[WCCF] AMD Radeon R9 390X Pictured

[3DVagabond]

I'm sure NV has a plan if Fury is much faster than the 980Ti.

Well they won't be able to pull their "slower but smoother" campaign again like they did the last time AMD had them beat in raw performance. It'll all be about them offering more RAM and them encouraging sites to do reviews to stress that. Or Gameworks and AMD's slower support in those titles. I doubt they can just pull a faster chip out of a hat though.

 

[Silverforce11]

Well they won't be able to pull their "slower but smoother" campaign again like they did the last time AMD had them beat in raw performance. It'll all be about them offering more RAM and them encouraging sites to do reviews to stress that. Or Gameworks and AMD's slower support in those titles. I doubt they can just pull a faster chip out of a hat though.

Full GM200, 6GB, custom models or water hybrid cooled models. >1.4ghz boost out of box.

Titan X +20% right there.

Think about the ratio of sales now for 980Ti vs Titan X. What is NV going to do with all the full GM200 dies?

 

[JDG1980]

Titan is already fully enabled.

Yes, I know. What I meant was that if Fury X beat Titan X by a relatively small margin, Nvidia could win back the performance crown by releasing a consumer-focused SKU which would basically be a cheaper, hot-clocked Titan X with 6GB of RAM. By combining GTX 980 Ti's higher boost and acceptance of aftermarket coolers with the Titan X's fully enabled chip, Nvidia has the margin to improve performance by ~5-10% without doing any substantial new engineering work.

 

[Face2Face]

Full GM200, 6GB, custom models or water hybrid cooled models. >1.4ghz boost out of box.

Titan X +20% right there.

Think about the ratio of sales now for 980Ti vs Titan X. What is NV going to do with all the full GM200 dies?

Very true. Maxwell has been no chump when it comes to overclocking. Seeing that even reference cards with a small voltage bump can hit 1400Mhz+. I would like to see an article or a consumer post some scaling numbers for big maxwell, for games specifically.

 

[raghu78]

Full GM200, 6GB, custom models or water hybrid cooled models. >1.4ghz boost out of box.

Titan X +20% right there.

Think about the ratio of sales now for 980Ti vs Titan X. What is NV going to do with all the full GM200 dies?

The problem there is you are assuming that Fiji Fury has very less OC headroom. Sorry until reviews come out I don't think we can say anything. btw it also depends on whether AMD has done significant improvements to perf/sp. If the perf/sp has gone up by >= 20% AMD will take the GPU crown with ease. I am confident Fury will have better OC headroom than Hawaii. My guess is Fury will be able to hit 1350-1400 Mhz with a bit of voltage tweaking. Anyway we will know in 3 weeks. so why jump to conclusions.

 

[maddie]

The problem there is you are assuming that Fiji Fury has very less OC headroom. Sorry until reviews come out I don't think we can say anything. btw it also depends on whether AMD has done significant improvements to perf/sp. If the perf/sp has gone up by >= 20% AMD will take the GPU crown with ease. I am confident Fury will have better OC headroom than Hawaii. My guess is Fury will be able to hit 1350-1400 Mhz with a bit of voltage tweaking. Anyway we will know in 3 weeks. so why jump to conclusions.

3 weeks? I thought it was the 16th.

 

[beginner99]

Think about the ratio of sales now for 980Ti vs Titan X. What is NV going to do with all the full GM200 dies?

Good question. Maybe yields aren't that good anyway or maybe enough are sold under quadro or tesla brand. But yeah full GM200 named GTX 990 or so seems possible.

 

[raghu78]

3 weeks? I thought it was the 16th.

I expect the reviews to go up on 24th. thats why i said 3 weeks. :biggrin:

 

[Tuna-Fish]

I expect the reviews to go up on 24th. thats why i said 3 weeks. :biggrin:

Why? I'd expect that the cards will be mailed to the reviewers before launch and that embargo goes down on the 16th.

 

[raghu78]

Why? I'd expect that the cards will be mailed to the reviewers before launch and that embargo goes down on the 16th.

no. I expect similar to Hawaii launch - the unveiling event happens first with tech specs and architectural details. Pricing maybe withheld to go up alongwith reviews. Reviewers get samples when they are at the event and the reviews go up in a week's time. I think the only question is are all R9 3xx cards launched at the same time or is it slightly staggered ?

 

[tential]

Why? I'd expect that the cards will be mailed to the reviewers before launch and that embargo goes down on the 16th.

I think the reason he is saying this is because AMD has been VERY VERY VERY tight lipped on this. If reviewers don't even have cards yet and we aren't getting leaks from anywhere else really then perhaps AMD is actually selling their high end model themselves and it will only be reference models and doesn't want word out until the 16th. 16th would be a reveal, then reviewers would get cards and then 24th would be release date.

Obviously this is all speculation though. I have no idea when reviews are going up, I really am just happy about a 16th reveal. I really hope this is a game changer card. I think if it's at $500-700 I can pick it up and sell it next year for only a $200 loss for 14nm cards. Not bad imo.

 

[Kenmitch]

I'd think at least some teasing of performance on the 16th. Leaks following once cards are in the wild.

 

[Kippa]

All this talk of "It has to beat the Titan X......" all depends on its price. If it is priced near the Titan X then yes I would expect it to at least equal or beat the Titan X. If however it is significantly cheaper than the Titan X then it doesn't have to beat or even equal the Titan X or the 980ti, it just has to be priced well relative to its performance.

 

[Face2Face]

The problem there is you are assuming that Fiji Fury has very less OC headroom. Sorry until reviews come out I don't think we can say anything. btw it also depends on whether AMD has done significant improvements to perf/sp. If the perf/sp has gone up by >= 20% AMD will take the GPU crown with ease. I am confident Fury will have better OC headroom than Hawaii. My guess is Fury will be able to hit 1350-1400 Mhz with a bit of voltage tweaking. Anyway we will know in 3 weeks. so why jump to conclusions.

It also has a lot to do with scaling. Not all GPUs scale the same when overclocked. Regardless, If Fury can do 1400Mhz+, it would be amazing feat for AMD.

 

[Headfoot]

All this talk of "It has to beat the Titan X......" all depends on its price. If it is priced near the Titan X then yes I would expect it to at least equal or beat the Titan X. If however it is significantly cheaper than the Titan X then it doesn't have to beat or even equal the Titan X or the 980ti, it just has to be priced well relative to its performance.

If we got another 290 -- e.g. 0-10% slower than the 980 Ti for 33% less money, I'd be ecstatic. ~$500 like that would be great. I imagine that's where the cut down Fiji will slot in

 

[Flapdrol1337]

All this talk of "It has to beat the Titan X......" all depends on its price. If it is priced near the Titan X then yes I would expect it to at least equal or beat the Titan X. If however it is significantly cheaper than the Titan X then it doesn't have to beat or even equal the Titan X or the 980ti, it just has to be priced well relative to its performance.

But we want to see faster gpu's.

And we want amd to make some money this time, so they don't slowly go bankrupt. It would be good if they don't have to undercut the 980Ti, or at least not by much.

 

[maddie]

But we want to see faster gpu's.

And we want amd to make some money this time, so they don't slowly go bankrupt. It would be good if they don't have to undercut the 980Ti, or at least not by much.

This is my main concern.

Consumers of both companies will benefit the most by having two financially healthy companies battling for sales.

Only a FOOL will want a monopoly, remember that the next time someone says "Die AMD".

 

[sm625]

CALCS:
Hawaii = 438mm2 [fact]
Shader portion = 292 mm2 [assume 2/3 area]
Non shader = 146mm2

Fiji = 600mm2 [Assumed]
Lets assume a 15% increase in density [middle range for GloFlo improvement]
Non shader < 146mm2 [HDL, HBM memory controller]
Say 15% for Gloflo and and say further 10% for memory controller = 146 x 0.85 x 0.9 =111mm
Shader = 600-111 = 489mm2
With 15% density increase, we get 482 x1.15 = 554mm2 [TSMC equivalent]
Compare with Hawaii = 554/292 = 1.9x2816 = 5350 shaders [80+ CU]

I think 10% for memory controllers is awfully conservative. GDDR5 memory controllers are huge and complex. They should be able to get rid of at least half of those transistor when using HBM. Can you rerun your calculations based on the possibility of 40% reduction in the non-shader portion of the die? On a 600mm^2 die, we should see at least 500mm^2 of pure compute.

If you take the 292mm^2 shader portion of hawaii, port it to GloFo, the size becomes roughly 15% smaller, or 248.2 mm^2. Seems to me you could easily fit two of these on 500mm^2, which puts Fiji at a nice and even 5632. This makes too much sense to me. Fiji has to be essentially a 295x2 on one die, with enough memory bandwidth to feed it.

The 4096 shader part could very well be just the 390X. Hawaii is more than 18 months old. Why would anyone think they havent found a way to bump up from 2816 to 4096 in all that time?

 

[ShintaiDK]

This is my main concern.

Consumers of both companies will benefit the most by having two financially healthy companies battling for sales.

If it only was that simple.

Remember that next time graphics cards goes up in price or they dont go to a lower node due to cost.

Shrinking dGPU volume, higher design cost, higher R&D cost.

Whats your solution? Because the current one isnt substainable.

Nobody says AMD should die. But from a business perspective only one of the two companies can survive in the long run until there is only IGPs. Then it doesnt matter who of AMD or nVidia survives. And if its AMD, then the next battle is who of AMD and Intel. If its nVidia, well, then they already lost the long run against Intel.

 

[njdevilsfan87]

Think about the ratio of sales now for 980Ti vs Titan X. What is NV going to do with all the full GM200 dies?

Put them on the 980Ti and let us flash to TX with 6gb of VRAM. One can always hope!

I really do hope Fury is faster than the 980 Ti. I can't see many going for a slightly cheaper Fury over a Ti due to the VRAM. Regardless if it's really needed or not.

 

[maddie]

I think 10% for memory controllers is awfully conservative. GDDR5 memory controllers are huge and complex. They should be able to get rid of at least half of those transistor when using HBM. Can you rerun your calculations based on the possibility of 40% reduction in the non-shader portion of the die? On a 600mm^2 die, we should see at least 500mm^2 of pure compute.

If you take the 292mm^2 shader portion of hawaii, port it to GloFo, the size becomes roughly 15% smaller, or 248.2 mm^2. Seems to me you could easily fit two of these on 500mm^2, which puts Fiji at a nice and even 5632. This makes too much sense to me. Fiji has to be essentially a 295x2 on one die, with enough memory bandwidth to feed it.

The 4096 shader part could very well be just the 390X. Hawaii is more than 18 months old. Why would anyone think they havent found a way to bump up from 2816 to 4096 in all that time?

Yes we will get 5880 shaders with 40%, but I don't want to say that much. It sounds too unbelievable.

I was trying to show that even with fairly conservative estimates we could get 2 x Hawaii shaders [5120 = 2 x R9 290]as a possibility.

Also remember the possible HDL benefit.

Tick tock, soon we will know.

Gentlemen, bets anyone?

 

[crashtech]

I think 10% for memory controllers is awfully conservative. GDDR5 memory controllers are huge and complex. They should be able to get rid of at least half of those transistor when using HBM. Can you rerun your calculations based on the possibility of 40% reduction in the non-shader portion of the die? On a 600mm^2 die, we should see at least 500mm^2 of pure compute.

If you take the 292mm^2 shader portion of hawaii, port it to GloFo, the size becomes roughly 15% smaller, or 248.2 mm^2. Seems to me you could easily fit two of these on 500mm^2, which puts Fiji at a nice and even 5632. This makes too much sense to me. Fiji has to be essentially a 295x2 on one die, with enough memory bandwidth to feed it.

The 4096 shader part could very well be just the 390X. Hawaii is more than 18 months old. Why would anyone think they havent found a way to bump up from 2816 to 4096 in all that time?

There are many questions and possibilities especially now that we have seen that huge die. I don't want to get my hopes up, but from all the rough arithmetic I've been doing, it would seem as if Fiji would need to be HawaiiX2 to have a real chance at being a "Titan Killer." If the 4096 number instead turns out to be true, AMD will probably have to settle for second best, unless the HBM turns out to be a really radical game changer (excuse the pun).

 

[ShintaiDK]

Also remember the possible HDL benefit.

Also remember HDL doesnt come for free. Just look at Carizzo. HDL basically killed Carizzo for desktop in favour of Kaveri due to speed and power consumption at that target. Hence why Carizzo is limited to 35W and down.

 

[Tuna-Fish]

I think 10% for memory controllers is awfully conservative. GDDR5 memory controllers are huge and complex. They should be able to get rid of at least half of those transistor when using HBM.

Just a minor nitpick: GDDR5 controllers are not big because they have a lot of transistors. They are big because they need complex analog elements that take a lot of space and which do not shrink well or almost at all as you move to smaller processes.

 

[maddie]

If it only was that simple.

Remember that next time graphics cards goes up in price or they dont go to a lower node due to cost.

Shrinking dGPU volume, higher design cost, higher R&D cost.

Whats your solution? Because the current one isnt substainable.

Nobody says AMD should die. But from a business perspective only one of the two companies can survive in the long run until there is only IGPs. Then it doesnt matter who of AMD or nVidia survives. And if its AMD, then the next battle is who of AMD and Intel. If its nVidia, well, then they already lost the long run against Intel.

Remember that next time graphics cards goes up in price or they don't go to a lower node due to cost.

Are you actually saying that a monopoly is the best way forward because costs are rising?

Engineering/scientific breakthroughs are never seen before they happen. If you read anything historical, note the dismay at how progress has slowed/stopped due to insurmountable barriers. This happens each time a technology matures.

I believe that competition pushes innovation to surpass these seeming barriers. No competition allows a complancy with the status quo and steady sales and profits.

Nobody says AMD should die. But from a business perspective only one of the two companies can survive in the long run until there is only IGPs

This sounds a long diplomatic way to saying that AMD should die.

By the way you forgot to include the last part of my post.
Only a FOOL will want a monopoly, remember that the next time someone says "Die AMD".