Will the R9 Fury X turn out like the 290X and the 7970 or forever be slower than the 980 Ti?

[PlayStation 2]

Title.

If history decides to repeat itself again, the Fury X will probably end up faster than the 980 Ti, kinda like how the 290X beats the 780 Ti easily now.

 

What's your opinion?

[Kobathor]

Once Pascal comes out, yes. 

 

[AMDisTheBEST]

Title.

If history decides to repeat itself again, the Fury X will probably end up faster than the 980 Ti, kinda like how the 290X beats the 780 Ti easily now.

 

What's your opinion?

You should check out AMD firepro. It is as good in gaming as it is in a workstation PC.

[VagabondWraith]

Possibly, but I don't mind nVidia's planned obsolescence since I upgrade each gen anyway.

[Kinda Bottlenecked]

Only time can tell. But it may not be the case since the 980 Ti is exceptionally strong. And the Fury X doesn't exactly overclock as good as the 290x or 7970 once did. AMD will need to optimize for the Fury X pretty hard.

[byalexandr]

You should check out AMD firepro. It is as good in gaming as it is in a workstation PC.

Really? What's the equivalent to the Fury X? If it's faster than a Fury X I'd totally get one

[AMDisTheBEST]

Really? What's the equivalent to the Fury X? If it's faster than a Fury X I'd totally get one

It is amd's version of titanx. Not for consumer but it can game equally well.

[byalexandr]

It is amd's version of titanx. Not for consumer but it can game equally well.

What is it? The W9100?

[ivan134]

Title.

If history decides to repeat itself again, the Fury X will probably end up faster than the 980 Ti, kinda like how the 290X beats the 780 Ti easily now.

 

What's your opinion?

I doubt it. I honestly believe what we're seeing with the Fury X is as good as it's going to get. For the 290x vs 780 ti, the problem was AMD didn't know how to do drivers well. Now they do, and every card they have now is at their maximum. Both vendors have done pretty well for 28nm.

[Najuno]

Title.

If history decides to repeat itself again, the Fury X will probably end up faster than the 980 Ti, kinda like how the 290X beats the 780 Ti easily now.

 

What's your opinion?

I wrote this already in other threads, but I wouldn´t have too high hopes for the Fury X in the future. That counts for all current 28nm chips. When the next gen comes along it will be due to the die shrink of 14/16nm be far more powerful and feature HBM2 with 8-16GB VRAM.

The Fury X already struggles in 4K with frametime stuttering in games like GTA V. It was a couraged step from AMD to put HBM1 on their big chip, but IMHO the time wasn´t right because it is only HBM1. And AMD would have loved to put more than 4GB HBM on their cards, that I´m sure, because otherwise why would the outfit their smaller cards with 8GB GDDR5 then?

[Dackzy]

I wrote this already in other threads, but I wouldn´t have too high hopes for the Fury X in the future. That counts for all current 28nm chips. When the next gen comes along it will be due to the die shrink of 14/16nm be far more powerful and feature HBM2 with 8-16GB VRAM.

The Fury X already struggles in 4K with frametime stuttering in games like GTA V. It was a couraged step from AMD to put HBM1 on their big chip, but IMHO the time wasn´t right because it is only HBM1. And AMD would have loved to put more than 4GB HBM on their cards, that I´m sure, because otherwise why would the outfit their smaller cards with 8GB GDDR5 then?

As it stands rigth now Nvidia will not use HBM 2.0, just so you know.

[Najuno]

As it stands rigth now Nvidia will not use HBM 2.0, just so you know.

Source?

 

EDIT: Your statement is nonsense. nVidia and AMD will feature HBM2.0 with their highend cards made in 14/16nm next year. Only weaker mid range cards will feature GDDR5X. Have you any idea what it would mean for nVidia not to use HBM in the first place? That would have dramtically influence on the PCB, speaking of memory controller, Interposer, etc. In other words those Pascal cards would be incompatible with HBM and for a HBM card they would have to redesign the card... sounds a bit ridiculous.

[Dackzy]

Source?

 

EDIT: Your statement is nonsense. nVidia and AMD will feature HBM2.0 with their highend cards made in 14/16nm next year. Only weaker mid range cards will feature GDDR5X. Have you any idea what it would mean for nVidia not to use HBM in the first place? That would have dramtically influence on the PCB, speaking of memory controller, Interposer, etc. In other words those Pascal cards would be incompatible with HBM and for a HBM card they would have to redesign the card... sounds a bit ridiculous.

I think i heard it on tek syndicate, but when are the new AMD and nvidia card's come out? (I am looking to get a new one since my 780 ti broke)

[Najuno]

I think i heard it on tek syndicate, but when are the new AMD and nvidia card's come out? (I am looking to get a new one since my 780 ti broke)

Yes they announced that both, nVidia and AMD, will build GPUs in 2016 with GDDR5X memory. For mid range cards, this will be more than sufficient. But for the top Tiers of their time (I´m sure both will start with a smaller chip and sell this as highend and then 12 month later bring the big chips) they try to use HBM2.0. The memory is not only way faster than already HBM1.0, but also needs a lot less Wattage compared to GDDR5(X), which is a big plus and PCB size will be also smaller. Now don´t expect neither of these cards to be cheap. I mean look at the Fury X, AMD can´t really dump prices, because the R&D and manufacturing costs are too high.

I think the first nVidia card that features HBM next year will cost min. 700USD.

 

Here´s a recent source about HBM2.0

 

http://www.extremetech.com/gaming/218224-nvidia-unveils-pascal-specifics-up-to-16gb-of-vram-1tb-of-bandwidth

[MMKing]

If you're in the Nvidia x70, x80 or the AMD x90(x) segment. GDDR5X should be very interesting to you.

 

In theory, it should be twice as fast as regular old GDDR5. The way they aim to do this is to widen the buss width to 64 bit, as opposed to the current 32 bit per memory module. In theory this should allow for twice the transfer rate, on the same amount of memory modules.

 

If current GPUs had double the bandwidth.

390x = 768GB/s

980 = 448GB/s

 

Though i doubt they will be able to keep up the impressive memory frequency of 3.5-4GHZ per second, so a more realistic scenario is an initial increase of 50%, but room for incremental growth to around 75%. Though i have no data to support this claim.

 

Faster memory is not important for 1080p, as you will note when the very similar 970 and the 390 is pitted against each other. But at 1440p and especially 4k, the 970 drops more than the 390 in terms of %.

[NuclearPeace]

Nobody knows for sure because absolutely nobody knew that the GCN would have such an dramatic increase of performance over time. Maybe that can't happen anymore because they nicked out all the performance they can with recent drivers, maybe Maxwell will get faster over time, nobody really knows for sure. This is why its best to buy what's the fastest/what you want now and not rely on conjecture.

[Dackzy]

The fury x with the latest driver update, it beats the 980 ti but it's not by that much, and yeah i know you can overclock the shit out of some gtx 980 ti.

 

http://www.overclock.net/t/1578881/fury-x-is-now-just-as-fast-as-gtx-980ti-in-1080p-1440p-and-faster-in-4k

 

It doesn't really matter what card you get they are both great cards.

[ApolloX75]

Considering Micron has GDDR6 taped out for 2016 already, I'd say HMB is going to be a hard sell for mainstream and even high end cards. The info I read on GDDR6 is it has more than double the bandwidth of GDDR5 but uses the same footprint.

It's likely they will skip the whole GDDR5x scheme entirely and jump straight to 6. Hopefully AMD sticks to their guns and keeps up the use of HBM as its an extraordinary achievement. Just needs more refining.

[xAcid9]

Considering Micron has GDDR6 taped out for 2016 already, I'd say HMB is going to be a hard sell for mainstream and even high end cards. The info I read on GDDR6 is it has more than double the bandwidth of GDDR5 but uses the same footprint.

It's likely they will skip the whole GDDR5x scheme entirely and jump straight to 6. Hopefully AMD sticks to their guns and keeps up the use of HBM as its an extraordinary achievement. Just needs more refining.

^ Fake news, Micron only working on GDDR5X atm.

[Prysin]

Really? What's the equivalent to the Fury X? If it's faster than a Fury X I'd totally get one

Fiji is not meant for workstation workloads. It lacks double precision.... its just as bad as Maxwell when it comes to DP...

 

W9100 featuring an Hawaii XT core will remain the top end Firepro GPU until the launch of Arctic Islands.

[byalexandr]

Fiji is not meant for workstation workloads. It lacks double precision.... its just as bad as Maxwell when it comes to DP...

 

W9100 featuring an Hawaii XT core will remain the top end Firepro GPU until the launch of Arctic Islands.

What exactly is it? A super beefed up 290X or what?

[Prysin]

What exactly is it? A super beefed up 290X or what?

W9100 is a 290X with 16GB of VRAM, lower clock speed. cherry picked GPU for maximally low power usage (like super super super low voltage at slightly lower then stock reference 290X clock speeds).

 

end result...

Hawaii XT (290X) that has a TDP of 175-200W rather then 275W....

 

However, unlike the 290X, the W9100 has some features unlocked through drivers. Such as a 1/3rd Double Precision mode.....

Given how beastly a 290X is at single precision (32bit float), imagine how massive 1/3rd of that is...

Spoiler alert - The theoretical maximum compute of a W9100 workstation GPU is higher then that of a Quadro 6100 and Tesla K60. The most beastly CAD/3D workload / compute card Nvidia offers today....

 

However, due to most programs being heavily optimized for CUDA workloads rather then OpenCL. AMDs theoretical performance is a far cry from their actual performance. In general, AMD is trailing Nvidia. however if you were to unleash a W9100 at its full potential. It would be the most powerful workstation card there is.

[ApolloX75]

^ Fake news, Micron only working on GDDR5X atm.

 

Ahh, thanks for the clarification. Did a little more digging and you are correct.

[byalexandr]

W9100 is a 290X with 16GB of VRAM, lower clock speed. cherry picked GPU for maximally low power usage (like super super super low voltage at slightly lower then stock reference 290X clock speeds).

 

end result...

Hawaii XT (290X) that has a TDP of 175-200W rather then 275W....

 

However, unlike the 290X, the W9100 has some features unlocked through drivers. Such as a 1/3rd Double Precision mode.....

Given how beastly a 290X is at single precision (32bit float), imagine how massive 1/3rd of that is...

Spoiler alert - The theoretical maximum compute of a W9100 workstation GPU is higher then that of a Quadro 6100 and Tesla K60. The most beastly CAD/3D workload / compute card Nvidia offers today....

 

However, due to most programs being heavily optimized for CUDA workloads rather then OpenCL. AMDs theoretical performance is a far cry from their actual performance. In general, AMD is trailing Nvidia. however if you were to unleash a W9100 at its full potential. It would be the most powerful workstation card there is.

So it's just like all of AMD's other cards. Just needs improvement software wise.

[Prysin]

So it's just like all of AMD's other cards. Just needs improvement software wise.

pretty much

 

IT IS however, A LOT cheaper...

a W9100 costs around 3000 USD, whilst the Quadro K6000 is around 3100 and the M6000 closer to 5000...

 

So you generally get a REALLY good deal....