• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

The next Nvidia Flagship card?

Is there a trend to releases?

Or do we think they will just try and destroy AMD before they can fix HBM or even get there next gen out proper?

Don't we kinda need another maxwell anyway? There where two keplers wherent their? 6xx and 7xx. Was it the same with 4xx and 5xx?
 
Kaapstad said:
I also think it is a safe bet that Pascal will be on the market before Arctic Islands.
Click to expand...

Based on what??

AMD has rebranded almost their entire range and on top of this we now know they cancelled their plans to use 20NM too.

Tonga is only available for Apple in fully enabled form,and the noise is that it was probably designed more for Apple to put in their iMacs anyway.

The rest of the range was launched years ago.

Fiji is not even in their professional card range,as it has worse DP compute than Hawaii and cannot do more than 4GB of RAM anyway.

Fiji looks more and more like a test for HBM integration, and I expect with a cheaper and smaller GPU it might have not been worth the R and D effort if HBM volumes are limited.

Considering they have experience of making largish GPUs with both GlobalFoundries and TSMC,if anything the evidence probably is pointing towards them trying to get to 14NM/16NM before Nvidia since it would give them a competitive advantage.

Unless GlobalFoundries fraks up in their timings,I expect AMD to use them and Nvidia will use TSMC.

D1craig said:
i cant see nvidia waiting until Q1 2016 to release another series of gpus.
Click to expand...

They are going to be more limited by what TSMC and GlobalFoundries can do with 14NM/16NM yields and costs.
 
Last edited:
whyscotty said:
I'd can't see that being true some how.
Click to expand...

Same here, I thought they had it this time hence why there the only ones with it, Also I read that they had a years leeway after which time Nvidia would be using it. Plus we are well into that year.

It's all rumours though, None of us really know.
 
If Nvidia are supposed to be using HBM2 by Q1 2016 then that would mean it is ready in which case AMD would use it before that. You guys are forgetting the Fury X2 which is already up and running. Once HBM2 is ready I would think AMD would use it for that than wait for Arctic Islands.
 
Kaapstad said:
Fiji has only just come to market 9 months after Maxwell, NVidia have not been idle since then.

Pascal is also ahead in development compared to Arctic Islands.
Click to expand...

Based on what?? Show me the dev roadmaps for Pascal and Arctic Islands then?

But using that logic,Nvidia should have launched the GTX680 before the HD7970 since the GTX580 was out two months before the HD6970. Oh wait...!

The HD6970 was the same situation - 32NM was cancelled and it was pretty much a 32NM product and it had a short lifespan and they rejigged it for 40NM.

Fiji is a scaled up Tonga with HBM which came out last year with some of the new power saving technology from the Carrizo APU.

Maxwell wasa far greater change in uarch from Kepler than GCN1.1 to GCN1.2 AFAIK.

The rest of the range is rebrands and refreshes of GPUs which came out years ago,plus they cancelled any plans for 20NM.

If they were intending to stay on 28nm for a long time,they would refreshed the entire range below Fiji with newer GPUs.

These are the bread and butter of the company,not Fury.

So it looks more like they decided to cut their losses on 28NM and launch the only GPU which made sense to them,ie, Fiji since it gave them experience with HBM,large scale production of interposers and GPU,HBM and interposer integration,and then prioritise on the next generation.

IMHO,Fiji will not be a large volume product. The fact it probably won't make the professional markets alone which are high profit margin is another indication of this.

Unless you happened to forget this:

http://wccftech.com/amd-launches-fi...ddr5-memory-grenadas-524-tflop-compute-power/

That launched in the same time-frame as Fury.

Fiji won't be able to have more than 4GB of RAM and it will not have decent DP performance anyway,so at this point in the highest profit margin market AMD serves,its useless. The only commercial use maybe if Apple uses it for one of its workstations for OpenCL stuff,but again they could just as easily use Hawaii for that.

Its more likely they will push to get to a new node as quickly as possible and considering they have already made a large 330MM2 PS4 SOC on both GF and TSMC processes which is mostly GPU,which is something you might not have realised. That means they already have the tools ready to make their designs more portable. This means they can probably hedge their bets a bit more next generation in case one of the companies screws up again.

Personally I would be surprised to see any largish gaming GPU in volume from either company before June 2016 on a new process node.

I could see another GTX750TI type card as a pipecleaner and even if Big Pascal were to get there in Q2 2016 I expect almost the entire production will be for commercial customers due to Xeon Phi.

Considering that the GTX980TI only came out last month,I would find it highly unlikely that in six to nine months it will be replaced.

Even the GTX480 was only replaced by the GTX580 which was a tweaked version of the same GPU.

It still meant the GF100/GF110 had a two year lifespan.

Placing the GM200 as be EOL within 6 to 9 months would be somewhat of a record for a new uarch Nvidia GPU launched in the last 8 years or so.

You might be right,it could happen.
 
Last edited:
D1craig said:
Is there a trend to releases?

Or do we think they will just try and destroy AMD before they can fix HBM or even get there next gen out proper?

Don't we kinda need another maxwell anyway? There where two keplers wherent their? 6xx and 7xx. Was it the same with 4xx and 5xx?
Click to expand...

There are 2 maxwell architectures:
V1 is used in the 750Ti and a load of 800 series mobile parts.
v2 is used in the 900 series parts.


Also, maxwell is a very forward looking architecture . Pascal will be very similar to maxwell but scaled up and using HBM. Nvidia put in huge R&D repair CES into maxwell because they expect it to have a long life time, right through Volta (pascal successor). They needed something low power for the continued 28nm process node, and that really shows in the 980Ti vs FuryX performance per watt, especially if you add 30-40w savings for the HBM memory and the 20+we saved by running it under water. Nvidia would probably be close to twice the performance per watt for the whole card if they used HBM versus Fiji.
 
D.P. said:
Nvidia would probably be close to twice the performance per watt for the whole card if they used HBM versus Fiji.
Click to expand...

Nope.

This is an air-cooled Strix Fury:

http://tpucdn.com/reviews/ASUS/R9_Fury_Strix/images/perfwatt_3840.gif
http://tpucdn.com/reviews/ASUS/R9_Fury_Strix/images/perfwatt_2560.gif

Its within the same ballpark as a GTX980TI for performance/watt.

If anything the AIO water cooler is adding power consumption to the set-up and TPU actually measures card power consumption not at wall figures. Even adding 30W to 40W to the Strix figure,probably wouldn't make a GTX980TI twice as efficient.

HBM saves power,but more importantly it saves a decent amount of die area,as there is a massive reduction in memory controller area in the GPU.

This means more space for other things like caches,more shaders,etc.

Edit!!

Looking at the Fury X power consumption figures adding 40W to the load power,would be a 15% increase in power consumption.

Even that does not make the GTX980TI twice as efficient.

Are you taking card figures running Furmark or something?? :confused:
 
Last edited:
CAT-THE-FIFTH said:
Based on what?? Show me the dev roadmaps for Pascal and Arctic Islands then?

But using that logic,Nvidia should have launched the GTX680 before the HD7970 since the GTX580 was out two months before the HD6970. Oh wait...!

The HD6970 was the same situation - 32NM was cancelled and it was pretty much a 32NM product and it had a short lifespan and they rejigged it for 40NM.

Fiji is a scaled up Tonga with HBM which came out last year with some of the new power saving technology from the Carrizo APU.

Maxwell wasa far greater change in uarch from Kepler than GCN1.1 to GCN1.2 AFAIK.

The rest of the range is rebrands and refreshes of GPUs which came out years ago,plus they cancelled any plans for 20NM.

If they were intending to stay on 28nm for a long time,they would refreshed the entire range below Fiji with newer GPUs.

These are the bread and butter of the company,not Fury.

So it looks more like they decided to cut their losses on 28NM and launch the only GPU which made sense to them,ie, Fiji since it gave them experience with HBM,large scale production of interposers and GPU,HBM and interposer integration,and then prioritise on the next generation.

IMHO,Fiji will not be a large volume product. The fact it probably won't make the professional markets alone which are high profit margin is another indication of this.

Unless you happened to forget this:

http://wccftech.com/amd-launches-fi...ddr5-memory-grenadas-524-tflop-compute-power/

That launched in the same time-frame as Fury.

Fiji won't be able to have more than 4GB of RAM and it will not have decent DP performance anyway,so at this point in the highest profit margin market AMD serves,its useless. The only commercial use maybe if Apple uses it for one of its workstations for OpenCL stuff,but again they could just as easily use Hawaii for that.

Its more likely they will push to get to a new node as quickly as possible and considering they have already made a large 330MM2 PS4 SOC on both GF and TSMC processes which is mostly GPU,which is something you might not have realised. That means they already have the tools ready to make their designs more portable. This means they can probably hedge their bets a bit more next generation in case one of the companies screws up again.

Personally I would be surprised to see any largish gaming GPU in volume from either company before June 2016 on a new process node.

I could see another GTX750TI type card as a pipecleaner and even if Big Pascal were to get there in Q2 2016 I expect almost the entire production will be for commercial customers due to Xeon Phi.

Considering that the GTX980TI only came out last month,I would find it highly unlikely that in six to nine months it will be replaced.

Even the GTX480 was only replaced by the GTX580 which was a tweaked version of the same GPU.

It still meant the GF100/GF110 had a two year lifespan.

Placing the GM200 as be EOL within 6 to 9 months would be somewhat of a record for a new uarch Nvidia GPU launched in the last 8 years or so.

You might be right,it could happen.
Click to expand...

So basically you reckon AMD's Fiji will be a short lifespan part and they will go full steam ahead to bring us a new line up next year. But NVidia wont do that because they have only just launched the 980ti.

Well that just makes no sense at all, NVidia has large R&D budget, and launched first wont make the 980ti a short lifespan card, and yet AMD, that has a much smaller R&D budget and launched later will make the Fiji a short lifespan card.

It might happen but it is just as likely to be the other way around or that they will both bring new cards to the table next year.

IMHO,Fiji will not be a large volume product.
Click to expand...

That bit did make me laugh though, especially considering the stock levels of the FuryX and very few buying the Fury nonX. ( just look at the ROH in the owners thread.)
 
bru said:
So basically you reckon AMD's Fiji will be a short lifespan part and they will go full steam ahead to bring us a new line up next year. But NVidia wont do that because they have only just launched the 980ti.

Well that just makes no sense at all, NVidia has large R&D budget, and launched first wont make the 980ti a short lifespan card, and yet AMD, that has a much smaller R&D budget and launched later will make the Fiji a short lifespan card.

It might happen but it is just as likely to be the other way around or that they will both bring new cards to the table next year.



That bit did make me laugh though, especially considering the stock levels of the FuryX and very few buying the Fury nonX. ( just look at the ROH in the owners thread.)
Click to expand...

What's even quite funny is that Titan,is,GK110 launched 5 months before for commercial customers before the Titan was released.

Look at the AMD cards - gaming versions have always come before commercial versions.

So yes they will both have cards by the summer next year but Nvidia is more worried about 14nm Xeon Phi. I didn't say that Nvidia wouldn't be launching new node cards next year - that's you saying that not me.

But kaapstad and others are making the claims Nvidia will be well before AMD going to a new node - so where are these new roadmaps from both companies?

Also, Nvidia has launched a full top to down range and AMD whereas Nvidia has not - even the GM200 is a Quadro part with 12gb of VRAM.

There is not going to be s corresponding firepro part due to it only having 4gb of VRAM and yet people forget that the gtx580 launched before the hd6970 and the grx680 came after the Hd7970. The only possible way it would happen is if Apple expresses an interest but HBM supplies are constricted and hbm2 is not far off AFAIK.

Plus when is the last time nvidia first moved to a new process node first since at least 2005?? Even if they did they are going to prioritise commercial customers first unless they do an HD4770 like pipe cleaner first.

Historically AMD has tended to 9/10 move to new nodes first whereas Nvidia tends to maximise their performance on a given node which is probably also why they have done better at the end of 28nm.


Nvidia has tended to spend AMD in graphics r and d considering how AMD has to simply cover more areas.


So you talk about r and d - so if AMD has a smaller r and d budget what do you think they will prioritise with a smaller marketshare?


A whole range of 28nm cards with a short lifespan,or cards made on a new node which might them more competitive?


Have you heard squat all about Nvidia wanting to use 20nm? AMD did and cancelled any products on it to push forward to the next nodes.

And you can laugh what you want about the volume thing since one if the AMD reps basically said HBM was the limiting factor in availability.

Plus considering that only two Fury models launched and only one is actually available I think stock is not really great either.

So for a big volume part it sure looks like it will be constrained,right?

AMDJoe said:
Limited stock is due to limited HBM supply
Click to expand...
 
Last edited:
Last edited:
You must log in or register to reply here.
Back
Top Bottom