The Fury(X) Fiji Owners Thread

bru · 7 Jul 2015 at 14:26

A more honest answer would have been:

Well the Fiji architecture is based on the same GCN 1.2 as Tonga and that isn't capable of HDMI 2.0, R&D constraints have meant we have been unable to engineer a new version of our GCN architecture, hopefully things will be better next year.

Gregster · 7 Jul 2015 at 14:49

I will do a Huddy style review (no offence meant Richard)

Ash0787 · 7 Jul 2015 at 14:52

The wait is a bit difficult at this point, ive wanted one of these things since it was just a rumour with no name

I think some people would be happy to buy them with faulty pumps for a price reduction because they will be putting waterblock on it anyway, not sure if ocuk would be allowed to do that though

The one I preordered was supposed to be around a few days ago but its overdue and ive looked around and cant see any information about it so I have no idea how long it will take, im even tempted to buy a cheap midrange card like a gtx 970 to use temporarily while I wait for the fury x because at the moment I am playing games at 15 frames per second

Gregster · 7 Jul 2015 at 14:53

LoadsaMoney said:
Got to agree, they should have had a plentiful supply for launch, unless its the pump issue thats holding them up, but saying that, thats something that should have been picked up on long before launch, then they could have had it sorted by.

Seeing how poor mine overclocks, I wonder if it isn't something there. I could just be donning a tin foil hat but mine isn't stable at 1120Mhz and that is a poultry 70Mhz over stock 1100Mhz seems to be good but not fully tested but 50Mhz doesn't fill me with confidence really and not the overclockers dream I was hoping for.

D.P. · 7 Jul 2015 at 15:04

LoadsaMoney said:
Got to agree, they should have had a plentiful supply for launch, unless its the pump issue thats holding them up, but saying that, thats something that should have been picked up on long before launch, then they could have had it sorted by.

From reading around on other forums with more experienced engineers I have come across the following hypotheses:
1) although 28nm process is mature Fiji is a very big chip by AMD GPU standards, much bigger than recent chips. They are also trying to push it to the limit, transistor density is higher than the GM200. Yields may simply be low, failure rates are proportional to area as well as transistor counts.

2) HBM supply issues. Although the components of HBM are not complex to make the manufacturing and assembly is very complex and relatively immature process. Bonding the stacks and the TSV connectors is challenging form an manufacturing perspective. Apparently 3 different companies are require to manufacture and bond a HBM chip. Testing methodologies may also be immature, e.g. it might not be easy to test the memory until it is all connected to the interposer. Also, with GDDR5 once a card card was assembled it could be tested and if a fault in a memory chip was found then the chip was easily removed and replaced. With HBM and the interposer it may mean the whole lot has to get chucked.

3) AMD launched too early, they simply weren't ready for production but Nvidia forced their arm. Ideally AMD would have built up a small stock pile over a couple of months and then launched but due to the delays already and the fact that 16mm FF is around the corner they had to do an almost paper launch to get things out the door.

we can add pump issues to that, although according to AMD this effects a tiny portion of cards to if you believe them then the pump issue shouldn't be a cause of stock issues.

it is unlikely that demand can be causing the issue just based on the lack lustre reviews and responses. I think Gibbo mentioned he had sold way more 980Ti during the FuryX launch even accounting for pre-orders.

D.P. · 7 Jul 2015 at 15:08

Dicehunter said:
One thing that made me giggle is when Mr Huddy was asked why it doesn't have HDMI 2.0 when even the lower end GTX 960 from Nvidia has it, His response just made me giggle -

"Well erm yes that's a *cough* good point and we ermm will be looking into that in the erm *Cough* future erm but there is a 3rd party *Cough* vendor releasing a erm *Cough* adapter that converts the HDMI 1.4 signal into erm HDMI 2.0"

The real reason is simply Fiji is architecturally the same as a Hawaii-Tonga hybrid, they didn't up date any of the chip related to HDMI and their resources were focused on A) HBM, B) getting more shaders on a 28nm process within the 600mm^2 restriction of the HBM interposer.

Gerard · 7 Jul 2015 at 15:11

D.P. said:
it is unlikely that demand can be causing the issue just based on the lack lustre reviews and responses. I think Gibbo mentioned he had sold way more 980Ti during the FuryX launch even accounting for pre-orders.

Well considering how they marked up the fury card its not surprising, £140 or thereabouts over what it was meant to sell for? Even now with a single card in stock its £649 whereas caseking.de have them in stock for £513.

TNA · 7 Jul 2015 at 15:15

D.P. said:
From reading around on other forums with more experienced engineers I have come across the following hypotheses:
1) although 28nm process is mature Fiji is a very big chip by AMD GPU standards, much bigger than recent chips. They are also trying to push it to the limit, transistor density is higher than the GM200. Yields may simply be low, failure rates are proportional to area as well as transistor counts.

2) HBM supply issues. Although the components of HBM are not complex to make the manufacturing and assembly is very complex and relatively immature process. Bonding the stacks and the TSV connectors is challenging form an manufacturing perspective. Apparently 3 different companies are require to manufacture and bond a HBM chip. Testing methodologies may also be immature, e.g. it might not be easy to test the memory until it is all connected to the interposer. Also, with GDDR5 once a card card was assembled it could be tested and if a fault in a memory chip was found then the chip was easily removed and replaced. With HBM and the interposer it may mean the whole lot has to get chucked.

3) AMD launched too early, they simply weren't ready for production but Nvidia forced their arm. Ideally AMD would have built up a small stock pile over a couple of months and then launched but due to the delays already and the fact that 16mm FF is around the corner they had to do an almost paper launch to get things out the door.

we can add pump issues to that, although according to AMD this effects a tiny portion of cards to if you believe them then the pump issue shouldn't be a cause of stock issues.

it is unlikely that demand can be causing the issue just based on the lack lustre reviews and responses. I think Gibbo mentioned he had sold way more 980Ti during the FuryX launch even accounting for pre-orders.

Sad to see really isn't it.

But can't blame consumers, at the current price points if I was buying I would probably get a 980ti too. Even though I would like to get my hand on new hbm tech, I think it's just best worth waiting on hbm2.

D.P. · 7 Jul 2015 at 15:15

Gregster said:
Seeing how poor mine overclocks, I wonder if it isn't something there. I could just be donning a tin foil hat but mine isn't stable at 1120Mhz and that is a poultry 70Mhz over stock 1100Mhz seems to be good but not fully tested but 50Mhz doesn't fill me with confidence really and not the overclockers dream I was hoping for.

The average overclock form reviews is around 70MHz.

My suspicion is that they have minimized the voltage necessary in order to help keep power and heat under control (and excess heat increases power even further). AMD designed the FuryX to keep to 60*C in order to limit electron leakage, in order to keep power usage reasonable at 40watts over 980Ti.

This may be good news in the future if voltage control is possible then the card could clock quite well but become a power hungry beast and generate a load of heat. that makes sense given that AMD claimed 400W board power and 500w cooling capacity. That may be required even for modest over clocks. The AIO should keep things quit enough.

Something I am slightly more worried abut is that performance didn't seem to scale well with overclocking. I may be mistaken but it seemed like an X% overlook got you a lot less than X% performance gain, while the 980Ti got closer to the X% gain. That might be something complex to do with the memory controller and interaction with HBM, or a bg in the power management that limits the effective clock speed.

D.P. · 7 Jul 2015 at 15:21

Gerard said:
Well considering how they marked up the fury card its not surprising, £140 or thereabouts over what it was meant to sell for? Even now with a single card in stock its £649 whereas caseking.de have them in stock for £513.

Point taken but honestly, given the reviews do you think AMD are experiencing a higher demand than expected, a lower demand, or demand as expected.

The FuryX does well but it isn't going to make a load of 980Ti/Tx/980 owners suddenly ditch their cards and buy 1-2 FuryX cards. Even some of the AMD faithful have decided to go for a 980Ti.

This is just speculation, we don;t have AMD and Nvidia official sales figures, but you have to look at the situation realistically. Excessive demand would be a great reason if the card smashed the 980Ti at all resolutions and cost £420 but that simply isn't case.

humbug · 7 Jul 2015 at 15:28

D.P. said:
The average overclock form reviews is around 70MHz.

My suspicion is that they have minimized the voltage necessary in order to help keep power and heat under control (and excess heat increases power even further). AMD designed the FuryX to keep to 60*C in order to limit electron leakage, in order to keep power usage reasonable at 40watts over 980Ti.

This may be good news in the future if voltage control is possible then the card could clock quite well but become a power hungry beast and generate a load of heat. that makes sense given that AMD claimed 400W board power and 500w cooling capacity. That may be required even for modest over clocks. The AIO should keep things quit enough.

Something I am slightly more worried abut is that performance didn't seem to scale well with overclocking. I may be mistaken but it seemed like an X% overlook got you a lot less than X% performance gain, while the 980Ti got closer to the X% gain. That might be something complex to do with the memory controller and interaction with HBM, or a bg in the power management that limits the effective clock speed.

The lack of Overclocking gain points to immature Drivers.

AMD have not done anything with the volts they don't do on all other cards, Fury actually overclocks better on Stock volts than a 290X.

Its not about trying to keep it within reasonable power levels, it does that anyway, its not much less efficient than Maxwell, Depending on where you look its uses about 10 to 45 Watts more than a 980TI and mostly in the middle of that at about 25 Watts.

Mtom · 7 Jul 2015 at 15:44

D.P. said:
The average overclock form reviews is around 70MHz.

My suspicion is that they have minimized the voltage necessary in order to help keep power and heat under control (and excess heat increases power even further). AMD designed the FuryX to keep to 60*C in order to limit electron leakage, in order to keep power usage reasonable at 40watts over 980Ti.

This may be good news in the future if voltage control is possible then the card could clock quite well but become a power hungry beast and generate a load of heat. that makes sense given that AMD claimed 400W board power and 500w cooling capacity. That may be required even for modest over clocks. The AIO should keep things quit enough.

Something I am slightly more worried abut is that performance didn't seem to scale well with overclocking. I may be mistaken but it seemed like an X% overlook got you a lot less than X% performance gain, while the 980Ti got closer to the X% gain. That might be something complex to do with the memory controller and interaction with HBM, or a bg in the power management that limits the effective clock speed.

This could come down again to the low volts and much more aggressive power tune than before.

humbug · 7 Jul 2015 at 15:48

What are the volts? PowerTune is adjustable in CCC, i believe its still +50% ?

Kaapstad · 7 Jul 2015 at 15:58

http://www.techpowerup.com/214114/ek-radeon-r9-fury-x-water-blocks-now-available.html

Mei · 7 Jul 2015 at 16:04

bru said:
A more honest answer would have been:

Well the Fiji architecture is based on the same GCN 1.2 as Tonga and that isn't capable of HDMI 2.0, R&D constraints have meant we have been unable to engineer a new version of our GCN architecture, hopefully things will be better next year.

if thats true they should have just said that str8 out
would have made them look less stupid
and they wouldnt of been asked about it in 100 different ways
the only answer to that is "ok"

Stu · 7 Jul 2015 at 16:34

Silent_Scone said:
Ultimately it's AMD that need to satisfy demand and currently this launch has been diabolical .

Disgustingly short supply honestly. They just couldn't get it out the door quick enough with what they had.

Sorry but it's true, really hasn't been good.

I think they only expected to sell 5 or 6 after the reviews got out, so no need to stock pile for the launch

Avenged7Fold · 7 Jul 2015 at 16:51

Dem fine ass blocks :cool:

Silent_Scone · 7 Jul 2015 at 16:54

Yeah and no stock lol.

Gerard · 7 Jul 2015 at 16:54

Got my gigabyte fury today and as luck would have it its one of the whiny ones. And Jesus Christ is it annoying. Even with headphones on its clearly audible, its just one of those pitches that seems to pierce everything and get incredibly annoying in a short space of time.