AMD's Polaris graphics cards rumored to launch in late May

It sounds pretty definitive to me:


Well, no they can't. The reason AMD are in the mess they are in is because their principle GPU competitor have gross margins over 100% higher than AMD (as does Intel). If you are content with AMD continuing to pare away R&D to allocate funds to debt servicing ahead of their 2019 fiscal armageddon, then fine...but without higher margins, there is no profit, and without profit there is no R&D, without R&D the product cadence slips and the holes need to be plugged with rebrands - and while rebrands are perfectly serviceable they don't drive revenue and profit like new architecture.

Look, I'm not going to restate myself for a 3rd time here. If you don't care to read the whole conversation then I don't care if it sounds definitive to you or not because you aren't even taking the time to read up.

I was more referring to the fact that Polaris is a smaller GPU, will be more per wafer, and will be more produced. They can sell at a lower margin because of this. I was not referring to their margins in general.

So long as Nvidia maintains it's mindshare people will buy their cards without second thought and they will get more money per card. Even when Fermi was out Nvidia managed to maintain 50% marketshare against AMD despite having hotter and slower cards.
 
Just found this

http://www.overclock3d.net/articles/gpu_displays/gtx_1080_ashes_of_the_singularity_benchmarks/1

It looks like one of two things is happening here. Either Nvidia still doesn't have good DX 12 support on Pascal or the 1080 isn't what it's chalked up to be. The fact that the 1080 can't beat the Fury X by much doesn't speak well of the card either way.
It has framerates that are around 10-12% higher than AMD's R9 Fury X from a game that has been known to work well with AMD Cards and we have no idea how Nvidia's Driver is playing with this game and the 1080.

I'm pretty certain there were several articles months ago about AMD vs Nvidia performance in this game and newer Nvidia drivers managed to beat AMD across the board when it wasn't running in DX12 and even in DX12 they caught up quite a bit.

You're also forgetting the price point the Fury X is currently being sold for. Just going by newegg (I'm from the UK but I went to the US site) I found one for $619.99 vs the 1080 which will apparently launch at $599.

So to conclude, what you've linked there is a cheaper Nvidia 1080 beating a more expensive AMD in an AMD favored game with a possibly un-optimized Nvidia driver. Yeah, you're really proving your point there...

Edit: I found one for $599.99 on newegg, so they're the same price.
 
Except they DO sell and they DO make money... They make far more profit off of more expensive cards than cheap ones....
They have higher margins. Not to be confused with revenue or total profit.

http://store.steampowered.com/hwsurvey/videocard/

So the #1 installed card (4.5% of users) is the 970 gtx which is a ~350 card. So AMD needs to have competition in this price range. but the first card you see that is 400+ is the 980 and 980 ti that are less than 1.5% combined of installed cards. So AMD can easily forego that end of the market if they want to, and still be very successful. But they do need a challenger/alternative to nvidias 1070 . . . the 1080? not so much.

note that there are a TON of cards in the middle that are 750, 760, 770, 960s, 7700s 7800s 7900s, etc etc etc. that range in price from 100-$350. these make up roughly 90% of installed cards. But it is clear that most people are not buying cards that are $400+.

Its always nice to be king of the hill, but not always worth the cost. And its not like AMD is ignoring this part of the market, its just that developing products takes time and $$. Do they want to bring the mid/low end to market first? or do they want to bring high end to market first. IMO it makes sense to bring the mid/low end to market first, as this is where all of the sales are. I can see how enthusiasts could be disappointed. But it is a business savy move.
 
I'm not saying their performance won't be better than AMD's, just that Nvidia's "faster than SLI 980s" claim is totally false.

Another thing to note, AMD's polaris 10 die is 33% smaller than Nvidia's 1070. The yields are going to be much better and it's going to be cheaper to produce. AMD will be able to sell them at a lower price and still make very good margins.
The fanboy is strrrong in you.
 
Last edited:
Just found this
http://www.overclock3d.net/articles/gpu_displays/gtx_1080_ashes_of_the_singularity_benchmarks/1
It looks like one of two things is happening here. Either Nvidia still doesn't have good DX 12 support on Pascal or the 1080 isn't what it's chalked up to be. The fact that the 1080 can't beat the Fury X by much doesn't speak well of the card either way.
First of all, even a noob should know that with AotS builds and their on-again-off-again code path alterations, the only real valid comparisons are between benchmarks using the same build. That is why they have a database with filtering options in the first place.
Secondly, AotS represents AMD absolute best case scenario as far as benchmarks are concerned. It is the one (worse than mediocre) game that AMD fans have been holding on to as "sure thing" win for the company. If AMD's best case scenario is now under threat at all from what amounts to Nvidia's first generation async compute, I wouldn't really wouldn't be calling for a day of celebration just yet...especially since AMD themselves seem to not solved the conservative rasterization support for Polaris which will be used in a lot more instances in actual popular games.
It has framerates that are around 10-12% higher than AMD's R9 Fury X from a game that has been known to work well with AMD Cards and we have no idea how Nvidia's Driver is playing with this game and the 1080.
Precisely.
The engine was developed on and for AMD GCN architecture
The engine has been developed from the outset to take advantage of AMD ACE structure and workgroup allocation and continues to be optimized for the architecture.
The engine was designed to be a core of demonstrator and marketing tool for AMD (Star Swarm)

The fact that the 1080 gets anywhere close to AMD's flagship card at all seems astounding given the 1080 is appreciably behind in every architectural metric excepting the raster - 60% less texture address units, 60% less ALUs, 60% less bandwidth, 24% less transistors, and an engine tailored for Fiji.
 
Last edited:
Yes but that only takes into consideration, the people that use their cards for gaming on Steam platform.
Still a strong indicator I think. The large majority of the million GTX 970/980 cards sold in the first four months from launch were 970's and that trend has only increased as the 980 fell out of favour with the high-end crowd which gravitated to the 980Ti. Just looking at Nvidia's financials, the 970 seems largely responsible for the year of record average selling price, revenue, and profit.
The GTX 970 has been for all intents and purposes a money printing press.

EDIT: I had been wondering where Nvidia's GP 106 was, since it seemed a good indicator of where AMD was with Polaris...well, it didn't take long to show itself. Mobile MXM GP106 inbound.
 
Last edited:
All this speculation and posturing in the comments over products that aren't even released yet. Just never ceases to amaze me. Until Nvidia and AMD come out with their respective new products, people need to get their panties out of a bind.
 
It has framerates that are around 10-12% higher than AMD's R9 Fury X from a game that has been known to work well with AMD Cards and we have no idea how Nvidia's Driver is playing with this game and the 1080.

I'm pretty certain there were several articles months ago about AMD vs Nvidia performance in this game and newer Nvidia drivers managed to beat AMD across the board when it wasn't running in DX12 and even in DX12 they caught up quite a bit.

You're also forgetting the price point the Fury X is currently being sold for. Just going by newegg (I'm from the UK but I went to the US site) I found one for $619.99 vs the 1080 which will apparently launch at $599.

So to conclude, what you've linked there is a cheaper Nvidia 1080 beating a more expensive AMD in an AMD favored game with a possibly un-optimized Nvidia driver. Yeah, you're really proving your point there...

Edit: I found one for $599.99 on newegg, so they're the same price.

Seeing as the GTX 1080 hasn't been released yet it is at best speculatively cheaper. It's really not a good point seeing as the Fury X will drop in price by the time or shortly after the 1080 releases.

FYI, they had Doom on Vulcan, not DX 12. DX 12 performance isn't something that can be fixed via drivers.

The fanboy is strrrong in you.

The fact that you read that comment and got "fanboy" from it is more of a sign of your own opinions than anything else. You know what they say about the person who throws the fanboy card first.

First of all, even a noob should know that with AotS builds and their on-again-off-again code path alterations, the only real valid comparisons are between benchmarks uasing the same build. That is why they have a database with filtering options in the first place.
Secondly, AotS represents AMD absolute best case scenario as far as benchmarks are concerned. It is the one (worse than mediocre) game that AMD fans have been holding on to as "sure thing" win for the company. If AMD's best case scenario is now under threat at all from what amounts to Nvidia's first generation async compute, I wouldn't really wouldn't be calling for a day of celebration just yet...especially since AMD themselves seem to not solved the conservative rasterization support for Polaris which will be used in a lot more instances in actual popular games.

Precisely.
The engine was developed on and for AMD GCN architecture
The engine has been developed from the outset to take advantage of AMD ACE structure and workgroup allocation
The engine was designed to be a core of demonstrator and marketing tool for AMD (Star Swarm)

The fact that the 1080 gets anywhere close to AMD's flagship card at all seems astounding given the 1080 is appreciably behind in every architectural metric excepting the raster - 60% less texture address units, 60% less ALUs, 60% less bandwidth, 24% less transistors, and an engine tailored for Fiji.

I do realize the shiftiness of the article I linked. It's probably pointless to continue to post anything further because precise details are scarce.

AOTS is optimized for AMD but it's not like a Project Cars deal here. The performance on Nvidia cards isn't bad.
 
Seeing as the GTX 1080 hasn't been released yet it is at best speculatively cheaper. It's really not a good point seeing as the Fury X will drop in price by the time or shortly after the 1080 releases.
Undoubtedly, but that probably has more to do with the Fury cards effectively going EOL. AMD has basically stated that Fiji inventory is being diverted to the Pro Duo (and I guess a very small number of S9300 X2's). Given the Pro Duo's reign as VR's sole heir looks to be a matter of weeks maybe AMD will rethink the Fury situation - it would be downright embarrassing to have to apply a one time charge to unsold inventory of their flagship GPU.

FYI, they had Doom on Vulcan, not DX 12. DX 12 performance isn't something that can be fixed via drivers.
No entirely true, but the onus falls on the developer to make sure that their code works as well as it can on the individual architectures. Any new introduced architecture will require its own coding considerations.
AOTS is optimized for AMD but it's not like a Project Cars deal here. The performance on Nvidia cards isn't bad.
AotS is actively optimized continually for AMD. The company is sponsoring Oxide after all. Oxide's involvement with Nvidia optimization is somewhat more cursory and seems generally to be blocking features that are detrimental.
The difference between AotS and Project Cars should be obvious. Project Cars isn't held up as the pinnacle of Nvidia's graphics architecture prowess and isn't the cornerstone of their marketing. AotS however is held by AMD and its marketing as just that.
 
Seeing as the GTX 1080 hasn't been released yet it is at best speculatively cheaper. It's really not a good point seeing as the Fury X will drop in price by the time or shortly after the 1080 releases.
Your right, it hasn't been released yet. Hence why I said "apparently" cheaper as we all know the price can change quickly. But currently. As it stands Nvidia have said to the world $599 on release. So it's the best we've got.

If the price drops on the Fury X. It's just to put it in line with it's performance vs the competition.

FYI, they had Doom on Vulcan, not DX 12. DX 12 performance isn't something that can be fixed via drivers.
I don't know why you're mentioning Doom. I never mentioned Doom? Only AoTS from the article you linked?

Anyway let me go find the article about Nvidia's driver updates and the effects they've had on AoTS. You are right. You cannot "fix" DX12 support if the architecture simply doesn't support certain features. But you can make DX12 run much better on the hardware you do have which is what they did with Maxwell. Hold on. Let me go find it and I'll link it. It's quite interesting to see the difference in driver revisions.

EDIT: turns out I wasn't going mad, just google "Ashes of the Singularity Nvidia Driver" and you'll find plenty of articles. Basically, Nvidia managed to get 20+ FPS extra out just by optermising at the driver level.
 
Last edited:
Except they DO sell and they DO make money... They make far more profit off of more expensive cards than cheap ones....
while the profit margins might be larger for individual cards in the high end market, the vast majority of sales come from low and mid range cards. these constitute the majority of the discrete video card profits.
high end margins are just used to make the investors happy since they look good on paper.
 
Your right, it hasn't been released yet. Hence why I said "apparently" cheaper as we all know the price can change quickly. But currently. As it stands Nvidia have said to the world $599 on release. So it's the best we've got.

If the price drops on the Fury X. It's just to put it in line with it's performance vs the competition.


I don't know why you're mentioning Doom. I never mentioned Doom? Only AoTS from the article you linked?

Anyway let me go find the article about Nvidia's driver updates and the effects they've had on AoTS. You are right. You cannot "fix" DX12 support if the architecture simply doesn't support certain features. But you can make DX12 run much better on the hardware you do have which is what they did with Maxwell. Hold on. Let me go find it and I'll link it. It's quite interesting to see the difference in driver revisions.

EDIT: turns out I wasn't going mad, just google "Ashes of the Singularity Nvidia Driver" and you'll find plenty of articles. Basically, Nvidia managed to get 20+ FPS extra out just by optermising at the driver level.

I must have gotten Doom and AOTS mixed up between two posts.

Undoubtedly, but that probably has more to do with the Fury cards effectively going EOL. AMD has basically stated that Fiji inventory is being diverted to the Pro Duo (and I guess a very small number of S9300 X2's). Given the Pro Duo's reign as VR's sole heir looks to be a matter of weeks maybe AMD will rethink the Fury situation - it would be downright embarrassing to have to apply a one time charge to unsold inventory of their flagship GPU.


No entirely true, but the onus falls on the developer to make sure that their code works as well as it can on the individual architectures. Any new introduced architecture will require its own coding considerations.

AotS is actively optimized continually for AMD. The company is sponsoring Oxide after all. Oxide's involvement with Nvidia optimization is somewhat more cursory and seems generally to be blocking features that are detrimental.
The difference between AotS and Project Cars should be obvious. Project Cars isn't held up as the pinnacle of Nvidia's graphics architecture prowess and isn't the cornerstone of their marketing. AotS however is held by AMD and its marketing as just that.

I really wish that AMD would just bring the price of the pro duo down. A dual gpu card that consumes that little power would be an awesome replacement for Multi-card setups. It would be killer at $1,000 or $1,100.
 
"Considering Nvidia's GTX 1070 launched at $379"

will launch or supposed to launch at. This hasn't happened yet so it can't be past tense.

WCCFTech has the die size of Polaris 10 at 1/3 smaller than the 1070. It would be very easy for AMD to undercut Nvidia and still make a very good profit.

Well, they announced the price, so I guess they figure that counts as "launching".... If the new AMD cards only match Fury, then they are in some trouble... Nvidia has found a way to take their $1000 card and beat it for almost a third the price... Fury was already behind the Titan, so it doesn't look good...

Do you believe everything you read or hear? The new GPUs do not beat the Titan in desktop gaming, the advantage was announced concerning VR performance. Samsung has already done that with a mobile phone.
 
while the profit margins might be larger for individual cards in the high end market, the vast majority of sales come from low and mid range cards. these constitute the majority of the discrete video card profits.
high end margins are just used to make the investors happy since they look good on paper.
I was talking about the 970 (and soon to be 1070)... that's the most used card on Steam... and the profit margins are very nice for Nvidia on that card - and it pretty much kills AMD at that price point...
 
I was talking about the 970 (and soon to be 1070)... that's the most used card on Steam... and the profit margins are very nice for Nvidia on that card - and it pretty much kills AMD at that price point...

I agree the 970 was a very nice card when it launched but lately it's about even with the 390 depending on your needs. I'd recommend the 970 for those with a smaller power supply, those who want to overclock, or those who want to SLI. Otherwise the 390 is a tad bit faster and does better at higher resolution gaming. I'd recommend the 390 for future proofing as well, it's hard to ignore the much larger memory bandwidth of the 390 over the 970 (384 GB/s vs 196 GB/s) and the larger memory size. I think a big reason the two cards are much closer today could be AMD's improved drivers. AMD still doesn't have perfect utilization of their cards under DX 11.
 
I'd recommend the 970 for those with a smaller power supply, those who want to overclock, or those who want to SLI. Otherwise the 390 is a tad bit faster and does better at higher resolution gaming.
I doubt at this point in time I'd recommend either unless buying second hand or sourcing the third-tier/grey markets (the traditional dumping grounds for EOL'd cards). If Polaris 10 offers anything like R9 390 performance it will do so cheaper, and judging by the clocking potential of Pascal, spending $300 on a GTX 970 makes little sense if for an extra $79 you can get double the vRAM and 2+GHz clocks. From VRWorld's sit down with Nvidia and board partners:
Stock GTX 1080 is clocked at 1.66 GHz, with Turbo Boost lifting it to 1.73 GHz. Founders Edition includes overclocking-friendly BIOS to raise the clocks to at least 2 GHz, and the presentation showed the chip running at 2.1 GHz. The main limiting factor for the overclocking beyond 2.2 GHz is 225 Watts, which is how much the board can officially pull from the power circuitry: 75 Watts from the motherboard and 150 W through 8-pin PEG connector. However, there are power supply manufacturers which provide more juice per rail, and we’ve seen single 8-pin connector delivering 225 W on its own. Still, partners such as ASUS, Colorful, EVGA, Galax, GigaByte, MSI are preparing custom boards with 2-3 8-pin connectors. According to our sources, reaching 2.5 GHz using a liquid cooling setup such as Corsair H115i or EK Waterblocks should not be too much of a hassle.
It obviously pertains to the GTX 1080, but it would be likely that the stock limit will also apply to the 1070, the same paradigm that applied to the GTX 980/970 where 20% factory overclocks on the 970 and beefier power ( 2 * 8-pin or 1 * 8-pin + 1 *6-pin) were relatively common.

Unless a 28nm card is an absolute steal, I wouldn't be rushing to recommend them.
 
I doubt at this point in time I'd recommend either unless buying second hand or sourcing the third-tier/grey markets (the traditional dumping grounds for EOL'd cards). If Polaris 10 offers anything like R9 390 performance it will do so cheaper, and judging by the clocking potential of Pascal, spending $300 on a GTX 970 makes little sense if for an extra $79 you can get double the vRAM and 2+GHz clocks. From VRWorld's sit down with Nvidia and board partners:

It obviously pertains to the GTX 1080, but it would be likely that the stock limit will also apply to the 1070, the same paradigm that applied to the GTX 980/970 where 20% factory overclocks on the 970 and beefier power ( 2 * 8-pin or 1 * 8-pin + 1 *6-pin) were relatively common.

Unless a 28nm card is an absolute steal, I wouldn't be rushing to recommend them.

That raises a good question. Isn't Nvidia afraid of people overclocking their cards too much? What's the point to buying the 1080 if you can get a 1070 and overclock it to 1080 performance? You're seeing a large jump in price for their top end cards but the performance jump isn't going to justify the price (see the fury x, 980 Ti, titan x). I mean the 970 is already good at overclocking. Give the card a performance boost and increase it's overclock and I don't see many people paying that large premium for the 1080. GDDR5x is a decent memory bandwidth boost but as we can see by AMD's cards with high memory bandwidth (390/x) it's just not a huge factor in many games.
 
That raises a good question. Isn't Nvidia afraid of people overclocking their cards too much? What's the point to buying the 1080 if you can get a 1070 and overclock it to 1080 performance?
Not really a good point when you consider overclocking both equally. The 1080 will still be the better performer.
 
Not really a good point when you consider overclocking both equally. The 1080 will still be the better performer.
Exactly. The always has been and always will be a segment of the buying public that value the increased performance over monetary consideration, and those that place added value in a complete (non-salvage) part.

When I was younger and my tech hobby was closer to obsession, I thought nothing of plunking down $3K+ on tri-SLI 8800Ultra's in 2007 for no other reason that Nvidia bought out a chipset that supported three card usage, and I simply wanted the best I could afford. I knew that Nvidia had no worthwhile tri-SLI scaling even in games and apps where it actually worked. I also knew that the cards would be obsolete within a year - but $3K represented little more than two weeks wages and I already had virtually everything else I really wanted, so what the hell. Performance per dollar? nope...performance per watt? nope - but man, that setup and months upon months of subsequent tinkering were virtually priceless in the enjoyment it bought. The next year I repeated the exercise with three BFG (later EVGA) watercooled GTX 280's, and the year after that three Sapphire HD 5850 Toxic's - chosen because their performance beat the hell out of the HD 5870 and they were 2GB cards.

Performance per dollar has never really entered into my own personal buying equation, and probably never will - and I am certainly not alone.
It really is a shame that this mentality needs explaining rather than just being intuited on a tech enthusiast site forum.
 
Not really a good point when you consider overclocking both equally. The 1080 will still be the better performer.

Assuming that the 1080 can overclock as well. We don't know how well the GDDR5x will react to an overclock. It's not a huge shift like HBM but it can still cause issues.
 
Back