand AMD being excited that Navi may be a lot less expensive to produce than Vega, with no HBM and all... and that better positions them to compete in a price war. If nVidia decides to take the bait or not remains to be seen: they have in the past, but I don't think they will be so quick to do so now that they are in a much more dominant position than they have in the past.
Even if the production price is ridiculously lower than Vega and they could afford to sell it at sub $200 price points (and that's a big hypothetical IF to make a point, not that I believe this necessarily).... If AMD doesn't charge as much as the market can bear for the performance level, they will get their share price crushed.
So I expect it to not make that many waves, and AMD to price it according to nVidia's benchmark for MSRP... but then be willing to be liberal with add-ins, rebates, and sales to help move the product as needed. I wouldn't be surprised if in 9 months time the actual sales price is 60-75% of the original MSRP for high end AIB cards (which is pretty much what we had seen from most all previous generations until the mining booms), unless of course another mining boom hits.
One other thing that could stop AMD from starting a price war here is if they can't get enough capacity from TSMC. While AMD will surely be able to get a lot of capacity there, it's plausible that AMD will soon want enough to supply half of the x86 server market just for EPYC Rome alone. That's really a lot, and if they're limited by fab capacity, then the CPU chiplets will be the priority. While unlikely, it's plausible that this could lead Navi to be such a soft launch as to be basically irrelevant to gamers for quite a few months after launch.
My if is the pricing. All indications point towards it being priced the same @ $499 USD on the higher of the the two SKU mentioned. On the last AMD investor call there was a promise to increase margins which is why I mention indications on pricing as well as rumoured $$$ leaks. If indeed it comes in at $499 and is say only 5% faster or less than the 2070 I could still see people going with Nvidia. So I hope they do price it right and take advantage of the opportunity here.
Guys from Anandtech reckon the die size is around 275^2 mm with other guesses stemming upwards of 250 mm^2.
Performance similar to an RTX 2070 with a die size of 275 mm^2 would be a huge disappointment. That would be performance per mm^2 worse than Vega 20, in spite of being a new architecture, being built against a more mature version of the same process node, and not having all the compute stuff (especially double precision) that bloats the die space of Vega 20. That would be a disaster for AMD unless it has some huge redeeming factor that is completely unexpected.
But exactly the same performance with a much smaller die would be a huge win for AMD. If they can get that performance in 150 mm^2, then you can scale up the die size and next year produce a 300 mm^2 die that beats the RTX 2080 Ti while costing far less to build.
For AMD to increase margins doesn't automatically mean higher prices on consumer GPUs. It's probable that AMD will increase their gross profit margin for the company as a whole even if Navi is a disaster just because EPYC Rome will carry the company. Even on GPUs, if they can get more market share for professional cards at thousands of dollars each rather than just consumer cards at hundreds each, that's a huge gain in margins. I'm not sure what they're charging for a Radeon Instinct MI60, but there's a whole lot of space to undercut a $9000 Tesla V100 while still being plenty profitable.
Even if they specifically meant increasing margins on consumer GPUs, just moving from HBM2 in Vega cards to GDDR6 will probably do that all on its own, even if the cost per GPU die didn't drop. And depending on the die size, the cost per GPU for that level of performance could drop a lot. If the initial Navi card is 2/3 of the price of a Vega card while costing 1/2 as much to build, that's a big increase in margins.
Right now on New Egg, you can get a Vega 64 brand new for $400 (two SKUs available) and several options of a Vega 56 for $300 or less. That wouldn't make any sense to do if they're about to offer a Navi cards with comparable performance that cost a lot more money. If they charge $500 for the top Navi card and it has performance comparable to an RTX 2070, then that would be worse on price/performance than the Vega cards that they're selling now. So I'd see that as a sign that the initial Navi card will be no more than $400.
Apparently out of the 2 5700 Navi's that will come first the one that is suppose to be around an RTX 2070 is running at 225W which is not encouraging (the other at 180W). The ASrock concepts that were shown at computex had 3 fan designs (they looked good though). Things are looking like the architecture is still not as efficient at 7nm which puts it in a bad spot against similar Turing offerings.
While the Radeon 7 was expensive to make, so had to be close the 2080 in price, Navi isn't and so doesn't have to be but the fear is it will be. Navi looks to be Polaris at 7nm.
When all this goes down soon there are also rumours of new turing models that are faster and a drop in price of the existing ones so if true, more segmentation.
Now, being a new architecture doesn't automatically mean that it's good. On the CPU side of things, Ryzen was an all new architecture, but so was Bulldozer. Rather, what it means is that how it will do in the various efficiency metrics is not publicly known right now, as previous architectures don't really give any guidance.
Certainly, 150 mm^2 would be a very optimistic die size. That's why I threw it out there as an example of something that would be an extremely favorable result for AMD. But 275 mm^2 is a pretty pessimistic estimate. There's a lot of space between those, and the difference between a 200 mm^2 die and a 250 mm^2 one matters a lot.
But we really don't know the die size yet. This is unlike third generation Ryzen, where we largely do know the die size because AMD's CEO has shown off a delidded version of it, and we know the exact size of Socket AM4.
Rumor is 255mm2
hard to compare directly with the 2070, since it’s loaded with RTX/tensor crap on top of the typical graphics architecture.
1) A picture of unknown provenance claiming that an unknown die is Navi is a lot less credible than AMD's CEO holding up a delidded CPU package and saying this is third generation Ryzen.
2) Delidding isn't really a thing on GPUs, so what you see is the full heatspreader that is intended to contact the heatsink. It's not really a clean comparison to CPU packaging.
3) If you're off slightly in dimensions of something, you can easily end up way off. Being off on your scale by 10% can mean you're off by 20% on the total area.