The 30 Year History of AMD Graphics, In Pictures (Archive)

[Mac266]

"And although it (RX 480) isn't as fast as Nvidia's Pascal-based cards, it does fare well against the GeForce GTX 970"

This sentence is so misleading. True, it isn't as fast as the 1070/80 but it does a damn good job against the 1060 6GB, and consistently beats it in DX12 titles. I'm not usually one to accuse bias, but come on guys.

 

[alidan]

"And although it (RX 480) isn't as fast as Nvidia's Pascal-based cards, it does fare well against the GeForce GTX 970"

This sentence is so misleading. True, it isn't as fast as the 1070/80 but it does a damn good job against the 1060 6GB, and consistently beats it in DX12 titles. I'm not usually one to accuse bias, but come on guys.

overall about 5% better then a 970, once you put it into dx12 and vulcan its able to close the gap and in some cases beat out a 1070

but then again, toms things project cars is a benchmark game, as much as I use toms as a primary news source, I don't use them as the place for gpu benchmarks.

 

[Guest]

ATI Radeon 9700 Pro and HD7970 are only two chips worth mentioning, everything was and is shit.

 

[bit_user]

ATI Radeon 9700 Pro and HD7970 are only two chips worth mentioning, everything was and is shit.

If you're making a short list of their best-of, then the HD 5870 should definitely be on there.

I'd also say Fury X. It's just bad luck, for them, that the GTX 980 Ti was so good. Otherwise, Fury X might've earned a place in their hall of fame.

 

[alidan]

ATI Radeon 9700 Pro and HD7970 are only two chips worth mentioning, everything was and is shit.

If you're making a short list of their best-of, then the HD 5870 should definitely be on there.

I'd also say Fury X. It's just bad luck, for them, that the GTX 980 Ti was so good. Otherwise, Fury X might've earned a place in their hall of fame.

its not that the 980ti was good, its that nvidia was willing to put out the 980ti just so amd didn't have a clear win. in dx12 and vulcan the fury, non x, is able to pull ahead of the 1080 in some games,

that said, you have an architecture held back by process too, as to my understanding it was made for 20nm apposed to 28 but that fell through.

but the entire 5000 lineup, especially the 5770, were amazing value for the price, something that today is hard to say of anything but a mid range card.

 

[InvalidError]

but the entire 5000 lineup, especially the 5770, were amazing value for the price, something that today is hard to say of anything but a mid range card.

Still using an HD5770 here

Yes, progress at the lower end has slowed down quite a bit over the past few years. AMD and Nvidia don't want their lower-end parts to overlap their mid-range and high-end anymore and are growing the gaps with each new generation.

 

[bit_user]

its not that the 980ti was good, its that nvidia was willing to put out the 980ti just so amd didn't have a clear win.

How can you say it wasn't good? Did you see the table, in my earlier post? How can you be anything but impressed that it roughly equaled Fury X with mostly inferior specs?

Not only that, but with the same process, TDP and only 1B more transistors, it almost doubled the performance of the 780 Ti! Maxwell was a huge leap, for Nvidia. It even wiped away most of their GPU-compute disadvantage.

 

[alidan]

its not that the 980ti was good, its that nvidia was willing to put out the 980ti just so amd didn't have a clear win.

How can you say it wasn't good? Did you see the table, in my earlier post? How can you be anything but impressed that it roughly equaled Fury X with mostly inferior specs?

Not only that, but with the same process, TDP and only 1B more transistors, it almost doubled the performance of the 780 Ti! Maxwell was a huge leap, for Nvidia. It even wiped away most of their GPU-compute disadvantage.

first off, flops are a theoretical max, it is shaders x clock speed x 2
in the case of the fury x, its 4096x1.05x2
This number is a very theoretical number, and is only really relevant to products within the same architecture when it comes to graphics.

im not saying it wasn't good, im saying in that context, its not that the 980ti was good its that nvidia was willing to put out the 980ti at all.

the 980ti if i remember right is a slightly cut down titan x, they basically sacrificed the 1000$ card so amd wouldn't have a price performance advantage.

and as for the 780ti being half as fast, remember nvidia pushed gameworks hard, and gameworks runs like hell on everything it touches, there are very small things you wouldn't even think about that are part of the render pipeline, such as a lightning effect, so even when you turn things off like tessellation, you are still in that pipeline.

you see it with tessellation how nvidia was willing to just throw the 700 line under the bus and not put a limiter in the driver and how they black boxed the tessellation for a very long time so it was either off or on 64x or equally stupid number because it hurt amd just a little more then it hurt their own cards, it wouldn't shock me if the same was true for every effect, if i remember right there were several games that had a 780ti performing worse or at least on par with a 960

I mean, its a mature 28nm process that they designed the chips for, they know the quarks, they know how to take advantage of the process to get more out of it then the 700 line, they gutted double precision even more then the 700 line was, and I would love to know exactly what made the 900 series as good as it was, because by all metrics, it should not be where it stands, but we likely won't know until that info becomes irrelevant to any competition. my money is half on knowing the process inside and out, making a dx11 asic and holding back driver improvements from the 700 line, I mean that's the most obvious one, somewhere between the end of the 600 and the start of the 900 they figured something out, never implemented it into the 700 and put the development entirely into the 900. give people with a 700 a reason to upgrade, and get a performance advantage on the same node while also culling back double precision a bit more, but then that is just all my understanding, I could be missing something completely.

 

[bit_user]

im not saying it wasn't good, im saying in that context, its not that the 980ti was good its that nvidia was willing to put out the 980ti at all.

the 980ti if i remember right is a slightly cut down titan x, they basically sacrificed the 1000$ card so amd wouldn't have a price performance advantage.

They're in the business to make money. They're a publicly traded company, and doing anything just to spite AMD that didn't make business sense would lead to shareholder lawsuits and possibly the removal of its CEO. They didn't sell this card at a loss. Beyond that, everything else is fair.

So, I really don't understand your phrasing that "they were willing to put this card out, at all." The reality is that most people aren't going to buy a $1k card. Furthermore, many 980 Ti's will clock higher than Titan X's, enough to deliver greater overall performance. The Titan X was really for the price-insensitive early-adopters and those who needed 12 GB of RAM (e.g. for deep learning - I know a non-gamer who bought one just for that). I really think the 980 Ti was done in a way to maximize their profits - not as a sacrifice. I think they'll do the exact same, with the 1080 Ti - put it out either when AMD launches Vega, or just when the new Titan X saturates the market of people willing to spend > $1.2k on a graphics card.

BTW, I bought my card for $460, around the time of the GTX 1080 launch. That they could sell for so little is further evidence that they weren't exactly losing money on them.

As for double-precision performance, AMD is doing the same thing.

 

[InvalidError]

BTW, I bought my card for $460, around the time of the GTX 1080 launch. That they could sell for so little is further evidence that they weren't exactly losing money on them.

You can get it for that price now because all of the R&D behind the 980Ti and other products sharing the same die has been paid off and amortized on previous sales with pressure from new product introductions to clear existing inventory while it is still marketable. If it had always sold at that price, then Nvidia might not have recovered its development and manufacturing costs.

 

[alidan]

im not saying it wasn't good, im saying in that context, its not that the 980ti was good its that nvidia was willing to put out the 980ti at all.

the 980ti if i remember right is a slightly cut down titan x, they basically sacrificed the 1000$ card so amd wouldn't have a price performance advantage.

They're in the business to make money. They're a publicly traded company, and doing anything just to spite AMD that didn't make business sense would lead to shareholder lawsuits and possibly the removal of its CEO. They didn't sell this card at a loss. Beyond that, everything else is fair.

Just on this point. Nvidia is the company that put out the founders edition. The effectively made the reference card the most expensive version, or at least should have. So lets go this way. Nvidia likely had bad titan x dies that they sat on, for a long time, just waiting on amd, because if a 980 wasn't powerful enough for you, you were either going to go with sli or pony up for a titan x, because if you need or want performance, you don't got an option. now, nvidia had between 550 and 1000, they had that price range to play with. where would you have dumped the 980ti given its performance? Its capable of beating out a stock cooler titan x, so performance wise you could even demand over 1000$ for it, but they didn't.

They announced and released it a week before the fury and fury x.

Keep in mind this is the company that brings us proprietary everything, and charges out the butt for it all, even when there are free options that are better. This is the company that made the founders editions. Why chose 650, when 750 or 800 would also work?

This card can only really be seen as a middle finger to amd. If it came out after amd put their fury, ok, if it was significant before, fine... but this was 1 week before the fury and effectively killed all hype for amd. They could have put the 980ti out any time after march 17th but they chose to hold it till just before the amd announcement.

 

[Martell1977]

@alidan - that's the nature of competition, get there first and grab the market share you can.

 

[bit_user]

Keep in mind this is the company that brings us proprietary everything, and charges out the butt for it all, even when there are free options that are better. This is the company that made the founders editions.

All valid complaints, but that's what a market leader can get away with. This sort of bad behavior is one reason I bought AMD 'till now. If both brands give me roughly comparable options, I'll reward AMD for their support of open standards (and before anyone says anything about Mantle, I think it's clear that was just put forth as a prototype to influence Direct X and OpenGL/Vulkan).

Having said that, the reason I bought a 980 Ti is that I wanted a VR-capable card that also had an analog out. AMD had nothing to offer me (and the outboard converters I've found are both expensive and inadequate). As a side-benefit, I believe the factory-overclocked 980 Ti that I got is probably faster than comparable air-cooled Fury X cards, and certainly for the money I paid. It's certainly been a delight, thus far.

This card can only really be seen as a middle finger to amd. If it came out after amd put their fury, ok, if it was significant before, fine... but this was 1 week before the fury and effectively killed all hype for amd. They could have put the 980ti out any time after march 17th but they chose to hold it till just before the amd announcement.

Here's where I call BS. I think you see malice where there's only ruthless competition.

@alidan - that's the nature of competition, get there first and grab the market share you can.

Totally agree.

 

[CaptainTom]

It's interesting how close the AMD vs Nvidia history mimics the Nvidia vs 3DFX story.

Just like Nvidia did, ATI at first struggled to produce truly high-end cards and focused solely on value/features. Then over time they finally learned enough and trounced their competition for a few years. But eventually this caused them to rest on their laurels and stop innovating.

Now it's just been a back and forth between Nvidia and AMD ever since the HD 5000 series. Wonder how 2017 will go, I suspect it will be a repeat of the 4000 vs 200 series.

 

[bit_user]

But eventually this caused them to rest on their laurels and stop innovating.

Now it's just been a back and forth between Nvidia and AMD ever since the HD 5000 series. Wonder how 2017 will go, I suspect it will be a repeat of the 4000 vs 200 series.

I agree that they under-invested in GPU tech, since GCN 1.0. It seems simplistic to say they "got lazy", but their CPU business almost sunk the company and the GPU division was probably too focused on consoles. The execs probably then assumed they'd naturally keep their GPU lead and didn't try to find ways to finance the R&D that would've helped solidified that lead.

I think they're now past a tipping point, where Nvidia is too far ahead on hardware efficiency and software tech for AMD ever again to surpass them. So, their GPU situation now mirrors their CPU situation, where the best they can hope to do is be in the same ballpark as the high-end products from their competitors.

For better or worse, their biggest hope now lies with their Chinese partners.

 

[Martell1977]

I agree that they under-invested in GPU tech, since GCN 1.0. It seems simplistic to say they "got lazy", but their CPU business almost sunk the company and the GPU division was probably too focused on consoles. The execs probably then assumed they'd naturally keep their GPU lead and didn't try to find ways to finance the R&D that would've helped solidified that lead.

I think they're now past a tipping point, where Nvidia is too far ahead on hardware efficiency and software tech for AMD ever again to surpass them. So, their GPU situation now mirrors their CPU situation, where the best they can hope to do is be in the same ballpark as the high-end products from their competitors.

For better or worse, their biggest hope now lies with their Chinese partners.

I think a big factor is when they combined the CPU and GPU divisions to focus on APU's and consoles. Now that they have separated the GPU division again, I expect we should be seeing some interesting improvements.

As for software, The latest "GeForce Experience" software makes you log in with nVidia, which is totally ridiculous. The software nVidia has is only a factor to those that would use it, which I suspect is not a huge number compared to overall sales numbers. Plus, driver support is pretty much equal between them now, so that's a non-factor.

Between Zen and Vega, AMD just might make a strong comeback, the RX series has been rather successful and they gained back a nice amount of market share, now they need some flagship models to get a bit more. I find it interesting that in the 1050(ti) launch it was said that those 2 complete the 10XX series, so there is a good chance there wont be a 1080ti after all, as many speculated.

 

[bit_user]

As for software, The latest "GeForce Experience" software makes you log in with nVidia, which is totally ridiculous. The software nVidia has is only a factor to those that would use it, which I suspect is not a huge number compared to overall sales numbers. Plus, driver support is pretty much equal between them now, so that's a non-factor.

I have totally unsubstantiated beliefs that their performance per Watt advantage isn't only due to the hardware, but also owing to a raft of clever software optimizations.

I find it interesting that in the 1050(ti) launch it was said that those 2 complete the 10XX series, so there is a good chance there wont be a 1080ti after all, as many speculated.

I think it'd be a mistake to read so much into that. They might've just been talking about the silicon. It'd be very easy for them to launch a 1080 Ti, and I'm sure they're getting some pressure from their board partners to do so.

 

[Martell1977]

I think it'd be a mistake to read so much into that. They might've just been talking about the silicon. It'd be very easy for them to launch a 1080 Ti, and I'm sure they're getting some pressure from their board partners to do so.

The language seemed very specific, but I wouldn't doubt the partners would like the ability to offer the extreme top end of the lineup. The reason I'm of the opinion that a 1080ti is unlikely is that it would only serve to cannibalize titan xp sales. As we know titan xp is already a cut down chip, so the 1080ti would be less VRAM and likely a further cut down chip. Given the performance gap between the titan xp and 1080, there isn't much room to drop a 1080ti in, without making the titan xp mostly irrelevant.

 

[bit_user]

The reason I'm of the opinion that a 1080ti is unlikely is that it would only serve to cannibalize titan xp sales. As we know titan xp is already a cut down chip, so the 1080ti would be less VRAM and likely a further cut down chip. Given the performance gap between the titan xp and 1080, there isn't much room to drop a 1080ti in, without making the titan xp mostly irrelevant.

Meh, they will quickly saturate the market of people willing to pay $1.2k for a video card. After that, it would make sense to price-drop Titan X and let their partners make overclocked versions with custom cooling. At that point, it now becomes the 1080 Ti.

And 980 Ti's were often faster than Maxwell Titan X, or so I've read. So, apparently, there's no reason 1080 Ti can't be faster than Pascal Titan X. What the early adopters got for their money was the fastest card.. at the time. Everyone knows that there'll eventually be something faster and cheaper.

It'll happen when Vega hits, if not before.

 

[ledhead11]

The reason I'm of the opinion that a 1080ti is unlikely is that it would only serve to cannibalize titan xp sales. As we know titan xp is already a cut down chip, so the 1080ti would be less VRAM and likely a further cut down chip. Given the performance gap between the titan xp and 1080, there isn't much room to drop a 1080ti in, without making the titan xp mostly irrelevant.

Meh, they will quickly saturate the market of people willing to pay $1.2k for a video card. After that, it would make sense to price-drop Titan X and let their partners make overclocked versions with custom cooling. At that point, it now becomes the 1080 Ti.

And 980 Ti's were often faster than Maxwell Titan X, or so I've read. So, apparently, there's no reason 1080 Ti can't be faster than Pascal Titan X. What the early adopters got for their money was the fastest card.. at the time. Everyone knows that there'll eventually be something faster and cheaper.

It'll happen when Vega hits, if not before.

I agree with you in all respects. Whatever the next thing is, NVidia is rumored to debut it at the next show in January.

 

[ledhead11]

I still have to put it out there again. Thank you Tom's for all these retrospective histories. From NVidia to AMD/ATI to the Others it's been a lot of fun to read. They may not be perfect but its such an ambitious, and IMO, a needed subject. Soooo much has happened in the last 10-20 years for computer technology and many take for granted what we have now.

 

[bit_user]

It'll happen when Vega hits, if not before.

I agree with you in all respects. Whatever the next thing is, NVidia is rumored to debut it at the next show in January.

You know that's when Vega is slated to launch, right? So, I think they'll upstage it with either GTX 1080 Ti or maybe even a consumer-level chip with HBM2.

 

[alidan]

But eventually this caused them to rest on their laurels and stop innovating.

Now it's just been a back and forth between Nvidia and AMD ever since the HD 5000 series. Wonder how 2017 will go, I suspect it will be a repeat of the 4000 vs 200 series.

I agree that they under-invested in GPU tech, since GCN 1.0. It seems simplistic to say they "got lazy", but their CPU business almost sunk the company and the GPU division was probably too focused on consoles. The execs probably then assumed they'd naturally keep their GPU lead and didn't try to find ways to finance the R&D that would've helped solidified that lead.

I think they're now past a tipping point, where Nvidia is too far ahead on hardware efficiency and software tech for AMD ever again to surpass them. So, their GPU situation now mirrors their CPU situation, where the best they can hope to do is be in the same ballpark as the high-end products from their competitors.

For better or worse, their biggest hope now lies with their Chinese partners.

Amd was so horrifically mismanaged along with one of the the ceos, if i remember correctly signing a asinine contract with what was it global foundries, then jumping ship to them, along with another selling their mobile gpu/cpu line off and signing a non compete.

then you put in that nvidia made a dx11 asic and here you go, current pascal is maxwell shrunk down to 16nm, mean while amd on the 28nm chips has the 290x/390x beating out the 1070 depending on game and dx12 implementation. personally i'm hoping for Vulcan to get a foothold, but almost everything amd had from gcn1-fury is hitting well above what it should have competed with.

Its not really sitting on their butt, its they got screwed by the processes falling through, and with vega my understanding is availability of hbm2, a good amount of their trouble also came from money issues caused by the cpu decision which also wasnt helped by intels monopoly bs they pulled too.

oh, and just a side note, once you equalize a 980 and 1070 to have the same theoretical flops, there is no difference between the cards, meaning there is not architectural advantage to a pascal over a maxwell, only a clock rate advantage due to the shrink, meanwhile over on amd, if you do the same thing, a 480 has about a 10-15~% advantage over the prior gen amd. you want to talk about sitting on your butt its nvidia, but they were ahead enough that they were able to.

I agree that they under-invested in GPU tech, since GCN 1.0. It seems simplistic to say they "got lazy", but their CPU business almost sunk the company and the GPU division was probably too focused on consoles. The execs probably then assumed they'd naturally keep their GPU lead and didn't try to find ways to finance the R&D that would've helped solidified that lead.

I think they're now past a tipping point, where Nvidia is too far ahead on hardware efficiency and software tech for AMD ever again to surpass them. So, their GPU situation now mirrors their CPU situation, where the best they can hope to do is be in the same ballpark as the high-end products from their competitors.

For better or worse, their biggest hope now lies with their Chinese partners.

I think a big factor is when they combined the CPU and GPU divisions to focus on APU's and consoles. Now that they have separated the GPU division again, I expect we should be seeing some interesting improvements.

As for software, The latest "GeForce Experience" software makes you log in with nVidia, which is totally ridiculous. The software nVidia has is only a factor to those that would use it, which I suspect is not a huge number compared to overall sales numbers. Plus, driver support is pretty much equal between them now, so that's a non-factor.

Between Zen and Vega, AMD just might make a strong comeback, the RX series has been rather successful and they gained back a nice amount of market share, now they need some flagship models to get a bit more. I find it interesting that in the 1050(ti) launch it was said that those 2 complete the 10XX series, so there is a good chance there wont be a 1080ti after all, as many speculated.

There is no way a 1080ti isn't going to happen, they have how many titans that failed to be perfect, they are cutting them down and biding their time to do what they did last time, amd is talking about vega or announcing it, nvidia drops the 1080ti

the difference being, by then enough games will be on dx12 for that to matter more than dx11 performance, amd should compair far more favorably this time around then they did with fury launch and 980ti... though looking at it now, fury and furyx users are getting more life out of their cards

The reason I'm of the opinion that a 1080ti is unlikely is that it would only serve to cannibalize titan xp sales. As we know titan xp is already a cut down chip, so the 1080ti would be less VRAM and likely a further cut down chip. Given the performance gap between the titan xp and 1080, there isn't much room to drop a 1080ti in, without making the titan xp mostly irrelevant.

Meh, they will quickly saturate the market of people willing to pay $1.2k for a video card. After that, it would make sense to price-drop Titan X and let their partners make overclocked versions with custom cooling. At that point, it now becomes the 1080 Ti.

And 980 Ti's were often faster than Maxwell Titan X, or so I've read. So, apparently, there's no reason 1080 Ti can't be faster than Pascal Titan X. What the early adopters got for their money was the fastest card.. at the time. Everyone knows that there'll eventually be something faster and cheaper.

It'll happen when Vega hits, if not before.

If i remember right, the 780ti was a more powerful chip then the titan at least if you only played games. as for the 980ti being faster then a titan x, only in non overclocking situations, once you put a third party cooler on a titan x it beats out a 1080 by about 5% and a 980ti is weaker than it by about 10% if i remember the numbers correctly.

 

[ledhead11]

It'll happen when Vega hits, if not before.

I agree with you in all respects. Whatever the next thing is, NVidia is rumored to debut it at the next show in January.

You know that's when Vega is slated to launch, right? So, I think they'll upstage it with either GTX 1080 Ti or maybe even a consumer-level chip with HBM2.

Yeah, I know about Vega and but have also noticed how NVidia keeps dodging the questions of Ti or a name/real specs for the next thing. It's like a cross between poker and go fish with these two lately.

 

[bit_user]

... selling their mobile gpu/cpu line off

I thought that happened well before most of the mismanagement. Possibly around the time of the ATI/AMD merger.

oh, and just a side note, once you equalize a 980 and 1070 to have the same theoretical flops, there is no difference between the cards, meaning there is not architectural advantage to a pascal over a maxwell, only a clock rate advantage due to the shrink

Yeah, I read that. I take it as further testament to what a huge step forward Maxwell was, for them. Other than that, I'm not sure it matters.

If you look at everything that improved with Pascal (die shrink, HBM2/GDDR5X, and I forget the 3rd main improvement - maybe something to do with power management or efficiency), I don't think you can fairly accuse them of sitting on their butts.

As for AMD's improvements, it all depends on what their baseline was. Fury X got less performance per Watt than Maxwell, so we'll have to see how Vega matches up against Pascal before deciding who's been lazy.

Yeah, I know about Vega and but have also noticed how NVidia keeps dodging the questions of Ti or a name/real specs for the next thing. It's like a cross between poker and go fish with these two lately.

Well, until AMD launches or announces anything, they really should keep quiet. In the face of no high-end competition, they can only hurt themselves with such rumors.