AMD Fury X Postmortem: What Went Wrong?

[AtenRa]

Updated ISA instruction set. That one can easily be split out. Just like the 2 renames for the scalar and UVD.

http://www.anandtech.com/show/8460/amd-radeon-r9-285-review/2

This is GCN 1.2. Sounds familiar?

If im not mistaken all those are over the GCN 1.2

 

[Silverforce11]

I don't think he is hyperventilating. Is that how you see him while reading his responses? I see him severely tired of the "IF" word being used to somehow makes AMD's situation lighter or show hope. If if if. It's kind of enough enough enough already.
There is always room for driver improvements for any architecture, however many feel this is just a larger Hawaii GPU, and some say it's a large Tonga GPU. If that is the case, then the drivers are probably pretty mature right about now. Maybe, and I use the term loosely because it reminds of "if", better memory management in the drivers is possible as HBM is relatively new. I can see room for improvement there. Of course this could be me totally overestimating the need for bandwidth. It could have all it ever needs and all it could ever take advantage of as of right now. Don't know.

I'm not talking about any of that stuff in the if if if statement.. I was responding to amenx regarding his comparison on Guru3d's OC test. I just presented the numbers because amenx felt the scaling wasn't good but its on par with their 75mhz OC.

Are you people aware that AMD GPU OC needs manual vcore control to be decent? Look at the recent 390/X, they behave the same as 290/X, with vcore they can hit 1.2ghz, but without it, nope, barely anything.

I'm going to end with this statement: AMD messed up without offering vcore mod tools to reviewers for the launch review. They didn't even use their major advantage with the water cooler, more thermal headroom at lower noise.

 

[ShintaiDK]

If im not mistaken all those are over the GCN 1.2

Unless you can find some edvidence. Then you are clearly mistaken.

Fury is GCN 1.2.

 

[Jaydip]

From a pure performance standpoint, I wonder if Tonga was a step backwards? If this was a tweaked Hawaii with HBM and 50% more everything, it likely would be better. We saw the 285 exhibit erratic performance and this seems to follow suit. Sometime amazing, sometimes meh. Just hypothesizing...

If you look at the various compute benches it seems Fury does exceptionally well there, I believe they designed Fiji to compete with NV in the pro space.You seriously don't make a ~9B xtor monster to tackle only games. Anyone knows when the pro version of this coming out?

 

[flopper]

the heat they got was due to mismanage the expectations.
no voltage control for review hurt them as I expected one to be there.
driver not mature enough was and is obvious should be better in coming iterations which also should been managed better.
etc...
great card just not optimal atm.

 

[YBS1]

Why oh why do people continue posting benchmarks that border on unplayable (~40 fps average, minimums will be much lower of course) to support their arguments? I have to imagine at least in general the majority of people who buy $500+ GPUs find 40fps unacceptable? Am I wrong? Neither of these cards are really suitable for 4K without multi card solutions.

 

[Silverforce11]

It's GCN 1.2 with some new functions, AMD is still calling it 3rd gen GCN.

"Also, the front-end of Fiji corresponds to Tonga. It has even been adopted 1: 1, although twice as large back-end to be supplied with data. It's like Hawaii designed fourfold and brings introduced by Tonga optimizations with it. AMD confirmed that that can be a bottleneck in low resolutions, in which the back-end is less demanded. Was developed in high resolutions, for Fiji, however, is also the smaller and towards Hawaii significantly improved front-end of Tonga sufficient to supply the shader units with the required data. The relationship to Tonga suggests: Fiji is once again on the "Third Generation of GCN" and thus GCN 1.2 and not 1.3 or even GCN GCN 2.0. Thus, the chip supports DirectX 12 with the feature level 12_0."

http://www.computerbase.de/2015-06/amd-radeon-r9-fury-x-test/

The biggest improvement is actually related to these:

Tremendous improvements in perf/w for these tasks as well as Compute tasks.

In fact, it's quite an achievement to have a chip so powerful at compute that manages to trade blows against a pure gaming chip, at the same die size and similar TDP.

Sadly, it's priced too high for the consumer variant and the gimped 4GB HBM will make it unwanted for HPC.

 

[AtenRa]

I'm going to end with this statement: AMD messed up without offering vcore mod tools to reviewers for the launch review. They didn't even use their major advantage with the water cooler, more thermal headroom at lower noise.

If im not mistaken there are Custom WC Furys on the way.

 

[Silverforce11]

If im not mistaken there are Custom WC Furys on the way.

Doesn't really matter, they messed up. Initial impressions stick. Remember the hot & noisy R290/X?

Now most gamers will think Fury X can't OC at all or its vcore locked permanently.

 

[railven]

If im not mistaken there are Custom WC Furys on the way.

Any idea on ETA and changes they may bring? I'm ready to buy a GPU now, I can't stand the noise levels from my GTX 680.

I was so ready to buy Fury X, but now that it's out of stock and I have to wait, I'm really losing that fanboy luster for it.

That 980 Ti G1 looks far more appealing all of a sudden.

Doesn't really matter, they messed up. Initial impressions stick. Remember the hot & noisy R290/X?

Now most gamers will think Fury X can't OC at all or its vcore locked permanently.

Yerp, this guy gets it. Fury X continues the AMD trend of breaking my heart.

 

[YBS1]

If im not mistaken there are Custom WC Furys on the way.

As in custom pcb Fury, or as in custom card and block only add it to your existing loop Fury?

 

[monstercameron]

I'm not talking about any of that stuff in the if if if statement.. I was responding to amenx regarding his comparison on Guru3d's OC test. I just presented the numbers because amenx felt the scaling wasn't good but its on par with their 75mhz OC.

Are you people aware that AMD GPU OC needs manual vcore control to be decent? Look at the recent 390/X, they behave the same as 290/X, with vcore they can hit 1.2ghz, but without it, nope, barely anything.

I'm going to end with this statement: AMD messed up without offering vcore mod tools to reviewers for the launch review. They didn't even use their major advantage with the water cooler, more thermal headroom at lower noise.

It's obviously a smart play on amds end, what's worse, high power usage and heat or no voltage mods?

Amd made the right choice, hell the card has dual bios for flashing. Once the tools get updated then the overclockers can have at it without the pressworks confusing the issue with heat and power distractions.

 

[Enigmoid]

Are you people aware that AMD GPU OC needs manual vcore control to be decent? Look at the recent 390/X, they behave the same as 290/X, with vcore they can hit 1.2ghz, but without it, nope, barely anything.

I'm going to end with this statement: AMD messed up without offering vcore mod tools to reviewers for the launch review. They didn't even use their major advantage with the water cooler, more thermal headroom at lower noise.

The power usage will jump massively with any voltage changes.

It's GCN 1.2 with some new functions, AMD is still calling it 3rd gen GCN.

"Also, the front-end of Fiji corresponds to Tonga. It has even been adopted 1: 1, although twice as large back-end to be supplied with data. It's like Hawaii designed fourfold and brings introduced by Tonga optimizations with it. AMD confirmed that that can be a bottleneck in low resolutions, in which the back-end is less demanded. Was developed in high resolutions, for Fiji, however, is also the smaller and towards Hawaii significantly improved front-end of Tonga sufficient to supply the shader units with the required data. The relationship to Tonga suggests: Fiji is once again on the "Third Generation of GCN" and thus GCN 1.2 and not 1.3 or even GCN GCN 2.0. Thus, the chip supports DirectX 12 with the feature level 12_0."

http://www.computerbase.de/2015-06/amd-radeon-r9-fury-x-test/

The biggest improvement is actually related to these:

Tremendous improvements in perf/w for these tasks as well as Compute tasks.

In fact, it's quite an achievement to have a chip so powerful at compute that manages to trade blows against a pure gaming chip, at the same die size and similar TDP.

Sadly, it's priced too high for the consumer variant and the gimped 4GB HBM will make it unwanted for HPC.

Musemage - barely better than the 390X

CompubenchCL - wins some loses some generally equal to the 980 ti.

F@H - worse than 980 ti in SP, better in DP (but worse than 390X).

Luxmark - barely better than the 390X, wins against the 980 ti.

Compute is a mixed bag. Its certaintly not much of an improvement and downright disappointing when comparing to the 390X.

 

[Silverforce11]

I was so ready to buy Fury X, but now that it's out of stock and I have to wait, I'm really losing that fanboy luster for it.

That 980 Ti G1 looks far more appealing all of a sudden.

Yerp, this guy gets it. Fury X continues the AMD trend of breaking my heart.

For gaming, there's no reason to pay $650 for Fury X when good custom 980Ti are $699. The gap OC v OC is massive.

I wouldn't buy it at that price, needs a price cut. Needs vcore mod tools last week for the press launch, its a joke their marketing people have no freaken clue, I could do a better job at it.

 

[Azix]

If you look at the various compute benches it seems Fury does exceptionally well there, I believe they designed Fiji to compete with NV in the pro space.You seriously don't make a ~9B xtor monster to tackle only games. Anyone knows when the pro version of this coming out?

I think the compute power does come into play later for gaming.

http://www.redgamingtech.com/asynch...eir-role-on-ps4-xbox-one-pc-according-to-amd/

 

[AtenRa]

Any idea on ETA and changes they may bring?

As in custom pcb Fury, or as in custom card and block only add it to your existing loop Fury?

Sorry, I dont have all this info yet.

 

[Azix]

If this does materialize AMD should advocate a complete retesting of their entire line-up. Since they were thinking ahead as far back as GCN 1

Unfortunately the ACE’s weren’t very useful in the DX11 API (because the API were incapable of running Compute and Graphics simultaneously), but despite this AMD decided to still add in the additional ACEs, believing they’d be useful in the future.

 

[Silverforce11]

If this does materialize AMD should advocate a complete retesting of their entire line-up. Since they were thinking ahead as far back as GCN 1

Useful for consoles since their lifespan on one uarch is so long, but PC gamers tend to upgrade every 2/3 years? More for enthusiasts who like the best constantly.

Still, with how well GCN mature, I have no reason to doubt that GCN will continue to improve further and especially with DX12's similarities with Mantle.

 

[Keysplayr]

Doesn't really matter, they messed up. Initial impressions stick. Remember the hot & noisy R290/X?

Now most gamers will think Fury X can't OC at all or its vcore locked permanently.

But why won't they learn this? What is blocking them from learning this concept that you can't do things half arsed? No o/c tools. Maybe the thing is an overclocking monster yet to be unleashed. Do you really believe that?
But more importantly, does AMD want to fail? I just can't get my mind around launching something that isn't ready when it (possibly) could have been a much better introduction. You're right, first impressions are emblazoned in peoples minds for the first showing. It's over for 90+ % (guess) of people as most of what I've read across forums is disappointment with a side order of disbelief that AMD would actually release this card in this state. Un-ready? Or, the best they could do?
You don't know how many posts I've read that read, "I was waiting for the Fury review to purchase one at launch, but now I'm just going with 980Ti.

AMD's fault. They should have waited for better drivers (assuming they aren't already tapped out and this is just a glorified Tonga), and properly functioning and extensive o/c tools.

 

[tviceman]

Hyperventilating much?

Everything I posted in that quote is spot on with how GCN behaves with & without vcore OC. Found anything wrong with that assessment?

I also ended by saying it needs a price cut to be competitive.

Grasping at straws much? Unlocked Vcore wouldn't have done too many favors (if any) for Fury X. I highly, highly doubt it'll get much more than 15% OC and even then power consumption will hit thermonuclear conditions. It won't be able to match GM200's OC'ing, so it was probably in their best interest to just mask this aspect and avoid the entire label of "worse overclocker AND polar bear raper."

Fury X was hyped up to be the next coming of Jesus. At the end of the day, it couldn't beat a cutdown GM200, let alone be price competitive. Anyone who thinks vanilla Fury is the card to own needs step back and see that Nvidia can easily release a further cutdown GM200 at the same price as vanilla Fury and just continue raining red tears down on AMD's Fiji.

Doesn't really matter, they messed up. Initial impressions stick. Remember the hot & noisy R290/X?

Now most gamers will think Fury X can't OC at all or its vcore locked permanently.

Exactly. You're speaking some sense now.

For gaming, there's no reason to pay $650 for Fury X when good custom 980Ti are $699. The gap OC v OC is massive.

Try $670-680
http://www.newegg.com/Product/Product.aspx?Item=N82E16814487142
http://www.newegg.com/Product/Product.aspx?Item=N82E16814487141
http://www.newegg.com/Product/Product.aspx?Item=N82E16814500378
http://www.newegg.com/Product/Product.aspx?Item=N82E16814127889

The EVGA SC+ is 21% faster than Fury X at 1440p out of the box.

 

[Azix]

Useful for consoles since their lifespan on one uarch is so long, but PC gamers tend to upgrade every 2/3 years? More for enthusiasts who like the best constantly.

Still, with how well GCN mature, I have no reason to doubt that GCN will continue to improve further and especially with DX12's similarities with Mantle.

There are tons of people on older GCN hardware that it would help. And it would continue to help on hardware going forward. Current hardware stands to benefit as well. Good all around I think.

That tiny GPU in the PS4 is putting out so much with 1.8tflops. The closer gaming PCs get to working like consoles the better for us. If dx12 delivers PC gaming could really finally be far ahead of consoles as it should be and on AMD hardware they've been ready. All previous benchmarks would have to be tossed

 

[Silverforce11]

But why won't they learn this? What is blocking them from learning this concept that you can't do things half arsed? No o/c tools. Maybe the thing is an overclocking monster yet to be unleashed. Do you really believe that?
But more importantly, does AMD want to fail? I just can't get my mind around launching something that isn't ready when it (possibly) could have been a much better introduction. You're right, first impressions are emblazoned in peoples minds for the first showing. It's over for 90+ % (guess) of people as most of what I've read across forums is disappointment with a side order of disbelief that AMD would actually release this card in this state. Un-ready? Or, the best they could do?
You don't know how many posts I've read that read, "I was waiting for the Fury review to purchase one at launch, but now I'm just going with 980Ti.

AMD's fault. They should have waited for better drivers (assuming they aren't already tapped out and this is just a glorified Tonga), and properly functioning and extensive o/c tools.

You could have asked the same thing when the R290/X launched. They learnt NOTHING at all from the horribly noisy 7970 and especially the 7970Ghz Ed.

They also learnt nothing from mistakes of NV with Fermi. Seriously who thinks 94C on a GPU is a good idea for consumers?

AMD is pure fail when it comes to marketing & management. I was hoping Lisa Su may change course but releasing Fury X with a water cooler with ample thermal room WITHOUT vcore mod tools is idiotic.

@tviceman
A 15% OC is still ~1.2ghz or ~150mhz. Better than the 50-75Mhz that review sites get. It would look less crap at OC for sure. Heck, maybe it can do a 20% OC with vcore for people who don't care about power use! Atm, it's not a possibility because they decided to launch it crippled!

 

[AtenRa]

Fury-X is like TITAN-X, no OC, no OEM custom designs etc etc.

Fury will be like GTX980Ti, custom PCBs, OCs, WC and AirCooled etc etc.

 

[tg2708]

All this back and forth about which card is more powerful yadda yadda when pc games are starting to become nothing but halfbaked ports of console games. Titan x, Fury x and 980ti are pointless cards at the moment and going forward because as 2015 shows games are not truly taking advantage of their power. You may argue that 4k is bringing some of these cards to their knees but whats the point when you need a high resolution monitor to actually distinguish pc quality from consoles? With previous games like crysis 3, metro last light etc. one could actually see the difference, but apart from running at higher resolutions and fps we have nothing. Top cards power are grossly underutilized if something does not change going forward. The PCmasterrace moniker is fading with each crappy games that are ported over.

 

[Silverforce11]

Even Titan X can do 15-20% OC on its crap reference cooler.

Fury X does ~75mhz extra, it's a joke.