AMD Fury X Postmortem: What Went Wrong?

[Silverforce11]

Only a small sample of games and Fury generally seems to perform better with AA turned up. I somehow doubt drivers have much to do with it so much as different settings. AF is ezpz for modern GPU so it shouldn't even be much of a factor.

I'm not saying either way, there's a ton of variability across the review sites in general. I've seen BF4 results where its 25% slower, to where its a tie, same for GTA V.

 

[DDH]

So you couldnt post a link to those claimed magic drivers?



Oh. now we get them

I thought the memory was locked and couldnt be overclocked? Is this even real?
Why were the charts being posted without a source?

EDIT--->
guess that is that



but i bet this wont be the last time i see these faked charts

There is absolutely two versions of drivers out there. 15.15 and 15.15.1004. Released 15/06 and 20/06 respectively

Whether there is any difference besides the batman changes who knows.



AMD have stated the mem cant yet be overclocked. Did these guys overclock it? Who knows.

You have a 980. Why do you even care?

 

[raghu78]

I'm not saying either way, there's a ton of variability across the review sites in general. I've seen BF4 results where its 25% slower, to where its a tie, same for GTA V.

hwc and pcper have commented that they suspect drivers are still not fully optimized. Even though the launch has been botched AMD can redeem themselves if they can get the drivers in much better shape before R9 Nano launches. Thats going to be the real high volume product. Anyway right now Fury X supply is just not enough to match demand. So AMD has not lost out in a big way. Atleast not yet. But after the early adopters are gone AMD needs a better show of performance if they want to get market share back.

Games like DAI show 10-15% performance improvement for 45% more shaders and much more bandwidth and improved ROPs. Something is definitely wrong with the perf scaling.

http://www.hardwarecanucks.com/foru...682-amd-r9-fury-x-review-fiji-arrives-10.html

http://www.techpowerup.com/reviews/AMD/R9_Fury_X/17.html

http://www.computerbase.de/2015-06/.../6/#diagramm-dragon-age-inquisition-2560-1440

Lot of games across multiple reviews are showing poor perf scaling from R9 390X to Fury X. With 45% more shaders and same core clocks with lots more bandwidth and improved ROPs and tesselation I would expect perf scaling to be atleast 25%. But there are so many games where scaling is 10-20%. Very disappointing. AMD has dropped the ball once gain when it comes to launch drivers. :thumbsdown:

 

[Larnz]

I would love for this "un-optimised" driver thing to be true. I was hoping the Fury X would be a bit better atleast on par at 1440p with the Ti.

I am in the market for a new card and its a bit of a toss up really I know the 980Ti is currently better but the pricing for cards here is a bit whack, AMD is always cheaper in NZ than Nvidia for some reason. I can get a Fury X for $100USD cheaper than a standard reference 980Ti... makes the decision a bit harder.

 

[ViRGE]

Sorry guys:



http://forums.overclockers.co.uk/showthread.php?p=28230087#post28230087

Thank you for pointing that out before this got too far out of hand. Reddit I can understand, but the AT Forums users are supposed to be smarter.

AMD driver strings are in YYMMDD format. So there can't be a build with 15.15-180612a. If you look at past AMD drivers, you can quickly see this.

 

[RampantAndroid]

Edit, fail...already posted.

https://www.reddit.com/r/pcmasterra...y_x_possibly_reviewed_with_incorrect_drivers/

Apparently AMD provided the wrong drivers for the reviews?

And a reply from AMDMatt:

Yep, no.

For reference on what driver strings mean by the way:
15.15-150611a-185358E (This is the real driver we provided)
15.15 - is the branch
150611 - is the date of the build, YY/MM/DD
185358 - is the build request from our system to create this driver based off the information above

In this case the review is suggesting a driver dated from June 12th 2018 and the build request that has a letter instead of a number for its last digit so it’s either a lot of typos or someone being misleading on purpose.

 

[bryanW1995]

How bad can it get? Pretty bad. If AMD is unable to tame the challenges of first-generation HBM, then Fiji could prove to be more of a tech demo than a viable commercial product. We could be looking at Bulldozer 2.0, and that's a blow AMD may not recover from. Fiji is a massive, go-for-broke chip; if despite its technical advancements it fails to recover the performance crown (falling short of Titan X by 5%-10%), then it will be a failure. AMD won't be able to sell it for more than $499, and that price won't be sufficient to recoup R&D costs. Meanwhile, let's suppose that the pessimistic rumors are true and all the cards except Fiji are straight rebrands. No port to GloFo process, just better binning, higher clocks, and more RAM. This leaves AMD's entire lineup in an uncompetitive position; without bringing anything new to the table, they'll continue to have to compete on price with the legion of ex-mining cards out there, not to mention the cheap R9 200 series clearance sales. AMD continues to hemorrhage market share for 18 months or more, waiting for the FinFET+ process to arrive... assuming they survive that long. The only real hope at that point is a buyout offer from Samsung.

Your "most pessimistic possible" scenario was pretty close to what my most optimistic one was.

You sure seem to have it in for AMD though, I remember seeing you write dozens of critical posts about them in between the announcement and benchmarks being released, with never a single one that was anything other than critical. I wonder why that is? Why do you have such a grudge against a corporation?

 

[yacoub]

4GB limitation, no HDMI 2.0, outperformed in most tests by 980Ti, yet they price it $650 like the 980Ti. Does not compute. I understand they are struggling and need cash. That doesn't change the fact that the card is not worth that price based on its performance.

 

[beginner99]

AMD made the same error they have been doing the last decade. They are trying to be too clever. Yes, Fury X is awesome in compute and water-coolers in high end might become more common or the standard. Yes, it will probably age better and demolish maxwell in dx12. Yes HBM is cool and will be required.

But all this helps 0 to perform NOW in current dx11 console ports. NV showed that GDDR5 is still good enough for current gen flagships. No need for potential expensive HBM increasing BOM. Compute doesn't matter that much in current game except Civ 5 and there Fury X blows the competition out of the water. But it's also the game you can play at or below 30 fps without much issues.

This happened with Phenom (first real quad-core), integrated memory controller (ok, that one was actually good but still no chance vs. a better CPU uArch, namley Core 2). And then there was Faildozer, the biggest mishap of them all. The idea wasn't bad but not well thought through (software licensing cost per core...) and terribly executed. Now we have Fury X. Compared to Bulldozer it can actually be recommended for some niche users (small cases, Civ 5 nerds) and performance is there but the competitions product offers a bit better performance/$.

All in all I get the vibe AMD is some kind of place where the kids (engineers) get to play and implement some cool features with little thought on having a business case. Too much focus on cool and fun than actually making money. Now this sounds like a great place to work at, just not so great for shareholders and getting money for R&D.

They should have ignored flagship performance and just improved upon Hawaii. Reduce power consumption, color compression, HVEC, add HBM ( to get experience before 14 nm shrink) and maybe add some shaders to get maybe 10-20% over custom 290x. Refresh rest of lineup at lower prices. They could have sold this new 390x at $400 and people would have bought it in masses. They could have actually made money from it which I doubt will happen with Fury X. This hypothetical GPU could then just be die shrunk to 14 nm and sold as low-midrange GPU, another market with many buyers.

 

[coercitiv]

All in all I get the vibe AMD is some kind of place where the kids (engineers) get to play and implement some cool features with little thought on having a business case.

Bubble burst: the place where "kids" are is here. We're the ones playing around, doing armchair analysis, and most importantly foretelling after the event.

There must be dozens of forum members that can conquer the world of computing if they have AMD, Intel, or Nvidia doing their bidding. Just imagine the power behind products made with the knowledge, skills, expertise and hindsight of a Forum Approved™ CEO. Endless possibilities.

 

[Jaydip]

7970Ghz is actually faster than 280X (which is faster than 770, which again is faster than 680) if you didn't know your history.

Again the current problem with Fury X is poor value against custom 980Ti, it's not a flaw that is hardware related. If AMD drops the price, problem solved. Unfortunately it looks like they won't, cos its sold out everywhere.

Ahem
http://www.techpowerup.com/reviews/Gigabyte/R9_280X_OC/26.html

 

[looncraz]

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!

AMD has never shown an ability to learn from the mistakes of others, let alone their own mistakes.

The entire design of Bulldozer is a testament to that. They chased the long-pipeline high-clock-rate dragon after watching a company with literally an order of magnitude greater R&D budget fail miserably at doing so despite many years of attempts at optimizing it.

The Pentium 4 was a terrible CPU and AMD's alpha-inspired Athlons were raking the floor with them. 3.4GHz hot-running Pentium 4, or a nice Athlon XP+? Your choice.

Hyper-threading, though, did make the Pentium 4 more pleasant to run due to the poor Windows scheduler.

 

[looncraz]

Fiji is new silicon. We don't know what it can do with more voltage. It could do much better than Hawaii, for example. It could very well have a similar behavior. We don't know. As it is, FuryX needs a price cut, that's all. Going forward it's gonna get better, as GCN did in the 7970 and 290x.

Consume more power after getting more voltage? Yes, this also happens to every piece of silicon out there. It could be worthwhile. It could be not. Still, it's a product that gets AMD on the 980Ti range of performance. It was needed, they couldn't have possibly gone much longer relying only on Hawaii. It's no bulldozer as it's been called in this thread. Subpar, I agree. A train wreck? No, not at all. A 2900XT vs 8800GTX repeat would've been a train wreck. It isn't this time.


If you're in a hurry there are a lot of juicy >1300MHz 980Tis out there that can get you going.


We can all agree nV has won the DX11 era, on both performance and drivers once Maxwell arrived. There's a new chapter about to be started in a month with W10+DX12 and AMD's hardware seems to be tailored for DX12 from day one, back to the 7970 almost four years ago. Things could get *very* interesting once DX12 games arrive. Meanwhile, Maxwell is still king, especially when overclocked, and Pascal/Arctic Islands are far, far away. In between, anything could happen.

I don't get why people keep calling the 2900XT a train wreck vs the 8800GTX. It was cheaper and performed about as well.

http://www.legionhardware.com/articles_pages/asus_radeon_hd_2900xt_vs_asus_geforce_8800_gtx,1.html

 

[selni]

The launch didn't seem that bad - the 980 Ti was a very effective spoiler but it's still a giant leap forward from the 290X and is at least competitive with highend NVidia cards. It's just really, really late and I wonder if this is the price AMD wanted to be able to charge...

 

[MattL]

Nvidia one upped AMD with the 980 Ti... imagine if the Fury X came out without a 980 Ti in the market... the Fury X would be competing in price with a 980 and be doing very well, comparing closer to a Titan at a much higher price point.

Nvidia was extremely smart.

With that said I'd say wait for the proper overclocking and further drivers to support it. Designed with the water block and running as cool as it is I really think this card is targeted for serious overclocking.

Also isn't their an air cooled version of basically the same card coming out at $550 soon? If so then that will be the real high performance value bargain (at least having a serious shot at it).

 

[flopper]

Fury X seems supreme as a card.
its a buy for sure

 

[Mako88]

With that said I'd say wait for the proper overclocking and further drivers to support it. Designed with the water block and running as cool as it is I really think this card is targeted for serious overclocking.

Also isn't their an air cooled version of basically the same card coming out at $550 soon? If so then that will be the real high performance value bargain (at least having a serious shot at it).

FX is voltage locked, which means you need to run a modded BIOS to unlock it and get even halfway decent clock rates, which equals warranty voided. So overclocking isn't something that can save the day here unfortunately.

And the air cooled card, which I also thought would be the best value personally, is a cut down chip apparently to save energy and heat. Much slower than the AIO water FX allegedly.

And that's a true a shame, because the best bang for the buck would have been two air cooled Fury X cards in Crossfire for just $1100. Nothing would have been able to touch that value proposition.

 

[raghu78]

Fury X seems supreme as a card.
its a buy for sure

Right now Fury X lacks the consistency of a GTX 980 Ti especially in the software department. Performance scaling from R9 390X ranges from very bad to really good. So AMD has its work cut out. Multiple reviewers have pointed out that Fury X could be held back by drivers. AMD needs to improve the performance through drivers and then they have a chance.

 

[Mako88]

Right now Fury X lacks the consistency of a GTX 980 Ti especially in the software department. Performance scaling from R9 390X ranges from very bad to really good. So AMD has its work cut out. Multiple reviewers have pointed out that Fury X could be held back by drivers. AMD needs to improve the performance through drivers and then they have a chance.

They've been saying that since...2003? The whole "oh just wait for teh betters drivers guyzzzz" thing is played out. Enough.

Besides, the core is similar enough to Tonga (40% I believe? Someone correct if wrong) that drivers aren't going to magically unlock anything down the road.

It is what is, and truthfully it's pretty darn good. It's just a same that the voltage is locked, no HDMI 2.0, 4GB, and the price being around $49 away from being a home run.

Lets wait and see if the air cooled card really is cut down, if it isn't that will be a great bargain when you think about it.

 

[Atreidin]

What's with the hilariously editorialized title? Who died again? Do we have to agree with the premise that something went wrong?

They certainly didn't hit it out of the park but performance that is, I'd assume, indistinguishable from a 980ti outside of benchmarks isn't anything to scoff at. Certainly not a "disaster" or needing a "post mortem" . Not enough to turn things around for them either though I suspect.

Lol I was thinknig the same thing. Some people just need to overdramatize.

 

[Ancalagon44]

I think the problem with Fury X is in some way HBM.

Why? AMD felt that, because the card has HBM, it needs to be the highest of high end, to target the 980Ti. However, its very difficult to make a card to target that level of performance. Not only that, but HBM's immaturity means that only 4GB of RAM is possible. Not that the amount really matters - we have seen that it still does amazingly well in 4K benchmarks. My point is that, not only is the performance slightly short of the 980Ti, but with the memory being smaller too, it means the Fury X is not as future proof.

I personally think they should have repeated the 4870. Made a card similar to the 290X, maybe a little more optimized if they can, with 4GB of HBM. Price it competitively, make it a card that can beat a 970 in all situations, and you will have hordes of fans. It would sell very well.

Very few people are going to buy the Fury X. The problem with aiming for a halo product like the 980 Ti, is that if you shoot and miss, you lose everything. Very few people will buy it, because it is not a convincing win.

If you're going to spend $650 on a graphics card, you want to know you are getting the best. The Fury X is not the best, so why spend $650 on it?

 

[Keysplayr]

Fury X seems supreme as a card.
its a buy for sure

(I'll bite.)

For who?

 

[wand3r3r]

(I'll bite.)

For who?

No need bait like that. (goes for both)

It does fine for 4k. If you don't overclock it's not far off the reference 980 ti, basically trading blows.

Small form factor.

Sure it's not the best option, but it's not a horrible card either like you try to imply.

 

[Vesku]

The launch didn't seem that bad - the 980 Ti was a very effective spoiler but it's still a giant leap forward from the 290X and is at least competitive with highend NVidia cards. It's just really, really late and I wonder if this is the price AMD wanted to be able to charge...

It's only 3 weeks behind the 980 Ti and that's mainly because Nvidia chose to release ahead of AMD's long scheduled event. I'd say the development cycle for Fury X just made launching to spoil the Titan X impossible, just took too long for them to get HBM up and running.

 

[Vesku]

They've been saying that since...2003? The whole "oh just wait for teh betters drivers guyzzzz" thing is played out. Enough.

Besides, the core is similar enough to Tonga (40% I believe? Someone correct if wrong) that drivers aren't going to magically unlock anything down the road.

GCN has continued to improve via drivers pretty much since the first update after 7970 launch. Odds are there is quite a bit of tweaking necessary to match GCN up with HBM rather than GDDR5 to take full advantage of the lower latency and wider bus.