AMD Delays Navi to Q4 2019

[Taf the Ghost]

8 minutes ago, leadeater said:

They still need something like Nvidia has been able to do with their archs, even if they don't hit the highest performance mark by 10%-15% something that can scale down easily would be more successful and cheaper for AMD. Being cheap to make alone isn't enough, products need to be desirable. Stressing archs outside of where they are well balanced just to compete doesn't make them desirable, they can still be well priced and meet the performance for the price point.

 

Getting general business aspects done very well unfortunately doesn't counter act being on a technical level behind your market competitors.

 

Polaris while good took too long to get there, binning on binning on binning is not as good as "Here's GP106 and it's good now, oh and here's a bunch of cut downs".

Fiji wasn't any good and hasn't aged well.

Vega is doing what Fiji didn't but similarly too late to matter.

Vega 20 is literally an entire generation too late. Seriously think how good and profitable it would have been competing in the same market as the 1080 Ti without having to do ultra cut rate pricing. Yes I know 7nm wasn't ready for that to happen but requiring 7nm to compete with 16nm is not a good look.

 

AMD may have been able to make Nvidia release reactionary products but it was in no way difficult for Nvidia to do so and utilized existing dies and cutting them down which is vastly cheaper and easier to do than what AMD had to do to drive Nvidia to do that, e.g. 1070 Ti.

 

Navi needs to be good, not good for AMD and compared to it's products but actually good for the products it will compete against. Navi won't just be competing with RTX 20 series because due to coming later to market than would be best it's in the timeline of Nvidia's potential 7nm generation. If Navi is going to be anything like Zen the performance jump required is significant. Timing just isn't on AMD's side right now.

AMD's problem in the technical space has been their design criteria. At the end of the day, Intel is low-end; AMD is mid-tier and Nvidia is high-end. That's the way each company designs their GPUs. Now, it was a massive risk for Nvidia to go that direction, and it actually didn't pay off until Maxwell, but, since then, they really make one design and cut it down until they reach the performance criteria of the category they want to hit. AMD made some bad design choices a while ago and it's still showing through.

 

The "Design Big then Cut Down" approach clearly works better in GPUs, but AMD will probably never get to that point. AdoredTV has been pretty harsh about the fact Nvidia has won the GPU Wars, and unless AMD gets to well over 30% of the Server CPU market, they're never going to have the funding to stay at pace, at the high end, with Nvidia. AMD's only real opportunity is to get to an actually scalable architecture then get MCM working for gaming GPUs. That's probably 2 full architectures away, so I wouldn't expect that until the 2024 range at the earliest. 

 

AMD clearly made the choice to prioritize Semi-Custom over Retail, and I really can't blame them. Nvidia is a cut-throat competitor with little qualm about screwing over the consumer to keep their profit margins, while at the same time investing the most money in driver development in the GPU industry.. AMD's tech, at this point, is old, but it's also sufficient for much of the current market. Which probably just means we're in a lull period for a bit.

[leadeater]

20 minutes ago, Taf the Ghost said:

AMD's problem in the technical space has been their design criteria. At the end of the day, Intel is low-end; AMD is mid-tier and Nvidia is high-end. That's the way each company designs their GPUs. Now, it was a massive risk for Nvidia to go that direction, and it actually didn't pay off until Maxwell, but, since then, they really make one design and cut it down until they reach the performance criteria of the category they want to hit. AMD made some bad design choices a while ago and it's still showing through.

Well the problem is ATI/AMD was also high-end and if they only want to compete in the mid to higher mid market they need to do a significant image and marketing shift because they are still trying to present themselves as a high end option when they just mostly aren't. GCN did actually start out promising, I guess it was another slight misstep where AMD thought the gaming market was going since future GCN developments focused on the compute side. I honestly think if we could go back to GCN1 and develop all over again without the assumption compute was going to be a big gaming market factor GCN now would be significantly different, better? Maybe, dunno.

 

In any case it's only recently AMD has been a mid-tier market sector company and that is only due to limitations in the architecture and not strategic choice planned out in advance.

 

Edit:

AMD wasn't wrong for where the market was going, just the time scale.

[pas008]

2 minutes ago, leadeater said:

Well the problem is ATI/AMD was also high-end and if they only want to compete in the mid to higher mid market they need to do a significant image and marketing shift because they are still trying to present themselves as a high end option when they just mostly aren't. GCN did actually start out promising, I guess it was another slight misstep where AMD thought the gaming market was going since future GCN developments focused on the compute side. I honestly think if we could go back to GCN1 and develop all over again without the assumption compute was going to be a big gaming market factor GCN now would be significantly different, better? Maybe, dunno.

 

In any case it's only recently AMD has been a mid-tier market sector company and that is only due to limitations in the architecture and not strategic choice planned out in advance.

i thought their compute side was their strong point maybe ahead of its time considering nvidia brought in mesh?

 

 

[bmichaels556]

This isn't a major issue in and of itself, but the issue is that it may end up as too little too late. Suppose they could someone hit that $250 GTX 1080 (or greater) performance. But what will Nvidia have right on the horizon? 

 

But this could also be a tactical movie for them. We could see a Playstation response to the One X with a Navi APU launched sooner than we think. I'm sure Sony isn't enthusiastic that their pro console gets stomped by the One X.

 

So perhaps we see a PS4 Pro v2 / PS5 at a great price point, competetive with One X with better performance. "Right now, we're only putting Navi in PS5. Wait just a bit longer, and you will be amazed at what our full Navi series will be able to do."

 

I wonder..

[leadeater]

15 minutes ago, pas008 said:

i thought their compute side was their strong point maybe ahead of its time considering nvidia brought in mesh?

Yea it's good but meaningless towards making it a product people actually widely want. In the last 4 years unless you were a miner strong compute was meaningless.

 

Not even the server sector cared because AMD failed to properly address CUDA, so where it could of be used there was no supporting framework to make people want to use it, all the while CUDA kept gaining adoption and market strength to now where it's near as much too late.

[pas008]

5 minutes ago, leadeater said:

Yea it's good but meaningless towards making it a product people actually widely want. In the last 4 years unless you were a miner strong compute was meaningless.

 

Not even the server sector cared because AMD failed to properly address CUDA, so where it could of be used there was no supporting framework to make people want to use it, all the while CUDA kept gaining adoption and market strength to now where it's near as much too late.

but now nvidia has it

it can be adopted to be used

you cant develop something and cut out 75% of the market because of it

 

its like rt, amd and nvidia cock blocked rt with intels larrabee, but now nvidias doing rt we will see that definitely come into the picture

 

thought amd came out with ROCm?

[leadeater]

34 minutes ago, pas008 said:

but now nvidia has it

it can be adopted to be used

you cant develop something and cut out 75% of the market because of it

 

its like rt, amd and nvidia cock blocked rt with intels larrabee, but now nvidias doing rt we will see that definitely come into the picture

 

thought amd came out with ROCm?

Nvidia has what? Nvidia always had compute capability along with 99% market control of the API used for compute, CUDA.

 

ROCm exists because AMD failed to address CUDA, it made no difference how good AMD cards were at it or that they had some cheaper cards with the full compute capabilities if you didn't have a development tool framework that allowed the use of that hardware. Worlds best engine is no good without a vehicle to put it in to.

 

Gaming sector didn't not use compute or avoid it they just used what Nvidia gaming hardware could do and a lot of the GameWorks features leverage compute capabilities, in a very specific Nvidia optimized way.

[pas008]

20 minutes ago, leadeater said:

Nvidia has what? Nvidia always had compute capability along with 99% market control of the API used for compute, CUDA.

 

ROCm exists because AMD failed to address CUDA, it made no difference how good AMD cards were at it or that they had some cheaper cards with the full compute capabilities if you didn't have a development tool framework that allowed the use of that hardware. Worlds best engine is no good without a vehicle to put it in to.

 

Gaming sector didn't not use compute or avoid it they just used what Nvidia gaming hardware could do and a lot of the GameWorks features leverage compute capabilities, in a very specific Nvidia optimized way.

https://devblogs.nvidia.com/introduction-turing-mesh-shaders/

 

I been drinking tonight but doesn't this help for amds compute side?

[leadeater]

31 minutes ago, pas008 said:

https://devblogs.nvidia.com/introduction-turing-mesh-shaders/

 

I been drinking tonight but doesn't this help for amds compute side?

Probably not directly as it'll be using all Nvidia hardware tech, firmware and drivers. It's still not directly compute usage and more hybrid but certainly a model that in theory should help AMD if they can implement similar in their drivers, extra hardware support may also be required but if that is the case I expect at a minimum Navi to have it and Vega 20 if not all Vega.

 

Quote

Simplicity Is Golden

Mesh shaders represent a radical simplification of the geometry pipeline. With a mesh shader enabled, all the shader stages and fixed-function features described above are swept away. Instead, we get a clean, straightforward pipeline using a compute-shader-like programming model. Importantly, this new pipeline is both highly flexible—enough to handle the existing geometry tasks in a typical game, plus enable new techniques that are challenging to do on the GPU today—and it looks like it should be quite performance-friendly, with no apparent architectural barriers to efficient GPU execution.

 

Like a compute shader, a mesh shader defines work groups of parallel-running threads, and they can communicate via on-chip shared memory as well as wave intrinsics. In lieu of a draw call, the app launches some number of mesh shader work groups. Each work group is responsible for writing out a small, self-contained chunk of geometry, called a “meshlet”, expressed in arrays of vertex attributes and corresponding indices. These meshlets then get tossed directly into the rasterizer, and Bob’s your uncle.

 

The appealing thing about this model is how data-driven and freeform it is. The mesh shader pipeline has very relaxed expectations about the shape of your data and the kinds of things you’re doing to do. Everything’s up to the programmer: you can pull the vertex and index data from buffers, generate them algorithmically, or any combination.

 

At the same time, the mesh shader model sidesteps the issues that hampered geometry shaders, by explicitly embracing SIMD execution (in the form of the compute “work group” abstraction). Instead of each shader thread generating geometry on its own—which leads to divergence, and large input/output data sizes—we have the whole work group outputting a meshlet cooperatively. This mean we can use compute-style tricks, like: first do some work on the vertices in parallel, then have a barrier, then work on the triangles in parallel. It also means the input/output bandwidth needs are a lot more reasonable. And, because meshlets are indexed triangle lists, they don’t break vertex reuse, as geometry shaders often did.

http://reedbeta.com/blog/mesh-shader-possibilities/

 

Sounds great on paper.

[pas008]

21 minutes ago, leadeater said:

Probably not directly as it'll be using all Nvidia hardware tech, firmware and drivers. It's still not directly compute usage and more hybrid but certainly a model that in theory should help AMD if they can implement similar in their drivers, extra hardware support may also be required but if that is the case I expect at a minimum Navi to have it and Vega 20 if not all Vega.

 

http://reedbeta.com/blog/mesh-shader-possibilities/

 

Sounds great on paper.

Think this is nvidia opening a door to close it?

Litigation? 

Remember nvidia will gain advantage with giving amd fans no reasons to upgrade 

Cut off reasons for them to spend again

If they get free performance

 

Release a product at 65% capacity

Give them 20% more for free

Competitor gives another 5%

Is there any reason for then to spend again?

Cash flow problems? 

Tell me how i can get anyone to bank on me with cash flow problems

[leadeater]

11 minutes ago, pas008 said:

Think this is nvidia opening a door to close it?

Litigation? 

Remember nvidia will gain advantage with giving amd fans no reasons to upgrade 

Cut off reasons for them to spend again

If they get free performance

Well I can certainly see how it could help cards like the Vega 64 be able to utilize much more of the hardware and increase the performance, enough to not need or want to upgrade? Have to wonder how much potential is left unused/untapped, if you look at the theoretical performance then a fair chunk but you know how theory vs reality often goes.

 

Looking at it from another angle it could be quite a master stroke to push AMD towards fixing architecture problems in their technology to have a shift already lined up that would have greatly suited AMD and now their planned changes are ineffective for future titles, though useful for existing. If that's the way it goes then I would have to say well played Nvidia, well played.

[pas008]

2 minutes ago, leadeater said:

Well I can certainly see how it could help cards like the Vega 64 be able to utilize much more of the hardware and increase the performance, enough to not need or want to upgrade? Have to wonder how much potential is left unused/untapped, if you look at the theoretical performance then a fair chunk but you know how theory vs reality often goes.

 

Looking at it from another angle it could be quite a master stroke to push AMD towards fixing architecture problems in their technology to have a shift already lined up that would have greatly suited AMD and now their planned changes are ineffective for future titles, though useful for existing.

To me amds maturing drivers is a problem

If you bought rx 480 why would you buy 580?

Same with 7970 then 7970ghz edition

Same with 290

 

They need to release full potential cards and release around 15% increase next gen 480 to 580

7970 to7970ghz editions aren't fitting

I can go on but you are not getting cash flow

Cash flow is what makes money cause it's all investment

[leadeater]

23 minutes ago, pas008 said:

To me amds maturing drivers is a problem

If you bought rx 480 why would you buy 580?

Same with 7970 then 7970ghz edition

Same with 290

 

They need to release full potential cards and release around 15% increase next gen 480 to 580

7970 to7970ghz editions aren't fitting

I can go on but you are not getting cash flow

Cash flow is what makes money cause it's all investment

I'm still running my crossfire 290Xs cos why would I upgrade? Those are old as hell cards.

[pas008]

36 minutes ago, leadeater said:

I'm still running my crossfire 290Xs cos why would I upgrade? Those are old as hell cards.

290 aged well

So you didn't spend in How many years?

[leadeater]

5 minutes ago, pas008 said:

290 aged well

So you didn't spend in How many years?

I got them near as much the day you could buy them, so like end of 2013. Reference models with EK full cover blocks. I would have upgraded by now if prices just generally didn't suck. People that invested back then in crossfire I figure are the worst off. Prices have basically doubled and a single current AMD card now is the same or slower than 2x290X.

 

So I either drop down to a single card and gain basically nothing or spend double+ on something actually faster like 2080/2080Ti.

[pas008]

9 minutes ago, leadeater said:

I got them near as much the day you could buy them, so like end of 2013. Reference models with EK full cover blocks. I would have upgraded by now if prices just generally didn't suck. People that invested back then in crossfire I figure are the worst off. Prices have basically doubled and a single current AMD card now is the same or slower than 2x290X.

 

So I either drop down to a single card and gain basically nothing or spend double+ on something actually faster like 2080/2080Ti.

Now you see?

They aren't releasing to full potential no reason to upgrade

Hurts their cash flow

If  they released near their 970 performance they'd be kings

 

People can't bet on fine wine

I know they don't allow over budget but if they complete the card amd would be killing it now with 290 if it was released @90% its potential

 

As you see cash flow hurts look at nvidia when they don't have it

[Humbug]

I too am still running my Sapphire R9 290 vapor-x. Best GPU I ever owned and does so well to this day with great driver support even for the newest games.

 

 

I too would have upgraded and gone to a higher resolution as well if the GPU market wasn't so messed up now. There is no point buying a RX 580 or a GTX 1060 because they are only marginally faster than what I have. Not worth the upgrade. And the really fast stuff is stupidly priced.

 

What we have seen from Nvidia is ridiculous price escalations every generation as if they are trying to kill pc gaming.

 

Navi really needs to shake up the GPU market and provide Vega56 performance at $200-250. It's about time.

 

 

 

 

 

[pas008]

3 minutes ago, Humbug said:

I too am still running my Sapphire R9 290 vapor-x. Best GPU I ever owned and does so well to this day with great driver support even for the newest games.

 

 

I too would have upgraded and gone to a higher resolution as well if the GPU market wasn't so messed up now. There is no point buying a RX 580 or a GTX 1060 because they are only marginally faster than what I have. Not worth the upgrade. And the really fast stuff is stupidly priced.

 

What we have seen from Nvidia is ridiculous price escalations every generation as if they are trying to kill pc gaming.

 

Navi really needs to shake up the GPU market and provide Vega56 performance at $200-250. It's about time.

 

 

 

 

 

If you didn't get all those free performance upgrades would you have spent more on amd?

[leadeater]

9 minutes ago, pas008 said:

Now you see?

See what? I haven't said anything counter to that and actually have already been saying that hence go back and read what I said about Navi.

[Humbug]

12 minutes ago, pas008 said:

If you didn't get all those free performance upgrades would you have spent more on amd?

It's not only performance upgrades. I have continuously got new features. Whether it's first class support for newest graphics APIs like Vulkan, or stuff like freesync that wasn't there when I bought my GPU, or various useful features in the Radeon Settings which were introduced later (e.g. frame rate target control). I also know that when new AAA games come out I will get a well optimized driver for my GPU as well even though it is old.

 

So if that stuff wasn't there I would not have spent more on AMD. It would mean that come time for my next upgrade I am more likely to switch to Nvidia.

 

In summary I know I can count on AMD to support me once I purchase their product. The reason I didn't upgrade is that the GPU market has stagnated in terms of price/performance.

[leadeater]

10 minutes ago, pas008 said:

If you didn't get all those free performance upgrades would you have spent more on amd?

Free performance upgrades? There has been none, the 290X hasn't gotten any faster, what it hasn't done is A) Gotten worse and B) AMD has released nothing until recently that is actually better, at double the price.

 

Since when is twice the performance at twice the cost actually a good thing? If you don't need the performance upgrade there is no reason to spend the money at all. If the market was healthy like it has in the past I could spend the same and get more.

[pas008]

7 minutes ago, leadeater said:

See what? I haven't said anything counter to that and actually have already been saying that hence go back and read what I said about Navi.

You spent awhile ago on 290s?

Have your contributed to amds current cash flow in gpu dept

Don't tell me it doesn't matter

Cash flow matters

 

[leadeater]

1 minute ago, pas008 said:

You spent awhile ago on 290s?

Have your contributed to amds current cash flow in gpu dept

Don't tell me it doesn't matter

I think you're trying to have a discussion I wasn't having or trying to prove a point for no reason. I haven't contributed to Nvidia either for the exact same reason, what you're trying to prove applies to both equally so it's a moot net zero point.

 

Honestly if prices were better I would have brought Nvidia, both have bad options relative to my current situation so they both get nothing. There's no major revelation there.

[pas008]

2 minutes ago, leadeater said:

I think you're trying to have a discussion I wasn't having or trying to prove a point for no reason. I haven't contributed to Nvidia either for the exact same reason, what you're trying to prove applies to both equally so it's a moot net zero point.

 

Honestly if prices were better I would have brought Nvidia, both have bad options relative to my current situation so they both get nothing. There's no major revelation there.

I'm saying if amd finished a product

Or nvidia opened up sooner

Public decision matters

That's why we wear nikes majority

We may try others but consistently we wear what is on top

480 or 7970 fuck even 5870

Came out on top with drivers and performance they'd kill it and we all know they became superior but in the their release they looked like shit

Huge no no

Rather buy matrox which is amd based lol

[leadeater]

11 minutes ago, pas008 said:

I'm saying if amd finished a product

Or nvidia opened up sooner

Public decision matters

That's why we wear nikes majority

We may try others but consistently we wear what is on top

480 or 7970 fuck even 5870

Came out on top with drivers and performance they'd kill it and we all know they became superior but in the their release they looked like shit

Huge no no

Rather buy matrox which is amd based lol

You should probably jump back to page 2/3 and read what I've written there