AMD speaks RDNA2, RDNA3, Zen3 and Zen4, announces new roadmaps

Is this the strategy that nvidia switched to years ago? Separate most of the double precision compute cores from the consumer graphics and maintain two lineups instead of one?
 
Up to 50% performance/watt gain but how much of that will translate to actual gaming gains? Doesn't seem like a 2080 Ti killer but maybe as fast or slightly faster at best. DXR performance is a big mystery. I don't think this will be competitive with Ampere at the top end and AMD will be forced to price a lot lower with low margins as usual.
 
So, AMD, what will performance look like with your hardware raytracing solution?

"we support raytracing."
???.
What date will these new much more efficient cards will become available ???.

Basically it looks like AMD is not sure what will happen in the near turn due to coronavirus.
 
Up to 50% performance/watt gain but how much of that will translate to actual gaming gains? Doesn't seem like a 2080 Ti killer but maybe as fast or slightly faster at best. DXR performance is a big mystery. I don't think this will be competitive with Ampere at the top end and AMD will be forced to price a lot lower with low margins as usual.

Kind of my take. They didn't actually talk about performance. Also, they indicated this is coming towards the end of the year. Meaning it's likely AMD could be releasing after ampere. Once again we have AMD trailing and Nvidia open to charge whatever they please.
 
I welcome AMD's realistic pronouncements as opposed to the skewed optimism of AMD 'of old'. Don't knock them for that.
 
Up to 50% performance/watt gain but how much of that will translate to actual gaming gains? Doesn't seem like a 2080 Ti killer but maybe as fast or slightly faster at best. DXR performance is a big mystery. I don't think this will be competitive with Ampere at the top end and AMD will be forced to price a lot lower with low margins as usual.
Kind of my take. They didn't actually talk about performance. Also, they indicated this is coming towards the end of the year. Meaning it's likely AMD could be releasing after ampere. Once again we have AMD trailing and Nvidia open to charge whatever they please.
That performance/watt came from note RX-325:
"RX-325:Testing done by AMD performance labs 6/1/19, using the Division 2 @ 25x14 Ultra settings. Performance may vary based on use of latest drivers, RX-325"

While it is good seeing that is an actual game being used but not much more is indicated. Was it done at the same frame rate? Power level? In other words there is a lot of wiggle room to explain less than 50% 200w RNDA compared to 200w RNDA2 gaming performance. Note the date, 6/1/19, which would seem to be an early engineering sample. Sufficient time to get production cards going I would think, so why don't we see any? My best guess is TSMC capacity which is expected to open more in June. Of course cororavirus can really affect things here.
 
Really does not tell us a whole lot yet. Should be a interesting year at least for video cards tho.
 
Yeah, no juicy nuggets, but it shows at least they are on track for what people are expecting.

50% boost perf/watt sounds pretty good though. I mean, could that be that RDNA2 is 50% faster than a 5700 XT? Cause that would be a pretty nice card and probably slightly better than a 2080 Ti.

Granted, Nvidia will have their next card out too, but AMD beating the 2080 Ti with a more reasonable price could be a good show (and at least keep them in the game). Then add in a competent ray-tracing solution and that could be a winner.

But we get hyped every time, and the past is not kind to AMD's success. But. You never know.
 
Yeah, no juicy nuggets, but it shows at least they are on track for what people are expecting.

50% boost perf/watt sounds pretty good though. I mean, could that be that RDNA2 is 50% faster than a 5700 XT? Cause that would be a pretty nice card and probably slightly better than a 2080 Ti.

Granted, Nvidia will have their next card out too, but AMD beating the 2080 Ti with a more reasonable price could be a good show (and at least keep them in the game). Then add in a competent ray-tracing solution and that could be a winner.

But we get hyped every time, and the past is not kind to AMD's success. But. You never know.

If AMD can get to 2080Ti levels of rasterized performance with better Ray Tracing and at a reasonable price, then I'd be happy with that.
 
Well, if they are not BS on the numbers (and also, we don't know the power levels, 50% perf/watt is not the same as 50% better perf) that would put them in 2080 Ti territory. So that is a real possibility.

Then on ray-tracing, I think they might surprise us. Well, we know both next-gen consoles have ray-tracing, and AMD must have found a way to make it work, maybe more efficient than Nvidia has so far. So this could be a big win here.

And then price. I don't know what they are targeting, but $1,200 or even $1,000 I think would be a miss. I would guess $799 for the top end, and then moving down from there. I hope they can pull it off.
 
And then price. I don't know what they are targeting, but $1,200 or even $1,000 I think would be a miss. I would guess $799 for the top end, and then moving down from there. I hope they can pull it off.

I would love to see a 72CU card with 12GB of HBM2e for US$899; oh, and make it a LC (Liquid Cooled) variant, thanks!
 
A lot of speculation about if AMD can reach 2080ti territory when Big Navi finally comes out.

I feel the real question should be can AMD afford not to be in 2080ti territory by then?
 
I would expect a Titan level card, like what the Vega Frontier Edition attempted to do, hopefully with a better name (Frontier -> exploring the wilderness?), maybe the only consumer level card with HBM2e but one can hope, >$1000, if with 32gb of HBM2e, maybe approaching $2000. The higher end professional cards and MI cards I would definitely expect HBM2e.

I just see most of the gaming cards except maybe a lower version of a HBM2e card would be using DDR6, what Nvidia did with the 2080Ti was utterly ridiculous. With those prices I would expect most will just go with the next generation of the console which the XBox X should be faster than a 2080 Super. Which is a strong consideration, will AMD charge more for a card that performs less or equal to the next generation console about the same cost? Or performs slightly better? I see AMD having a good opportunity to pick up some large user numbers if they can price right, hit the manufacturing numbers needed with expected performance numbers. AMD is also aiming for a 50% profit margin up from 45%. So they are not necessarily looking at having historically lower prices but will compete with Nvidia price wise that will make sense.

I take the 50% perf/watt improvement using the least efficient Navi GPU -> Navi 5700XT, at 225w RNDA2 will perform 50% better at 225w, if that power envelope is expanded to 250w (if it scales linearly) 2.5/2.25 x 1.5 = 68% faster than a 5700XT. Any headroom beyond that will just be a plus. If AMD VRS is more efficient than Nvidia and more automatic or if newer titles on consoles that are available to the PC pretty much all use it, that will stretch the performance improvement. Add in improved Radeon Boost that work with more games and particulary DX12 and Vulkan games will give a usable performance improvement which can be rather large when it counts. If Microsoft/Sony/AMD came up with efficient RT methods and software tools (which I would expect) RT rendering enhancements should be more widely used.

I see at this time Nvidia investment into Tech particularly AI may just maintain their lead and it could be huge. I expect Nvidia to at least have a 30% low side and more like 50% performance improvement. Combined that with working tech that can be easily be incorporated, DLSS (particularly if more automatic with driver updates for more and more games and developer support and updates). I expect Nvidia to expand upon RT tech and more notably new ways of doing things, most likely using their AI advancements, DLSS is only the first now successful implementation, I expect that to be much expanded and possibly blow our socks off. If so, then I don't expect prices be lowered that much unless AMD start kicking their ass in sells.

I see this year, depending upon world events as well, could be the most interesting, best year ever to be a reviewer/tester of gaming cards as well as professional cards. Maybe a year where everything you own you might just want to throw away.
 
A lot of speculation about if AMD can reach 2080ti territory when Big Navi finally comes out.

I feel the real question should be can AMD afford not to be in 2080ti territory by then?

This happened last time.

AMD should not be looking to compete with the 2080 Ti, it should be looking to compete with the 3080 Ti.
 
This happened last time.

AMD should not be looking to compete with the 2080 Ti, it should be looking to compete with the 3080 Ti.

That's highly unlikely to happen. That was my reasons for only mentioning the 2080Ti level. I'd be very happy though to eat my words and say I was wrong. But history says different.
 
That's highly unlikely to happen. That was my reasons for only mentioning the 2080Ti level. I'd be very happy though to eat my words and say I was wrong. But history says different.
It seems AMD is trying to not hype themselves up like they did with bulldozer. Just realistic goals, and going out and doing them.
 
I would love to see a 72CU card with 12GB of HBM2e for US$899; oh, and make it a LC (Liquid Cooled) variant, thanks!
at 899 it would have to be GDDR6.. HBM2e is expensive as fk and AMD likely learned from their mistake with vega/vega 7nm that HBM has no purpose for gaming. now that they've split their consumer card line up from their enterprise lineup the new instinct cards will probably have HBM2e and RDNA 2 cards will probably have 14gbps GDDR6 with hopefully a better memory controller to allow higher memory clock speeds.

This happened last time.

AMD should not be looking to compete with the 2080 Ti, it should be looking to compete with the 3080 Ti.

consider it more of a proof of concept.. if RDNA 2 actually works and is competitive then they know they're on the right track which makes it much easier to release RDNA 3. no point in swinging for the fences based on rumors with zero context to them.. they just keep throwing around the 50% figure but still have not said whether it's ray tracing or raster performance, for all we know the 3080ti is 50% faster at ray tracing but performs exactly the same as the 2080ti in games that don't support ray tracing.
 
Last edited:
I would love to see a 72CU card with 12GB of HBM2e for US$899; oh, and make it a LC (Liquid Cooled) variant, thanks!

at 899 it would have to be GDDR6.. HBM2e is expensive as fk and AMD likely learned from their mistake with vega/vega 7nm that HBM has no purpose for gaming. now that they've split their consumer card line up from their enterprise lineup the new instinct cards will probably have HBM2e and RDNA 2 cards will probably have 14gbps GDDR6 with hopefully a better memory controller to allow higher memory clock speeds.

Some of us want more from our GPUs than just gaming; I run Cinema 4D, Substance, & Resolve; all can benefit from the added bandwidth HBM2e would provide...?
 
I just want the RDNA 2.0 version of the Vega Frontier Edition LC...
 
  • Like
Reactions: noko
like this
at 899 it would have to be GDDR6.. HBM2e is expensive as fk and AMD likely learned from their mistake with vega/vega 7nm that HBM has no purpose for gaming. now that they've split their consumer card line up from their enterprise lineup the new instinct cards will probably have HBM2e and RDNA 2 cards will probably have 14gbps GDDR6 with hopefully a better memory controller to allow higher memory clock speeds.



consider it more of a proof of concept.. if RDNA 2 actually works and is competitive then they know they're on the right track which makes it much easier to release RDNA 3. no point in swinging for the fences based on rumors with zero context to them.. they just keep throwing around the 50% figure but still have not said whether it's ray tracing or raster performance, for all we know the 3080ti is 50% faster at ray tracing but performs exactly the same as the 2080ti in games that don't support ray tracing.
The AMD 50% perf/watt was based off a real game, Div 2 per note on slide. Since comparison was RNDA vs RNDA2 that would not be with ray tracing anyways.
 
Back
Top