AMD CEO Lisa Su: High-end Navi GPU and Radeon Raytracing is coming

Ironically, i could se the opposite, and say that you're trying to diminish the problems AMD has. How is that argument nonsense? What logic are you following? Because that honestly sounds like the only logic thing. It's statistics nothing more, and it's how everything works, if i encounter many many less (or none) problems with an X product, same for the people i know, so let's say 20 different persons, in 99% of the case they have no problems with nvidia cards, and had at least 1 problem (even stupid) with AMD cards, what does this tell you? That if you take 20000 people the average won't be much far from that, unless me and all my friends have been lucky for years getting nvidia cards, and unlucky with AMD cards each freaking time, honestly a bit unrealistic.
Never had any regression honestly, and if there was it was hard to spot unless you benchmark at every driver release, and even there it's nothing related to "problems" the fact a newer driver could perform is probably something they might do on purpose, which is surely a s**tty move, but still it has nothing to do with problems or unreliability, and it's not like your getting half the performance from a driver version to another, actually, i still need to see an appropriate, accurate test about this matter because i've seen people trying to prove this but fail miserably because their test methodology wasn't consistent and unreliable. Trying older drivers isn't working around drivers, because even if you use internet as your source, most of the people who roll back drivers is just because in some cases older drivers make a videogame run slightly better, and not because part of it doesn't work or is totally broken, like it happens with AMD cards, like it happened to me with my friend recently, we went over 5 versions of drivers across 4 or 5 months and when some of them fixed something, broke something else, it was a game of getting the lesser evil. You keep recommending whatever you want, i'll keep recommending nvidia, until at least i can trust AMD with their cards, and people i build a computer for don't call me back because something related to the videocard doesn't work.

It's not like I FEEL THE NEED to tamper with the radeon settings, it's just that i want to set my own fan curve and disable a couple of features i can't use or don't need, it's not tampering, it's choosing, and like it's possible with nvidia, i'd like it to be possible without causing problems with AMD too. As i already said OEM has little to do when it's the stock bios, and when that OEM is also the one which manufactures reference cards for AMD. I claim i DON'T FEEL THE NEED to modify anything because the stock settings in that case are what suits me best, but even modifying those settings i don't encounter problems, like i do with AMD, and like i said before, nvidia drivers/control panel don't interfere in the same way Radeon Settings does, and it also doesn't cause problems like Radeon Settings does. I don't understand the part where you talk about terrible fan curves on different cards, it's not like i leave the default fan curve, i always use a more aggressive curve, because most of the time people don't mind the extra noise as long as the card runs cooler and doesn't cause problems reaching high temps, that's what i usually do on systems with a nvidia card, now if i try to do the same on a system with an AMD card strange stuff happens, and as much as i try to deal with it, there's always something that will come up in exchange for solving a problem. Not all etailers or shop allow for reviews, that's why forums are a more direct way to access users' opinions and experiences. You are playing favoritism actually, you ARE saying AMD is better than it is, when saying it's on the same level of nvidia, because i've only seen them consistently decrease prices of their products in order to stay competitive, and that's not actually a real merit, decreasing profit margin is bad for them too, but in a way or another they have to sell those GPUs, because nobody but fanboys would buy them if they were at the same price of nvidia counterparts and that's a fact, no matter how hard you try and make it look otherwise, i can say i do prefer nvidia, but not because i am a fan, these things don't deserve fans, it's not like they're a football team, they make a direct profit over us and they don't make you a special deal because you're their fanboy, i prefer them because they've been extremely more reliable compared to AMD cards in these last years, so extremely that each time i owned or make someone buy an AMD card, i had to deal with multiple problems, and i haven't had to deal with any problems when it was nvidia in play. So how should i drive my opinion if not from direct personal experience then?

 

I haven't owned a Radeon/AMD GPU in almost 20 years. But I acknowledge AMD getting back into the high end GPU business as crucial for market balance. That's the main reason why NVidia has been able to increase the prices on their high end GPUs over the years, because there is no competition. So I fervently hope that AMD makes a good showing in the high end bracket and is competitive with Ampere in not only rasterisation, but ray tracing as well.

 

I do expect 3080 Ti to overpower 2080 Ti, but by how much, hard to say. I'd be very, very surprised if it was 50% like the other thread suggested. 15% would be realistic, if conservative, though I'm sure disappointing for many. I can't see how Nvidia could switch to 7nm with great fanfare and then release a product that's only better with RTX On. At the end of the day pretty much nobody still cares enough about RT to judge cards based on it. Of course they could instead drop the prices, but I don't see that happening. Haven't the current high prices worked just fine for them? I haven't studied any sales numbers possibly available, but I reckon they sold as many 2080 Tis as they wanted, as well, despite the outrageous price. The huge monoliths are troublesome, so they don't even want to make them cheap and sell truckloads of them.

 

Nvidia has been on top for too long and could use a good whooping. We could use a high end card with the old $600-$800 prices to stabilize the pricing on the entire lineup of video cards. Or perhaps it will just be a bit underwhelming again.

 

It's true that current game console have a business model that involves selling heavily subsidised hardware to gamers, then making up the costs from game publishing through hefty revenues for every sale in the official store.
Steam and iOS/Androd stores have similar policies of demaning high percentages, even though PC and mobile games do not need to subsidise end user's hardware and end users can top-performing or entry-level configurations, which is not possible on consoles.
So the prices of console and PC games are very closely matched nowadays, since articstic content and game programming for current platforms is really differernt that much, unlike what it used to be.

I doubt we'll see such high prices because electronic gaming is very demand-driven and relatively elastic - i.e. there are millions of customers but very few of them will want to pay unusually high prices to play even top-quality games.
So publishers know very well that it's always better sell million copies for MSRP $40 each, than sell maybe ten thousand for $80 each.

If 'big Navi' has double the CUs of the Navi10, it too shall have no problem outperforming RTX 2080 Ti by 15% - tests show that the 5700 XT is mostly less than 30% behind.

 

Besides, games are very rarely $60 right now. Once you get all the DLC (and future promised DLC) that make up the second half of the game, most AAA games are actually $90-$120. Sneaky.

 

Anywhere close to that would be excellent for the consumer and it would force badly needed price cuts. Probably unlikely that AMD has the financial headroom / risk-taking capability to reduce pricing to that aggression level. In reality, they only need to reduce pricing to where the public favour their products while, of course, staying true to the price-to-performance rule (that rule always in direct context to a competitor's equivalent product). And, with a current high price bar (thanks Ngreedia!!), AMD, as a business, will want to maximise profits.

 

Haha. I have to admit that's an excellent point you are making there!

I don't quite understand that comparison to 2080 Ti anyway. The big Navi isn't even out yet, who knows when it will be, so what's the point of comparing it to the relatively old 2080 Ti anymore? It's actually quite condescending or pitying, depending on your pov, to make a big deal about AMD perhaps being able to beat a last gen Nvidia GPU. It's the reality, of course, since of late AMD hasn't been able of anything more, but you can't hype an upcoming product that way at all. If we assume the big Navi and Ampere 3080 Ti will be released this year, then those two should be compared to each other.

 

Yes, but they need to have a product in the game to know where they need to focus improvements at.

 

I just can't understand why 'big Navi' is dismissed out of hand by some posters. The point is, if AMD manages to double the CUs and the memory bandwidth, this shall provide twice the performance, given internal fabrics and power efficiency can keep up - and AMD had 2 years to improve on the RDNA architecture, if you consider multiple production delays for Navi1x.

 

Things do not scale that linearly. Not even Nvidia, which has a more efficient architecture, can scale that way.

And then you need to take into account other factors such as power consumption, temperatures and clocks. The RX5700XT, which is a 250mm chip, already has a 225W TDP and a good amount of AIBs custom cards are already around 270W~290W.

Even if it comes out with the better optimized 7nm EUV and a more refined version of the RDNA, there are no miracles. I believe AMD can come this year with something on 2080Ti level or slightly above (after all, the 2080Ti is a 2018 card), but twice the performance of a RX 5700X It´s simply a daydream.

 

An additional 40% would be impressive depending on how it goes with possible scaling ,additional hardware and potentially HBM though that would then also be a big factor on how much a bigger Navi GPU would potentially cost.
But yeah it would somewhat pull ahead of the existing 2080Ti's lead but not by much and expecting a 50% or higher increase is a bit unrealistic.
(Well maybe if AMD goes all out on a Navi20 and RDNA2 but it would also greatly increase the cost for such a GPU so it would need to really scale up to make it competitive if it's in the same price range instead of a slightly lower level.)

30% is about how it went from Vega to Navi so about the same is what I'd expect possibly a bit higher but then Vega was also limited by it's hardware to where the more powerful Radeon VII was almost surpassed by Navi10 / 5700 which I expect not to be the case again so who knows though AMD does need a fairly reasonable large performance jump but cost and power draw is also going to be a factor and whenever NVIDIA has something to unveil or they might simply lower prices of the existing lineup of GPU's to make these an attractive alternative depending on how AMD's big Navi positions itself against the 2080 or the 2080Ti but we'll see and for now it's just speculative although I wouldn't expect too much the way current hardware scales.

EDIT: Actually for a few titles it might still not be enough to take the lead of the 2080Ti if that's what AMD is actually attempting with a big Navi GPU.


EDIT: The memory, cores, scaling overall and drivers, power delivery and clock speeds.
30 - 50% perhaps bit up and down depending on what's holding the card back and how well it can utilize the hardware particularly how well the drivers can actually leverage said hardware though overall it's a 5% - 10% performance difference or so unless some title is just completely broken performance wise so driver wise it can result in a boost but I would not expect later drivers to completely change how the card performs.

So for a few titles it'll be a big improvement and for others it might be a bit less it'll be interesting to see more on how it all ends up once some info is available on Navi20 and RDNA2 later this year.
(Though I think my expectations are a tad optimistic here.)

 

@Denial I agree that in the future games will all use RT and we will probably migrate from the current form of making graphics to that.
What the Nvidia approach means is that, at some point, a new GPU will come out that doesn't play the old games anymore: that would suck.
Crytek RT demo doesn't need RTX cores, so I'm hoping we either progress with RT software or the hardware just transforms into a hybrid form.
It doesn't make sense having dedicated hardware for a feature that can be turned off, I'll gladly turn it off to gain performance and in turn paid for hardware that I don't use.
We need better solutions, it's just what I'm trying to say. Nvidia made RTX to try and sell cards, I don't think that will work for the majority of people.

 

Trust me, it'll be "50% performance increase in specific scenarios", not 50% raster performance increase, that's basically impossible. I'm guessing they'll say that it's 50% increased RT performance which sounds logical due to maturity of the tech.

 

The Truring RT cores simply off-load the ray tracing it's not really a proprietary tech. My 1080ti can turn RTX on but the performance hit is muck more massive. HW RT will be the way forward AMD has in a round about way has admitted this.

Also eventually DX9 support will need to be dropped. How many modern cards support DX8 or later?

 

Incorrect.

Crytek RT is a path tracer, has no founding in any standardized api and only works in their limited tech demo.

 

I'm not trying to diminish anything; I'm able to acknowledge AMD's problems. I'm just saying Nvidia isn't as perfect as you're making them out to be. The fact you haven't encountered any issues is more luck than a sign of their development skills. It is utterly absurd to say 99% of Nvidia users don't encounter problems. Even people who give 5 star reviews for an Nvidia product still still find an issue. You are so fixated on your personal anecdotes, where all you have to do is look at the bugfixes in Nvidia's driver update changelogs. Those bugs had to come from somewhere. I am not saying Nvidia's drivers are worse, because (at least on Windows) they're definitely not, but, encountering issues isn't as rare as you think.

I too would want to see proof of that. I'm not suggesting anything that severe.

Yet you say I'm the one who is biased...

Wait so you're having an issue with a reference card? Like with those awful blower fans? Seems a little unfair to use a known crappy model to influence your decisions.

Yeah, I don't buy that. I've never had issues controlling fan profiles regardless of what GPU brand I had. I have, however, also had issues trying to control fan speeds of models from both brands.

*sigh* then look at a different one... It's like pulling teeth with you. But even I recommended to look at forums, and trust me, you'll find plenty of posts about Nvidia.

No actually, I'm saying Nvidia is worse than you think it is. AMD is not great, but Nvidia is barely better.

By looking at all data, not just your own. If I recommended Nvidia for every person I built PCs for, I know for a fact I would encounter complaints. I also know that if I recommended AMD for every PC I built for someone, I'd encounter more complaints. There is no one perfect choice.

 

normally, 99% of nvidia users don't encounter problems.

normally.... 99% of AMD users don't encounter problems.


crap happens though.

it just happens more often with AMD at new launches than it does nvidia.
the extremely flawed turing drivers at launch were an anomaly that usually doesn't get seen.
but nvidia was playing with a new programmable scaler, the first of its kind and no other geforce card has had one before - it was fixed within 2 months.

Navi has been out how long now, and how are the drivers going?

You had since launch poor performance at 1080p fixed
You had since launch crashes on 75hz displays when browsing the internet.

Steve from GN couldn't even get an AMD employee to pass feedback on with the issues he has been talking about for months.


Atleast they fixed that LTT discovered bug where the 5700XT was boosting itself till it was on fire


Lets not forget that time AMD forgot they were required to implement Memory mapping support for DX9 and previous applications leading to apps texture allocating themselves into a crash on R9 3x0's - for 6 months.

The good news is that NAVI's success seems to have resulted in software engineer hires at AMD recently, reported bugs aren't taking half a year to get a look at, fixing them though......