AMD Confirms Delay of R9 Fury X2/Gemini to 2016 for Better Alignment with VR Market

[patrickjp93]

What do they seem to be leaning towards in terms of the cluster replacement, if you can share? I am sure the testing is still early, but I am very curious.

A combination of KNL-exclusive nodes for scale-up workloads and some Skylake-EP/EX (iGPU vs. I FPGA-accelerated not yet decided) with Arctic Islands for scale-our due to the near 1.3TFLOPs DP compute advantage in real-world performance. For anyone who saw the Pascal DP article earlier this month, I was correct. Pascal is a 1/3 DP vs. SP performance architecture. It's 12.6 vs 4.2 for Pascal theoretically with 11.3 and 3.97 being about the best in RWP achieved under OpenCL 2.0. It's a little better under CUDA. Arctic Islands is 11.8 vs. 5.9 theoretically and comes out to 10.95 and 5.75 in RWP. KNL lives up to its theoretical numbers a bit better and is a better accelerator for AI training, but in raw compute the Xeon Phi version is already beaten handily. The socketed package form has no competitor anywhere on the market.

[Nurmi]

"to 2016 for Better Alignment with VR Market"

 

Any clues when and what is this VR market?

Mean aside from this routine fucking teams everywhere talk, hows that VR launch going to roll out like what goggles and games am I expecting to see?

[patrickjp93]

"to 2016 for Better Alignment with VR Market"

Any clues when and what is this VR market?

Mean aside from this routine fucking teams everywhere talk, hows that VR launch going to roll out like what goggles and games am I expecting to see?

Mostly the Oculus 2.0, but I think VR won't really be taking off until 2018. 2016 will still be too expensive.

[marldorthegreat]

Pascal is not around the corner. More like Q4 2016.

The Fury X2 will still be a beast after Pascal. After all, Pascal isn't some kind of magic GPU architecture that will be 200% faster than Maxwell.

no, itll be summer, not q4. 

[Prysin]

Well for one, @Prysin for damn sure is happy with his 295x2.

It may not be the most practical card, or even the smartest card to buy, but look at where the Titan X beats the 295x2. It takes two of them and OC'd and it BARELY edges out the 295x2. By 0.2 fps and that's the margin of error.

Just look. People who would rather have two separate GPUs are gonna go that route. But you have people like me who'd rather prefer one chunk of PCB.

Please ignore @Enderman

he is suffering from confirmation bias in quite some cases. Sure, anyone can find a single benchmark that proves their point. Finding more benchmarks that prove your point then those that disprove them however. That is more difficult.

 

people in general dont buy these cards.

 

BECAUSE THEY COST A FUCKING METRIC TONN...

 

but you see, @Enderman

if you had seen a R9 295x2 for 575 USD. Would you pass it up if you had THE BEST PSU ON THE MARKET, AND A CASE THAT SUPPORTS IT?

 

i get TitanX/980Ti performance, for 980 price.

Even when CF doesnt work, i have set GPU1 to be overclocked. So it will always perform better then stock R9 290X. Always.

 

Titan Z = 3000 USD at launch

R9 295x2 = 1500 USD at launch

Titan X = 999 USD at launch

980Ti/Fury X = 649 USD at launch

 

Most bought cards

R9 390/GTX 970 = 329 USD at launch

R9 380/GTX 960 = 229 USD at launch

 

 

the market for the "Enthusiast" level GPUs such as 980Ti and up, are just that. For Enthusiasts. You do not buy a 980Ti because you NEED it. you buy it because you CAN.

A R9 390/GTX 970 can power most games at high FPS with the right settings. you do not NEED a 980TI/Fury X to play a game. You WANT a 980Ti/Fury X to play a game at ULTRA SETTINGS.

[Starelementpoke]

So.. Because it produces heat, there's no reason why you should own it?

You should tell that to JayzTwoCentz with his SLI Titan X build (skunkworks.) It's watercooled, and spits all the CPU and GPU heat straight into his room. Guess he shouldn't do that setup, because "people don't want all that heat in their room."

 

1) you need to learn to read without making assumptions

i never said people should not buy it because it produces heat...wtf...where did you get that shit from

 

2) if you read with your eyes instead of your feet, you will notice that I'm talking about REASONS why MOST PEOPLE buy other graphics cards

and yes, its a fact that the 295x2 is a niche product and does not sell very well

 

3) heat was ONE out of SEVERAL points i mentioned

not because it produces heat, but because it produces UNNECESSARY heat

 

4) titan X SLI performs better and produces less heat, and since j2c is not limited to one GPU he went with a better option

make sense?

 

5) why have twice as much heat than other options?

 

6) why am i even arguing with an AMD fanboy? of course someone like you won't understand why most people dont buy dual GPU cards, I'll just stop arguing with a brick wall now.

 

Dual-GPU cards typically don't sell well for a variety of reasons, the best of them being heat and the power reasons.

Granted, the 295x2 might be one of the coolest running dual-GPU cards, but eh

Mom, Dad, Grandpa, Please stop fighting...

But on a serious note, Enderman, you still haven't made your point clear to me. What are you arguing exactly? That the 295x2 is a bad card because of the heat? That dual gpu's in general are bad? And on another note, isn't pascal only supposed to improve compute performance majorly? Or am I thinking about this wrong?

[Starelementpoke]

you've derailed this thread so far even YOU dont remember what the topic was

LOL

 

1) titan X = 250W TDP

290X = 290W

titan X SLI = less heat + more performance than 295x2

 

 

2) the ENTIRE POINT of this is that dual GPU cards sell very little because of their niche market, and delaying the fury X2 is hurting the sales because most people will be looking forward to the next AMD/nvidia GPU generation rather than "oh most powerful graphics card in the world for $1500 for the next year hurr durr"

 

so by waiting an extra 6 months for "better alignment with VR" they are going to be selling even fewer than they already would releasing it now

or maybe they're just using that an excuse instead of saying that its not ready yet

You aren't actually showing yourself to be much more than just a fanboy yourself you known either right?

[Doobeedoo]

Hmm, I really though they'd release it by the end of the year. Though I wonder how they'll price it being their new lineup is not so far off and all.

[Prysin]

You aren't actually showing yourself to be much more than just a fanboy yourself you known either right?

he doesnt know math and he doesnt know his facts

 

http://anandtech.com/show/7930/the-amd-radeon-r9-295x2-review

R9 290X is supposed to be 250w TDP, albeit real world power draw IS higher. (however, the electricity going in is NOT equivalent to the HEAT going out)

R9 295x2 is 500W TDP because the GPUs are cherry picked to have lower voltage consumption then normal 290Xs....

 

its sort of like R9 Nano... much lower power usage due to cherry picked GPUs.

[Prysin]

Pascal is Q2 2016. The HPC clients are already being sampled. Miami has KNL, AI, and Pascal samples already since we're looking at replacing our entire cluster.

<s>

And actually Pascal is 1000% the performance of Maxwell. CEO Jen Hsun Huang already said 10x performance gains.

</s>

1000% of barely anything = "finally decent"

[Enderman]

Please ignore @Enderman

he is suffering from confirmation bias in quite some cases. Sure, anyone can find a single benchmark that proves their point. Finding more benchmarks that prove your point then those that disprove them however. 

I have never seen a 295x2 for $600...

I have seen them for $800+

and I wouldnt buy one used

 

also they are just two 290Xs in crossfire

so a lot of stuff that doesn't perform well in SLI/crossfire will perform horribly or not perform at all, because iirc you cant just disable crossfire on a dual GPU card like you can on two separate GPUs

 

and i use shadowplay, so personally I wouldnt buy a last-gen GPU just because it gets a few fps higher performance

maybe you didnt know this, but not everyone is an "fps per dollar" gaming fanatic

 

either way, my point is that dual GPU cards are a nice market, few people buy them, and even fewer will buy the fury X2 if its released so late

 

Mom, Dad, Grandpa, Please stop fighting...

But on a serious note, Enderman, you still haven't made your point clear to me. What are you arguing exactly? That the 295x2 is a bad card because of the heat? That dual gpu's in general are bad? And on another note, isn't pascal only supposed to improve compute performance majorly? Or am I thinking about this wrong?

Did you not read the OP???

AMD is hurting sales by delaying the fury X2 becuase of "better alignment with vr"

like what kind of stupid excuse is that?

I think they're just using that as an excuse instead of saying "its not ready" and lowering their stock value even more

 

not that many people buy dual GPU cards in the first place, by releasing it closer to other generations they are losing that already small nice customer base

[MEC-777]

Thinking AMD should have maybe scrapped the 2x Fury card, or just simply release it when it's ready. By the time this launches (nearly a year since Fiji launch) Arctic Islands and Pascal won't be too far behind. I don't see how waiting to launch it a long side VR products will help sell more. not a good marketing strategy IMO.

[Notional]

Honestly I think the issue is two fold:

 

1. AMD has scrapped TSMC for Samsung/GloFo for Arctic Island. This is apparently due to poor yields and quality from TSMC. This might also mean there aren't enough chips to actually make this product in high enough numbers.
2. They might be postponing this product to await the courts judgement of the Asetek vs Cooler Master case. Releasing a product, just to see it be removed again is a bad idea.

Either way, it seems like Arctic Island cards won't be made until Q2, so expect a Q3 release in 2016. I highly doubt NVidia will be out sooner with Pascal. Maybe their quadro cards will see an earlier launch, but that is irrelevant for gaming or people wanting to spend less than 5000$ on a card.

[Prysin]

I have never seen a 295x2 for $600...

I have seen them for $800+

and I wouldnt buy one used

 

also they are just two 290Xs in crossfire

so a lot of stuff that doesn't perform well in SLI/crossfire will perform horribly or not perform at all, because iirc you cant just disable crossfire on a dual GPU card like you can on two separate GPUs

 

and i use shadowplay, so personally I wouldnt buy a last-gen GPU just because it gets a few fps higher performance

maybe you didnt know this, but not everyone is an "fps per dollar" gaming fanatic

 

either way, my point is that dual GPU cards are a nice market, few people buy them, and even fewer will buy the fury X2 if its released so late

 

Did you not read the OP???

AMD is hurting sales by delaying the fury X2 becuase of "better alignment with vr"

like what kind of stupid excuse is that?

I think they're just using that as an excuse instead of saying "its not ready" and lowering their stock value even more

 

not that many people buy dual GPU cards in the first place, by releasing it closer to other generations they are losing that already small nice customer base

 

Again, mate... i own this card. Ive been using AMD for a while...

And i can happily inform you that on most of your points.

 

YOU ARE WRONG!!

 

Well, you do not live in Norway and or you havent looked for the best sales

Mine was brand new out of the box. Box still had original, untampered seals.

 

They are two slightly overclocked 290Xs.

Reference 290X is 975MHz core clock with 1GHz boost. 295x2 runs at 1018 "core" clock... it doesnt really "boost". It runs any range between then 300MHz "idle" or 1018 "max".

CF can be disabled with dual GPU cards in crimson drivers. Game by game if you so want to. It was a bit more cumbersome to disable pre-crimson drivers. BUT IT WAS POSSIBLE.

 

Shadowplay is a fair argument. I have not used it myself, but every video/review ive seen of it indicates that it is simply better then what AMD has. Especially on the streaming side.

 

 

In terms of SLI/CF issues:

my HD 7950s using the CF Bridge had stuttering. Not a whole lot. but they had it. My 295x2 which uses a PLX chip as interconnect still have some microstutters, but much unlike the HD 7950s, my current issues is game by game. Not as a whole.

 

Some games will have more microstutters then others. It seems this is mostly tied to the post processing effects used in some titles.

I recently found out, by modding skyrim, that different implementations of Depth of Field and Adaptive Lightning can cause flicker/stutter on AMD (these are known issues). Whilst other similar effects, that just have a few minor tweaks mostly timing wise. Will NOT cause stutters or flickers.

 

This indicates to me that SOME. but by far not all stuttering issues with GCN 1.1 and GCN 1.2 cards (these cards uses the PCIe for CF rather then a bridge interconnect). Are mostly caused by timing/execution issues in the game code rather then drivers.

 

I say SOME because far from all games will have this issue, and saying AMD drivers are fine and without any fault would be a blatant lie.

[CTR640]

"to 2016 for Better Alignment with VR Market"

 

Any clues when and what is this VR market?

Mean aside from this routine fucking teams everywhere talk, hows that VR launch going to roll out like what goggles and games am I expecting to see?

It's clearl Lisa Su can't run the company for shit. She is helping AMD to go to bankrupt. "Better alignment with VR"...?? The fuck??

[ivan134]

They should just scrap it at this point.

[niofalpha]

 

its just two fury Xs

 

Don't it's specs match like 2 Fury Nano's/ normal Furys?

It'll be extremely highly binned Fury X, since it's rumored TDP is 375w and the Fury Nano TDP is less than 1/2 that (yes, I know Rumors. WCCF Tech, and TDP  != power consumption).

This said the Gemni GPU will have to be either Nano's, or a 140mm+ radiator since the Fury X runs at ~60 C, 2 of them would be probably be too much for that small of a radiator.

[Enderman]

Don't it's specs match like 2 Fury Nano's/ normal Furys?

It'll be extremely highly binned Fury X, since it's rumored TDP is 375w and the Fury Nano TDP is less than 1/2 that (yes, I know Rumors. WCCF Tech, and TDP  != power consumption).

This said the Gemni GPU will have to be either Nano's, or a 140mm+ radiator since the Fury X runs at ~60 C, 2 of them would be probably be too much for that small of a radiator.

yes, this was mentioned already later on in the discussion

[Guest Kloaked]

Doesn't matter what anyone says. It's releasing way too late. They may sell some units but they won't sell a lot at all.

[shawno]

Doesn't matter what anyone says. It's releasing way too late. They may sell some units but they won't sell a lot at all.

This.

By the time its released nvidias line up will be on the horizon and Arctic islands wouldnt be too far off.

[Prysin]

even with the die shrink, the Fury X2 will be in line of the next equivalent of 2x 980s or so...

[dimitriianghelov]

i wonder if this has something to do with amd not having good enough cpus to pair them with the powerful dual gpu and promote it

[Sharkyx1]

This said the Gemni GPU will have to be either Nano's, or a 140mm+ radiator since the Fury X runs at ~60 C, 2 of them would be probably be too much for that small of a radiator.

Well as a rad heats up, the Delta worth the outside air increases, making the rad more efficient.

It might be that fury X hits 60c 2 won't hit 120

295x2 runs under 75c remember and that's 500w

[HKZeroFive]

i wonder if this has something to do with amd not having good enough cpus to pair them with the powerful dual gpu and promote it

 

I don't think so. This is supposed to be the fastest video card ever and there wouldn't be any sense to try and promote it with a new CPU. I'm just wondering why they're delaying it because I don't buy the VR market alignment excuse.

[deviant88]

Pascal is not around the corner. More like Q4 2016.

The Fury X2 will still be a beast after Pascal. After all, Pascal isn't some kind of magic GPU architecture that will be 200% faster than Maxwell.

It will be 200% faster, it must be faster, after all 17billion transistor vs 8billion in 980ti, with HBM2 on top of that and other arhitecture improvements on 16nm it must be a huge leap else it will be a fail.

We need it for 4k and VR so they better not hold back on any performance.

From 28nm old outdated stuff to 16nmFF i have very high expectations.