Radeon FURY thread

[Jono8]

Not in NVidia's interests to do that.

They will want little Pascal to be about 20% faster than the GM200 cards and will achieve it by limiting the cores to the very minimum to do this on the GP204 chips. This in turn will increase their profit as the cards will be cheaper to make and will leave a nice big performance gap for big Pascal to clean up with.

Just like they did with the Maxwell cards.

Well I would class 20% faster than a Titan X a decent increase

I bet it will have about half the TDP as well what with the die shrink and HBM

 

[Davedree]

If AMD had got the styling right on the Tahiti cards like the HD7970 and got the drivers right on day one they would have wiped the floor with the GTX 670 and 680 giving them a much bigger market share.

The problem is always ''IF they had''with Amd, which is a shame. There's been a few missed opportunities like the 970 issue where Amd could have pounced but all they could do was a slight price drop on their inferior 290's.
Also maybe not so much previous gcards, but in particular this release feels desperate and unrefined almost like a paper launch.

I agree Tahiti was a good architecture as it provided a balanced dp compute and gaming performance on a medium sized die. Bar the stupid clocks and late wonder driver and then the missed opportunity with the mining craze (if they had ), i'm still to this day using my 7950. Fiji fp64 is 1/16 a big cut down from Hawaii and Tahiti, it's also a big ~600mm2 die which surely can't be cheap to produce with possible low yield rates, (hence all those tonga 285's) which is battling in between a gm204 398mm2 with only 2048 shaders, and 600mm2 gm200 with 2816 3072 shaders.


There's always something that so clumsily knocks them off their perch when it comes to the big review day for Amd. Hawaii was the cooler, Fury is the lateness, un prepared overclocking support and the slight cooler whine issue. By my calculations Fiji should have enough in the tank to compete with a 20% oc 980ti, but after 20% It becomes Nvidia's territory.

Performance for the Fiji is exactly where I predicted it would be, I predicted 3584 224 64 back in Sep 2014. It's not that Fiji is a bad card it performs reasonably well at 1440p and pretty good at 4k, it's just a year too late and overshadowed by Nvidia's Gm200 as they are more refined as a whole.

 

[Kaapstad]

Well I would class 20% faster than a Titan X a decent increase

I bet it will have about half the TDP as well what with the die shrink and HBM

Now you mention HBM I think the Titan X will still be faster @1080p using old fashioned GDDR5.

 

[Boomstick777]

I think you'll probably plump for a Pascal Titan when it's released from its cage

Hell, I might even treat myself to it!!!

Haha, take the word probably out mate. I'm def getting the highest end Pascal card. Got to try that myself. hopefully it comes early next year.

So do you think 980 Ti will remain on top until then?

 

[Jono8]

Now you mention HBM I think the Titan X will still be faster @1080p using old fashioned GDDR5.

No way. Pascal's equivalent of the 980 will be a good bit faster than a Titan X, I am sure of it.

 

[Boomstick777]

^^

All bets are off with Pascal. Nvidia made the Titan X / 980 Ti on 28nm with GDDR5.

Imagine what they can do with a die shrink, new architecture and HBM 2.0

 

[D.P.]

Not in NVidia's interests to do that.

They will want little Pascal to be about 20% faster than the GM200 cards and will achieve it by limiting the cores to the very minimum to do this on the GP204 chips. This in turn will increase their profit as the cards will be cheaper to make and will leave a nice big performance gap for big Pascal to clean up with.

Just like they did with the Maxwell cards.

There is more to it than just profits though.
he 16nmFF processes will be very expensive to begin with so they will want to make smaller chips, not jump in with the 600mm monsters. Also being a new process yields will be low so they will want smaller chips to raise yield rates.

This then allows them to make bigger chips a year later when the process is more mature and with all the lessons they have learned form the 104 part.

Also, die shrinks will be slow to come going forwards, i fully expect Pascal to follow in maxwell's foot steps and have 2 versions released at 16nm.

 

[Kaapstad]

No way. Pascal's equivalent of the 980 will be a good bit faster than a Titan X, I am sure of it.

It will be interesting how NVidia deal with HBMs low clockspeed or they will find themselves in the same situation as AMD using it on the Fiji cards. The low clockspeed kills 1080p performance but is not a problem @2160p where the fps are lower but the workload per frame is higher.

 

[Kaapstad]

There is more to it than just profits though.
he 16nmFF processes will be very expensive to begin with so they will want to make smaller chips, not jump in with the 600mm monsters. Also being a new process yields will be low so they will want smaller chips to raise yield rates.

This then allows them to make bigger chips a year later when the process is more mature and with all the lessons they have learned form the 104 part.

Also, die shrinks will be slow to come going forwards, i fully expect Pascal to follow in maxwell's foot steps and have 2 versions released at 16nm.

Fully agree with all the above.

 

[Scougar]

^^

All bets are off with Pascal. Nvidia made the Titan X / 980 Ti on 28nm with GDDR5.

Imagine what they can do with a die shrink, new architecture and HBM 2.0

Get low performance at 1080p and wait for other people to criticize?

EDIT: Dangit... Kaap beat me to it with the non-tongue n cheek version of my comment. D'oh!

 

[Ash0787]

^^

All bets are off with Pascal. Nvidia made the Titan X / 980 Ti on 28nm with GDDR5.

Imagine what they can do with a die shrink, new architecture and HBM 2.0

thats what concerns me, nvidia seems to have an engineering edge which amd were able to compensate for by introducing HBM, but now they dont have that card up their sleeve anymore I'm concerned it could be game over for amd, which is partly why im trying to support them this time around

 

[Kaapstad]

thats what concerns me, nvidia seems to have an engineering edge which amd were able to compensate for by introducing HBM, but now they dont have that card up their sleeve anymore I'm concerned it could be game over for amd, which is partly why im trying to support them this time around

Providing people more to higher resolutions it is AMD who will have the engineering edge with HBM. HBM2 with higher capacity memory will be very interesting at high resolution.

 

[Scougar]

I think the reviews were pretty good for the Fury (Pro). What doesn't make sense is the price of the 390/390x now though.

 

[Telecaster]

hopefully it comes early next year.

Really? You honestly think it will be so soon after all this wait and kerfuffle we've had waiting for Big Maxwell?

I think you're looking at 12 months at the very earliest, I reckon they'll want to milk 28nm for all it's worth, especially now that AMD have shown their hand and it's a bit meh.

If Fury had turned out to be a monster then maybe you'd have seen Pascal sooner, but now they've seen it I think Nv will be kicking it back with a cigar on and taking their time, maybe we'll even see a 990 yet?

 

[Boomstick777]

thats what concerns me, nvidia seems to have an engineering edge which amd were able to compensate for by introducing HBM, but now they dont have that card up their sleeve anymore I'm concerned it could be game over for amd, which is partly why im trying to support them this time around

AMD will have a new architecture and a die shrink as well as HBM 2.0. I imagine they will have something very good as well.

As for supporting AMD, I do try lol. I bought 7970's, 290X's and an AMD Fury X, but that pump noise killed it for me. Didn't want to risk trying another one and the Fury (Non X) basically Gibbo's comment means only card I want to try next is custom 980 Ti.

I'll have another go with AMD card at some point, I wont be so keen to be an early adopter though

Really? You honestly think it will be so soon after all this wait and kerfuffle we've had waiting for Big Maxwell?

I think you're looking at 12 months at the very earliest, I reckon they'll want to milk 28nm for all it's worth, especially now that AMD have shown their hand and it's a bit meh.

If Fury had turned out to be a monster then maybe you'd have seen Pascal sooner, but now they've seen it I think Nv will be kicking it back with a cigar on and taking their time, maybe we'll even see a 990 yet?

I wrote 'hopefully' I've got no idea when it will actually arrive.

I do hope big Pascal comes early next year though.

 

[Kaapstad]

 

[bru]

One thing worth thinking about, yes I'm sure NVidia could come out next year with a an absolute killer GP200 card and absolutely wipe the floor with whatever AMD could bring to the table, looking at their currant bus width and TDP advantage. But is it in their best interest to do that, I'm not sure that it is.

 

[ALXAndy]

One thing worth thinking about, yes I'm sure NVidia could come out next year with a an absolute killer GP200 card and absolutely wipe the floor with whatever AMD could bring to the table, looking at their currant bus width and TDP advantage. But is it in their best interest to do that, I'm not sure that it is.

I missed the currant bus last week. I was raisin hell.

 

[D.P.]

One thing worth thinking about, yes I'm sure NVidia could come out next year with a an absolute killer GP200 card and absolutely wipe the floor with whatever AMD could bring to the table, looking at their currant bus width and TDP advantage. But is it in their best interest to do that, I'm not sure that it is.

Its the same with Intel and CPUs.

 

[Orangey]

I'm facebook friends with the Sultana of Brunei