As far as I know, they won't delay the launch. But it will have an affect on the stock availability, and quantities will remain low for the first 6 months of its lifecycle. The more interesting tidbit is the pricing. We know Microsoft is huge, has cash reserves to take the hit, but not Sony.
Sony was planning on making a $400 console, and sell it for $500. So there's little ($100) to ($0) no profit margin there. However, it seems now the costs have increased to $550. That means they need to take a small ($50) to ($100) medium loss per unit. Or they could sell it for $600, and maintain their minimal profit margin, but that seems unlikely based on sour-history with the PS3 launch.
The ideal scenario is to delay the launch by 6 months, since the PS4 and PS4 Pro are young enough, powerful enough, and popular enough to allow this. Though that won't happen as Sony Executives do not want to miss the Holiday Period, unlike Nintendo which doesn't care. Maybe they can announce the PS5 a week before release, and in the announcement state that its RRP is $499.... but for the first 6-months the consoles will be sold as "Founders Edition" with a premium $600 price. It think that's the best approach.
As long as they do not use these circumstances to downgrade the console, I'm all for it. Since price can change, but hardware specs are locked-in for over 6 years. We have been expecting the console to ship with specs similar to a Midrange Gaming PC: r7-3700X, RX 5700, 32GB RAM (shared GDDR6), and 1TB SSD (or instead a soldered 256GB SSD + 1TB HDD removable).
Your're confusing a lot of information. The 128 CUs/8192 ALUs belong to Arcturus, an HPC-only GPU that is based on an adjusted Vega architecture. And the 17% IPC were a rumor regarding Zen 3.
So what I've heard is most likely 80 CU, and if they're able to do that with only about 60% of the power from RDNA1 that should put them within reach of those 40% better than the 2080 TI numbers we saw from that VR benchmark a while back, without requiring three eight pin power connectors.
depends on FLAT 300w TDP/TBP or "if they are smart" can be "max of" but not will always use. If they take Zen2 into account of the design, 65/105w competing neck and neck with "others" that claim same but in reality chew into the 180+ ACTUAL watt range (when not idling)
I always wondered about this, yes bunch of transistors take bunch of power to do their thing, CPU have long been able to "settle down" per ms to not chew power galore, GPU with their many thousands of "small not so smart" shaders (whatever you prefer to call them) have not been able to do this?
seems to me, that is a ripe feature to be targeting eg. loads of performance when you need it, while sipping power when you really do not..Nv managed to do something like this (with use of fancy chip/VREG) for GTX 1xxx generation, not so much for RTX 2xxx, AMD "for the most part" was pretty decent overall in their rated TDP vs actual TDP/TBP ratings, but for some reason have not done much to "harness" power usage when not truly needed, beyond limited user control (through things like afterburner..when you are/were able to use it) to downclock and/or downvolt without forcing a bios flash.
anyways.....
raw 300w, yes sir, that would be a toasty card, thankfully, it is "only" 7nm, vs say if it were 28, 45 etc nm designs, that would make the GTX 480 look like it was a freezer pop (^.^)
Top end cards are usually toasty. On paper, it looks really good. But again, AMD is still playing catch up with Nvidia. With RDNA2, it seems they are catching up (on paper) with Turing. But we do have to remember we are comparing 7nm chip from AMD with a 14nm class GPU from Nvidia. So the true competition will only heat up later this year.
Since RDNA2 is only coming out in late 2020, it will directly compete against RTX 3000 cards, which will be out by then. In short, even if these numbers translate to the high-end quite well, the crown seems to remain out of reach.
RDNA 2 will be matched with Ampere. RDNA already competes with Turing, it just lacks ray-tracing. (5700XT is only 5% slower than 2070Super. At much lower price).
I don't think it scales linearly. So there will be good 35%-45% improvement on the low-power side, then a great 45%-55% improvement in the midrange segment, then a decent 30%-40% improvement on the high-power end.
Overall, I think RDNA2 will slightly surpass Turing on a "IPC" comparison, but it will be doing so whilst using 7nm versus their 10nm wafers. If Nvidia were to port Turing directly to 7nm, they would have a noticeable advantage, at least when thinking on a "IPC" style format. Yet, we know Nvidia will make tweaks and get even further gains. So I think the RTX-30 series is going to enjoy a healthy lead in the market, just like Nvidia did in their GTX-10 series period.
I guess the good news is that progress is happening. But it's going to become even more expensive to get into PC Gaming, whereas the consoles might actually be decently competitive (ie Recall the launch of the PS2 or the Xbox 360).
Turing is 12nm. Moving to 10nm or 7nm would give them parity or an advantage over RDNA2 by dropping power, boosting clocks, or allowing for more compute units.
Ampere is purported to also be a 50% improvement over Turing thanks to 7nm, allowing them to boost clock, cores, and lower power.
People keep doing funny maths here. Here's how I see it: In IPC terms, RDNA on 7nm is already at relative parity with Turing on 12nm. In performance per watt terms, when you factor in that node difference it's an obvious loser. Ampere is supposed to gain by at least 50% over Turing - that's including the 7nm shrink. RDNA 2 is alleged to have a similar improvement, without any significant shrink. They're also going to finally be releasing products based on larger dies.
The end result should be real competition at the high end again, albeit likely with Nvidia still holding the performance crown at the bleeding edge. I don't care about £1000+ cards, so this sounds like the proper competition I've been waiting for since Maxwell overturned the apple cart.
Will be interesting to see, if the close relationship between the console and PC GPUs will give AMD an advantage with the optimization of games for their platform.
AMD already enjoys optimization for its own architectures with respect to Nvidia. In fact you can't see any new Nvidia feature supported by the developer, and I'm speaking about VSR, mesh shading, packet math, improved geometry handling, voxel accelerations. We are stuck to AC tricks to improve GCN throughput with low polygons count, pumped up textures (as AMD enjoys bigger memory bandwidth) and nothing more. Hey, but those boring improvements work better on GCN than on Nvidia HW.. WOW, viva la revolution... viva el progeso!
Nvidia co-designs many many games with developers. So they are indeed optimized. Nvidia has software engineers that personally help developers optimize for the game. And I'm sure all the implementations of "Variable shading" will have a similar API for developers to target, and the drivers for each card (Intel, Nvidia, AMD) will do the corresponding stuff.
AMD also provides staff members to assist major game developers, but that's not all. They also provide full source code to their accelerated libraries. See: https://gpuopen.com/games-cgi/
I believe both Nvidia and AMD also provide some form of financial support for major game titles, in exchange for optimizations targeting their respective hardware, though I don't know the specifics.
That was three years ago, and AMD still hasn't released anything that comes close to backing that up. I'm all for advancement no matter which company is doing it, but let's say AMD's GPU claims have earned them a wait until we see it status at best.
In realistic terms their existing parts are capable of excellent efficiency, they simply clock them out of their sweet spot chasing nVidia. When these parts hit, if performance isn't lining up the way they wanted I fully expect they will push the clocks and kill the efficiency once more.
In realistic terms their existing parts are capable of excellent efficiency
That's completely wrong. Power efficiency is measured as the work done / energy used. If you clock your big beast at a low clock you get fantastic efficiency, but that is useless if you then have the performance of the smallest chip made by the competitors that consumes as much but costs 1/10th.
The first AMD's problem since the introduction of GCN has been power efficiency, which goes together to computing efficiency (work done / transistor needed). Both these indexes were extremely low for AMD (with respect to the competition) and this has opened a widening gap since RDNA. To compete in term of performances, AMD has to use bigger and much power hungry chips. To limit the dimension gap between its chips and the competitor's similar performer ones they had to pump up the clocks, worsening the power efficiency (but avoiding to go into negative gain margins).
With Turing Nvidia passed to be the one using the bigger dies, but kept power efficiency crown. With RDNA AMD increased their computing efficiency, that normalized with respect to the production node used, it is now equal to Pascal. This at a cost, however: they worsened a lot their computation efficiency, by almost doubling the number of transistors while not adding any new feature (but packet math) with respect to Polaris. A RDNA chip is almost as big as Turing one (in number of transistors) but features Pascal capabilities.
With RDNA2 AMD is now in a different scenery with respect to RDNA vs Turing. 1. The PP, even if using 7Np or 7N+ is not going to give much improvements in terms of density and very limited in terms of power efficiency. 2. A bunch of new features have to be added to close the gap with the competition, from RT HW acceleration, possible AI acceleration (for filtering and new algorithms that could be exploited for new advanced features), VSR and geometry engine improvements (this has been a serious problem for AMD since Terascale, now widened by the implementation of dynamic mesh shading by Nvidia). This is going to cost transistor. 3. To reach the top of the list in performance, AMD needs to add much more computing resources, which means many more transistors.
Now, seen the number of transistors that they are using with RDNA to be at Pascal level, I can't imagine how big (in term of transistor counts) the new GPUs needs to be to be competitive with Nvidia's biggest beasts. And how they can maintain power efficiency, seen they now can't go wide as they did with RDNA. It's a real tough task what is trying to do AMD.
The good news is that they finally have decided to split their GPUs architectures: one for gaming and one for computing. As it previously was, they were good in neither of the two jobs.
If you clock your big beast at a low clock you get fantastic efficiency, but that is useless if you then have the performance of the smallest chip made by the competitors that consumes as much but costs 1/10th. --Navi's efficiency is around 1500mhz. That's not 1/10th the speed of the competition. A 5700XT at 1500mhz would be around a 2060 super, and consume way less power.
With RDNA2 AMD is now in a different scenery with respect to RDNA vs Turing. 1. The PP, even if using 7Np or 7N+ is not going to give much improvements in terms of density and very limited in terms of power efficiency. --RDNA is just the Vega die shrink with a new scheduler. So Vega 7nm was 40mT/mm2. RDNA is 40.5mT/mm2. The new scheduler allowed AMD to target Wave 16, 32, and 64. Vega's issue, was games weren't coded for their wave setup, meaning the best programmed games only hit about 60-70% of the use of the cores, while for computation they were amazing. And with a node shrink, RDNA2 will hit 60-80mT/mm2. RDNA2 is a ground-up 7nm design, RDNA is just vega with a new scheduler.
2. A bunch of new features have to be added to close the gap with the competition, from RT HW acceleration, possible AI acceleration (for filtering and new algorithms that could be exploited for new advanced features), VSR and geometry engine improvements (this has been a serious problem for AMD since Terascale, now widened by the implementation of dynamic mesh shading by Nvidia). This is going to cost transistor. --RDNA2 already has hybrid ray-tracing, with a bvr path to accelerate it, similar to Nvidia. Why does it need AI cores? It's not for computation, it's for gaming. VSR is already implemented if you watched their presentation. Microsoft also has a patented version. Ray-tracing itself takes less than 10% of the transistors of a die. Your very mis-informed.
3. To reach the top of the list in performance, AMD needs to add much more computing resources, which means many more transistors. --AMD can already reach the top with RDNA, if they made an 80CU die. But its only 7 months until RDNA2 comes out, so making it now, the masks costs wouldn't pay off before the next-gen, and the investment would be negative. RDNA2 will be around for 18months and it will recoup the investment and possibly take the performance crown.
Good thing your not an analyst. You're not very good at analyzing. But your arguments sound good if you don't look into it.
> Why does it need AI cores? It's not for computation, it's for gaming.
Nvidia relies on its Tensor Cores for good denoising performance with Global Illumination. Except, the performance of GI in Raytraced games is still so bad that most don't use it.
Yeah, you should really judge GPUs' efficiency at their stock specs. And if those specs push a design past its efficiency sweet spot, then too bad, cuz that's how the manufacturer chose to clock it.
It's a moot exercise to play the "what if it had been clocked x% lower" game. Perhaps of academic interest, but it's of virtually no consequence if that's not how they actually shipped it.
It's of interest when you're comparing the efficiency of an architecture, rather than the performance/value/etc. of a specific product. That is pretty academic, but then what are we here for if not to be nerdy about these things? :)
> they worsened a lot their computation efficiency, by almost doubling the number of transistors while not adding any new feature (but packet math) with respect to Polaris.
It's not accurate to say it added no new features. RDNA added: * Hardware GCN compatibility * 2x scalar throughput * primitive shaders * PCIe 4.0 * GDDR6 * Image sharpening * Improved DSBR
I do agree that it would probably do well to drop GCN support. As improvements in area-efficiency tend to provide benefits in energy-efficiency, and we know RDNA 2 will have significant amounts of the latter, I'm also expecting it to feature the former.
"That's completely wrong. Power efficiency is measured as the work done / energy used. If you clock your big beast at a low clock you get fantastic efficiency, but that is useless if you then have the performance of the smallest chip made by the competitors that consumes as much but costs 1/10th."
You're exaggerating here. Radeon VII is a great example of a design that stinks at its default clocks and voltages, but when the clocks are scaled back a little the voltage scales back a *lot*. The end result is something that's actually quite competitive both in terms of PPW and overall performance, so not "completely wrong" at all. See also: how well Vega integrated graphics compete with Intel's iGPU in thermally constrained situations, even with a node deficit.
"Now, seen the number of transistors that they are using with RDNA to be at Pascal level, I can't imagine how big (in term of transistor counts) the new GPUs needs to be to be competitive with Nvidia's biggest beasts."
When they're talking about a 50% PPW increase - and their previous problems with performance scaling were high temperatures resulting from hitting a voltage wall - then they shouldn't need to rely entirely on making their designs larger just to compete. Based on recent history, they'll probably end up with slightly smaller designs that can't reach Nvidia's top performance levels but compete well in their class.
"part of that advantage comes from an atypical advantage in manufacturing nodes"
Actually, I would more say that its more of a return to form for AMD, because until 16/14nm, AMD was usually first on a new (TSMC) node for GPU's. Global foundries might have been the reason they weren't first with 16/14nm.
That fourth slide (AMD RDNA 2 PERF/WATT IMPROVEMENT) is a bit of a mess. Why are CPU and GPU on the same graph? Can they really maintain that their GPUs are tracking teh industry trend for perf/watt? The choice of colors for fonts and colors for graph lines also appears to imply things, but if you consider closely does not.
they're showing how AMD buckled both the trend on the CPU side, and they're doing it with the GPU side also.
Makes perfect sense really. Their CPU division is doing amazing, and really rose the company back into the hearts of enthusiast, and to see them say that RDNA2 should repeat that success is a good thing to see. But we'll have to wait until the cards come out, of course.
They just did a graphics trick. The industry trend line passes from their Zen2 and RDNA2 points in the graph, not through Zen1-RDNA1. From Zen1 to Zen2 they just followed the industry trend = using a new PP to increase efficiency. Form RDNA to RDNA 2 they are doing the same reaching what the competition was 3 years ago. So, beside Zen that is in line with industry trend, RDNA is a catch up game with what are the perf/performance of the competition. Unfortunately they forgot to trace where the competition is. That is a point near their RDNA2 in the graph but using an older PP. When switching to the new PP, the industry trend line will go steeper and they will be again a step behind.
The step above the trend line is the one from Polaris to RDNA, where they improve perf/W a lot by just spending all their transistor budget for the new features (that they do not have).
RDNA already caught up to Pascal, and then some, matching Turing in practice because of the node advantage. RDNA 2 should perform *at least* like a Turing shrink to 7nm would. That's not exactly "just catching up to 3 years ago".
Perf/watt increases are great, but for a lot of consumers, it's perf/$ that drives purchasing decisions - and both AMD and Nvidia haven't been great on that front. AMD's 4-year-old RX at $240 (launch) still dukes it out with the ~$220 RX 5500 XT if you push clocks to make your 480 a quasi-580 - and don't tell me that "overclocking doesn't count," because that's EXACTLY what AMD did with its own 5600 XT launch.
It's great that AMD is offering products that move up the stack, which forced Nvidia to offer more performance via its "Super" semi-update, but so far, we haven't really seen the competition drive prices down, or offer consumers meaningful performance increases over a GPU they bought years ago if they're willing to spend the same amount.
The perf/$ only comes in play if the perf/watt is good enough. Meaning there is nothing AMD could do with Vega to make a 2080 TI consumer card and the thermals be able to be cooled with anything short of a custom water loop. This is to say if the perf/watt is good then the cost of the cards in each segment will come into play. Today AMD can't even step into the high end.
perf/W affects perf/$ because it lets you reach higher performance levels with a smaller die (which you can clock higher). So, improving efficiency can also create room to lower prices.
RDNA2 for sure will be slower than the RTX 3000. The best they can do is beat the 2080 Ti by a few or 10+ percent and sell it for less. I hope that's the case and if not RTX 3000 will be even more expensive than the 2080 Ti 1200+$. I have an OCed 2080 Ti and can play any game at over 60FPS at 4K so no need any upgrade soon. GPUs nowdays are relatively way more powerful than before. When Crysis came out even the most expensive cards were struggling running it at Ultra 1080p. I don't see need for faster GPUs now unless 4K120+Hz monitors become affordable which isn't the case.
"The best they can do is beat the 2080 Ti by a few or 10+ percent and sell it for less." Citation needed.
"GPUs nowdays are relatively way more powerful than before." Additional citations needed. Crysis is a terrible example to generalise from because it was specifically built to utilise GPUs more powerful than those in existence when it released. Games designers don't really do that anymore because it doesn't make any sense.
Zen 3 looks impressive with less game latency, improved IPC and improved clocks. I will upgrade my 1800X to a 4800X when it comes out. That will be a huge upgrade.
If the 17.5 TFLOPS number is true, this may actually be an Nvidia killer. Well, okay, Nvidia isn't going anywhere soon, but they will need to realize about a 34% increase in TFLOPS over a 2080 Ti to match that. TFLOPS are a rough measure, but AMD is currently sitting at 9.75 TFLOPS w/ RX 5700 XT, which suggests they will be getting +80% RX5700 XT performance with their top end Navi 2x card. It will be interesting to see which company has better RT implementation as well. That could be a big deal: with both companies offering RT and RT leaving its infancy, the RT value will matter.
I reckon Nvidia can easily exceed that level of performance - the question is more about how they'll choose to do it, and what the cost of that would be. If AMD can do it with a smaller die and no HBM then we might finally get solid price competition again.
I found some code relating to Navi 21 and Van Gogh (Renoir's replacement?) in Adrenalin version 20.1.3. Wasn't much and mostly related to power delivery system, but there's definitely proof of their development in the drivers.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
46 Comments
Back to Article
sorten - Thursday, March 5, 2020 - link
Whoa, I wasn't even expecting AMD news today. Sounds awesome, and this year even!Thanks for the coverage.
milkywayer - Thursday, March 5, 2020 - link
I wonder if Covid19 is going to impact the ps5 and nexbox launch. I mean it will but wondering how much.Kangal - Friday, March 6, 2020 - link
As far as I know, they won't delay the launch.But it will have an affect on the stock availability, and quantities will remain low for the first 6 months of its lifecycle. The more interesting tidbit is the pricing. We know Microsoft is huge, has cash reserves to take the hit, but not Sony.
Sony was planning on making a $400 console, and sell it for $500. So there's little ($100) to ($0) no profit margin there. However, it seems now the costs have increased to $550. That means they need to take a small ($50) to ($100) medium loss per unit. Or they could sell it for $600, and maintain their minimal profit margin, but that seems unlikely based on sour-history with the PS3 launch.
The ideal scenario is to delay the launch by 6 months, since the PS4 and PS4 Pro are young enough, powerful enough, and popular enough to allow this. Though that won't happen as Sony Executives do not want to miss the Holiday Period, unlike Nintendo which doesn't care. Maybe they can announce the PS5 a week before release, and in the announcement state that its RRP is $499.... but for the first 6-months the consoles will be sold as "Founders Edition" with a premium $600 price. It think that's the best approach.
As long as they do not use these circumstances to downgrade the console, I'm all for it. Since price can change, but hardware specs are locked-in for over 6 years. We have been expecting the console to ship with specs similar to a Midrange Gaming PC: r7-3700X, RX 5700, 32GB RAM (shared GDDR6), and 1TB SSD (or instead a soldered 256GB SSD + 1TB HDD removable).
Crazyeyeskillah - Thursday, March 5, 2020 - link
I've never seen this much extremely specified info dropped at a fin an day.If Big Navi didn't have people's ears pinned, this will certainly get the job done.
8k compute units, with something like 17% ipc gain, increased clocks, and the like - we're looking at a serious powerhouse.
tuxRoller - Thursday, March 5, 2020 - link
Where did you come across these numbers?Novacius - Thursday, March 5, 2020 - link
Your're confusing a lot of information. The 128 CUs/8192 ALUs belong to Arcturus, an HPC-only GPU that is based on an adjusted Vega architecture. And the 17% IPC were a rumor regarding Zen 3.HarryVoyager - Thursday, March 5, 2020 - link
So what I've heard is most likely 80 CU, and if they're able to do that with only about 60% of the power from RDNA1 that should put them within reach of those 40% better than the 2080 TI numbers we saw from that VR benchmark a while back, without requiring three eight pin power connectors.Novacius - Thursday, March 5, 2020 - link
Yeah. 50% better performance per watt would enable a card with 80 CUs/twice the performance of the 5700XT at around 300W.Flunk - Thursday, March 5, 2020 - link
300w TDP would make for a toasty card.Dragonstongue - Thursday, March 5, 2020 - link
depends on FLAT 300w TDP/TBP or "if they are smart" can be "max of" but not will always use. If they take Zen2 into account of the design, 65/105w competing neck and neck with "others" that claim same but in reality chew into the 180+ ACTUAL watt range (when not idling)I always wondered about this, yes bunch of transistors take bunch of power to do their thing, CPU have long been able to "settle down" per ms to not chew power galore, GPU with their many thousands of "small not so smart" shaders (whatever you prefer to call them) have not been able to do this?
seems to me, that is a ripe feature to be targeting eg. loads of performance when you need it, while sipping power when you really do not..Nv managed to do something like this (with use of fancy chip/VREG) for GTX 1xxx generation, not so much for RTX 2xxx, AMD "for the most part" was pretty decent overall in their rated TDP vs actual TDP/TBP ratings, but for some reason have not done much to "harness" power usage when not truly needed, beyond limited user control (through things like afterburner..when you are/were able to use it) to downclock and/or downvolt without forcing a bios flash.
anyways.....
raw 300w, yes sir, that would be a toasty card, thankfully, it is "only" 7nm, vs say if it were 28, 45 etc nm designs, that would make the GTX 480 look like it was a freezer pop (^.^)
watzupken - Friday, March 6, 2020 - link
Top end cards are usually toasty. On paper, it looks really good. But again, AMD is still playing catch up with Nvidia. With RDNA2, it seems they are catching up (on paper) with Turing. But we do have to remember we are comparing 7nm chip from AMD with a 14nm class GPU from Nvidia. So the true competition will only heat up later this year.nevcairiel - Friday, March 6, 2020 - link
Since RDNA2 is only coming out in late 2020, it will directly compete against RTX 3000 cards, which will be out by then. In short, even if these numbers translate to the high-end quite well, the crown seems to remain out of reach.Fataliity - Friday, March 6, 2020 - link
RDNA 2 will be matched with Ampere. RDNA already competes with Turing, it just lacks ray-tracing. (5700XT is only 5% slower than 2070Super. At much lower price).Cellar Door - Friday, March 6, 2020 - link
50% perf per watt increase, allows for 2080Ti(+20%) perf - if AMD sticks to a reasonable power envelope(TDP).Kangal - Friday, March 6, 2020 - link
I don't think it scales linearly.So there will be good 35%-45% improvement on the low-power side, then a great 45%-55% improvement in the midrange segment, then a decent 30%-40% improvement on the high-power end.
Overall, I think RDNA2 will slightly surpass Turing on a "IPC" comparison, but it will be doing so whilst using 7nm versus their 10nm wafers. If Nvidia were to port Turing directly to 7nm, they would have a noticeable advantage, at least when thinking on a "IPC" style format. Yet, we know Nvidia will make tweaks and get even further gains. So I think the RTX-30 series is going to enjoy a healthy lead in the market, just like Nvidia did in their GTX-10 series period.
I guess the good news is that progress is happening. But it's going to become even more expensive to get into PC Gaming, whereas the consoles might actually be decently competitive (ie Recall the launch of the PS2 or the Xbox 360).
michael2k - Friday, March 6, 2020 - link
Turing is 12nm. Moving to 10nm or 7nm would give them parity or an advantage over RDNA2 by dropping power, boosting clocks, or allowing for more compute units.Ampere is purported to also be a 50% improvement over Turing thanks to 7nm, allowing them to boost clock, cores, and lower power.
Spunjji - Friday, March 13, 2020 - link
People keep doing funny maths here. Here's how I see it:In IPC terms, RDNA on 7nm is already at relative parity with Turing on 12nm. In performance per watt terms, when you factor in that node difference it's an obvious loser.
Ampere is supposed to gain by at least 50% over Turing - that's including the 7nm shrink.
RDNA 2 is alleged to have a similar improvement, without any significant shrink. They're also going to finally be releasing products based on larger dies.
The end result should be real competition at the high end again, albeit likely with Nvidia still holding the performance crown at the bleeding edge. I don't care about £1000+ cards, so this sounds like the proper competition I've been waiting for since Maxwell overturned the apple cart.
Nozuka - Friday, March 6, 2020 - link
Will be interesting to see, if the close relationship between the console and PC GPUs will give AMD an advantage with the optimization of games for their platform.CiccioB - Friday, March 6, 2020 - link
AMD already enjoys optimization for its own architectures with respect to Nvidia.In fact you can't see any new Nvidia feature supported by the developer, and I'm speaking about VSR, mesh shading, packet math, improved geometry handling, voxel accelerations.
We are stuck to AC tricks to improve GCN throughput with low polygons count, pumped up textures (as AMD enjoys bigger memory bandwidth) and nothing more.
Hey, but those boring improvements work better on GCN than on Nvidia HW.. WOW, viva la revolution... viva el progeso!
Fataliity - Friday, March 6, 2020 - link
Nvidia co-designs many many games with developers. So they are indeed optimized. Nvidia has software engineers that personally help developers optimize for the game. And I'm sure all the implementations of "Variable shading" will have a similar API for developers to target, and the drivers for each card (Intel, Nvidia, AMD) will do the corresponding stuff.mode_13h - Sunday, March 8, 2020 - link
AMD also provides staff members to assist major game developers, but that's not all. They also provide full source code to their accelerated libraries. See: https://gpuopen.com/games-cgi/I believe both Nvidia and AMD also provide some form of financial support for major game titles, in exchange for optimizations targeting their respective hardware, though I don't know the specifics.
SydneyBlue120d - Friday, March 6, 2020 - link
Any updates on the Samsung mobile project?BenSkywalker - Friday, March 6, 2020 - link
"Poor Volta"That was three years ago, and AMD still hasn't released anything that comes close to backing that up. I'm all for advancement no matter which company is doing it, but let's say AMD's GPU claims have earned them a wait until we see it status at best.
In realistic terms their existing parts are capable of excellent efficiency, they simply clock them out of their sweet spot chasing nVidia. When these parts hit, if performance isn't lining up the way they wanted I fully expect they will push the clocks and kill the efficiency once more.
CiccioB - Friday, March 6, 2020 - link
That's completely wrong.
Power efficiency is measured as the work done / energy used.
If you clock your big beast at a low clock you get fantastic efficiency, but that is useless if you then have the performance of the smallest chip made by the competitors that consumes as much but costs 1/10th.
The first AMD's problem since the introduction of GCN has been power efficiency, which goes together to computing efficiency (work done / transistor needed).
Both these indexes were extremely low for AMD (with respect to the competition) and this has opened a widening gap since RDNA.
To compete in term of performances, AMD has to use bigger and much power hungry chips. To limit the dimension gap between its chips and the competitor's similar performer ones they had to pump up the clocks, worsening the power efficiency (but avoiding to go into negative gain margins).
With Turing Nvidia passed to be the one using the bigger dies, but kept power efficiency crown.
With RDNA AMD increased their computing efficiency, that normalized with respect to the production node used, it is now equal to Pascal. This at a cost, however: they worsened a lot their computation efficiency, by almost doubling the number of transistors while not adding any new feature (but packet math) with respect to Polaris.
A RDNA chip is almost as big as Turing one (in number of transistors) but features Pascal capabilities.
With RDNA2 AMD is now in a different scenery with respect to RDNA vs Turing.
1. The PP, even if using 7Np or 7N+ is not going to give much improvements in terms of density and very limited in terms of power efficiency.
2. A bunch of new features have to be added to close the gap with the competition, from RT HW acceleration, possible AI acceleration (for filtering and new algorithms that could be exploited for new advanced features), VSR and geometry engine improvements (this has been a serious problem for AMD since Terascale, now widened by the implementation of dynamic mesh shading by Nvidia). This is going to cost transistor.
3. To reach the top of the list in performance, AMD needs to add much more computing resources, which means many more transistors.
Now, seen the number of transistors that they are using with RDNA to be at Pascal level, I can't imagine how big (in term of transistor counts) the new GPUs needs to be to be competitive with Nvidia's biggest beasts. And how they can maintain power efficiency, seen they now can't go wide as they did with RDNA.
It's a real tough task what is trying to do AMD.
The good news is that they finally have decided to split their GPUs architectures: one for gaming and one for computing. As it previously was, they were good in neither of the two jobs.
Fataliity - Friday, March 6, 2020 - link
If you clock your big beast at a low clock you get fantastic efficiency, but that is useless if you then have the performance of the smallest chip made by the competitors that consumes as much but costs 1/10th.--Navi's efficiency is around 1500mhz. That's not 1/10th the speed of the competition. A 5700XT at 1500mhz would be around a 2060 super, and consume way less power.
With RDNA2 AMD is now in a different scenery with respect to RDNA vs Turing.
1. The PP, even if using 7Np or 7N+ is not going to give much improvements in terms of density and very limited in terms of power efficiency.
--RDNA is just the Vega die shrink with a new scheduler. So Vega 7nm was 40mT/mm2. RDNA is 40.5mT/mm2. The new scheduler allowed AMD to target Wave 16, 32, and 64. Vega's issue, was games weren't coded for their wave setup, meaning the best programmed games only hit about 60-70% of the use of the cores, while for computation they were amazing. And with a node shrink, RDNA2 will hit 60-80mT/mm2. RDNA2 is a ground-up 7nm design, RDNA is just vega with a new scheduler.
2. A bunch of new features have to be added to close the gap with the competition, from RT HW acceleration, possible AI acceleration (for filtering and new algorithms that could be exploited for new advanced features), VSR and geometry engine improvements (this has been a serious problem for AMD since Terascale, now widened by the implementation of dynamic mesh shading by Nvidia). This is going to cost transistor.
--RDNA2 already has hybrid ray-tracing, with a bvr path to accelerate it, similar to Nvidia. Why does it need AI cores? It's not for computation, it's for gaming. VSR is already implemented if you watched their presentation. Microsoft also has a patented version. Ray-tracing itself takes less than 10% of the transistors of a die. Your very mis-informed.
3. To reach the top of the list in performance, AMD needs to add much more computing resources, which means many more transistors.
--AMD can already reach the top with RDNA, if they made an 80CU die. But its only 7 months until RDNA2 comes out, so making it now, the masks costs wouldn't pay off before the next-gen, and the investment would be negative. RDNA2 will be around for 18months and it will recoup the investment and possibly take the performance crown.
Good thing your not an analyst. You're not very good at analyzing. But your arguments sound good if you don't look into it.
mode_13h - Sunday, March 8, 2020 - link
> RDNA is just the Vega die shrink with a new scheduler.Definitely not. Wave32 is much deeper than "a new scheduler". See my other post for some of NAVI's other changes.
mode_13h - Sunday, March 8, 2020 - link
> Why does it need AI cores? It's not for computation, it's for gaming.Nvidia relies on its Tensor Cores for good denoising performance with Global Illumination. Except, the performance of GI in Raytraced games is still so bad that most don't use it.
mode_13h - Sunday, March 8, 2020 - link
Yeah, you should really judge GPUs' efficiency at their stock specs. And if those specs push a design past its efficiency sweet spot, then too bad, cuz that's how the manufacturer chose to clock it.It's a moot exercise to play the "what if it had been clocked x% lower" game. Perhaps of academic interest, but it's of virtually no consequence if that's not how they actually shipped it.
Spunjji - Friday, March 13, 2020 - link
It's of interest when you're comparing the efficiency of an architecture, rather than the performance/value/etc. of a specific product. That is pretty academic, but then what are we here for if not to be nerdy about these things? :)mode_13h - Sunday, March 8, 2020 - link
> they worsened a lot their computation efficiency, by almost doubling the number of transistors while not adding any new feature (but packet math) with respect to Polaris.It's not accurate to say it added no new features. RDNA added:
* Hardware GCN compatibility
* 2x scalar throughput
* primitive shaders
* PCIe 4.0
* GDDR6
* Image sharpening
* Improved DSBR
I do agree that it would probably do well to drop GCN support. As improvements in area-efficiency tend to provide benefits in energy-efficiency, and we know RDNA 2 will have significant amounts of the latter, I'm also expecting it to feature the former.
Spunjji - Friday, March 13, 2020 - link
"That's completely wrong. Power efficiency is measured as the work done / energy used. If you clock your big beast at a low clock you get fantastic efficiency, but that is useless if you then have the performance of the smallest chip made by the competitors that consumes as much but costs 1/10th."You're exaggerating here. Radeon VII is a great example of a design that stinks at its default clocks and voltages, but when the clocks are scaled back a little the voltage scales back a *lot*. The end result is something that's actually quite competitive both in terms of PPW and overall performance, so not "completely wrong" at all. See also: how well Vega integrated graphics compete with Intel's iGPU in thermally constrained situations, even with a node deficit.
"Now, seen the number of transistors that they are using with RDNA to be at Pascal level, I can't imagine how big (in term of transistor counts) the new GPUs needs to be to be competitive with Nvidia's biggest beasts."
When they're talking about a 50% PPW increase - and their previous problems with performance scaling were high temperatures resulting from hitting a voltage wall - then they shouldn't need to rely entirely on making their designs larger just to compete. Based on recent history, they'll probably end up with slightly smaller designs that can't reach Nvidia's top performance levels but compete well in their class.
I guess we'll have to wait and see. :)
The_Countess - Friday, March 6, 2020 - link
"part of that advantage comes from an atypical advantage in manufacturing nodes"Actually, I would more say that its more of a return to form for AMD, because until 16/14nm, AMD was usually first on a new (TSMC) node for GPU's. Global foundries might have been the reason they weren't first with 16/14nm.
CrystalCowboy - Friday, March 6, 2020 - link
That fourth slide (AMD RDNA 2 PERF/WATT IMPROVEMENT) is a bit of a mess. Why are CPU and GPU on the same graph? Can they really maintain that their GPUs are tracking teh industry trend for perf/watt? The choice of colors for fonts and colors for graph lines also appears to imply things, but if you consider closely does not.Xyler94 - Friday, March 6, 2020 - link
they're showing how AMD buckled both the trend on the CPU side, and they're doing it with the GPU side also.Makes perfect sense really. Their CPU division is doing amazing, and really rose the company back into the hearts of enthusiast, and to see them say that RDNA2 should repeat that success is a good thing to see. But we'll have to wait until the cards come out, of course.
CiccioB - Friday, March 6, 2020 - link
They just did a graphics trick.The industry trend line passes from their Zen2 and RDNA2 points in the graph, not through Zen1-RDNA1.
From Zen1 to Zen2 they just followed the industry trend = using a new PP to increase efficiency.
Form RDNA to RDNA 2 they are doing the same reaching what the competition was 3 years ago.
So, beside Zen that is in line with industry trend, RDNA is a catch up game with what are the perf/performance of the competition.
Unfortunately they forgot to trace where the competition is. That is a point near their RDNA2 in the graph but using an older PP.
When switching to the new PP, the industry trend line will go steeper and they will be again a step behind.
The step above the trend line is the one from Polaris to RDNA, where they improve perf/W a lot by just spending all their transistor budget for the new features (that they do not have).
Spunjji - Friday, March 13, 2020 - link
RDNA already caught up to Pascal, and then some, matching Turing in practice because of the node advantage. RDNA 2 should perform *at least* like a Turing shrink to 7nm would. That's not exactly "just catching up to 3 years ago".sing_electric - Friday, March 6, 2020 - link
Perf/watt increases are great, but for a lot of consumers, it's perf/$ that drives purchasing decisions - and both AMD and Nvidia haven't been great on that front. AMD's 4-year-old RX at $240 (launch) still dukes it out with the ~$220 RX 5500 XT if you push clocks to make your 480 a quasi-580 - and don't tell me that "overclocking doesn't count," because that's EXACTLY what AMD did with its own 5600 XT launch.It's great that AMD is offering products that move up the stack, which forced Nvidia to offer more performance via its "Super" semi-update, but so far, we haven't really seen the competition drive prices down, or offer consumers meaningful performance increases over a GPU they bought years ago if they're willing to spend the same amount.
FreckledTrout - Friday, March 6, 2020 - link
The perf/$ only comes in play if the perf/watt is good enough. Meaning there is nothing AMD could do with Vega to make a 2080 TI consumer card and the thermals be able to be cooled with anything short of a custom water loop. This is to say if the perf/watt is good then the cost of the cards in each segment will come into play. Today AMD can't even step into the high end.Spunjji - Friday, March 13, 2020 - link
This!mode_13h - Sunday, March 8, 2020 - link
perf/W affects perf/$ because it lets you reach higher performance levels with a smaller die (which you can clock higher). So, improving efficiency can also create room to lower prices.Zizo007 - Friday, March 6, 2020 - link
RDNA2 for sure will be slower than the RTX 3000. The best they can do is beat the 2080 Ti by a few or 10+ percent and sell it for less. I hope that's the case and if not RTX 3000 will be even more expensive than the 2080 Ti 1200+$. I have an OCed 2080 Ti and can play any game at over 60FPS at 4K so no need any upgrade soon. GPUs nowdays are relatively way more powerful than before. When Crysis came out even the most expensive cards were struggling running it at Ultra 1080p. I don't see need for faster GPUs now unless 4K120+Hz monitors become affordable which isn't the case.Spunjji - Friday, March 13, 2020 - link
"The best they can do is beat the 2080 Ti by a few or 10+ percent and sell it for less."Citation needed.
"GPUs nowdays are relatively way more powerful than before."
Additional citations needed. Crysis is a terrible example to generalise from because it was specifically built to utilise GPUs more powerful than those in existence when it released. Games designers don't really do that anymore because it doesn't make any sense.
Zizo007 - Friday, March 6, 2020 - link
Zen 3 looks impressive with less game latency, improved IPC and improved clocks.I will upgrade my 1800X to a 4800X when it comes out. That will be a huge upgrade.
philosofool - Friday, March 6, 2020 - link
If the 17.5 TFLOPS number is true, this may actually be an Nvidia killer. Well, okay, Nvidia isn't going anywhere soon, but they will need to realize about a 34% increase in TFLOPS over a 2080 Ti to match that. TFLOPS are a rough measure, but AMD is currently sitting at 9.75 TFLOPS w/ RX 5700 XT, which suggests they will be getting +80% RX5700 XT performance with their top end Navi 2x card. It will be interesting to see which company has better RT implementation as well. That could be a big deal: with both companies offering RT and RT leaving its infancy, the RT value will matter.Spunjji - Friday, March 13, 2020 - link
I reckon Nvidia can easily exceed that level of performance - the question is more about how they'll choose to do it, and what the cost of that would be. If AMD can do it with a smaller die and no HBM then we might finally get solid price competition again.JasonMZW20 - Saturday, March 7, 2020 - link
I found some code relating to Navi 21 and Van Gogh (Renoir's replacement?) in Adrenalin version 20.1.3. Wasn't much and mostly related to power delivery system, but there's definitely proof of their development in the drivers.