AMD Radeon R9 Fury X Video Card Review @ [H]

madgun said:
So Relayer would you care the same amount for NViidia's employees? Why the double standard. I don't believe an ounce of what you said.
Click to expand...

In the same situation and circumstance, of course I would. Why wouldn't I? There might be certain personalities associated with these companies that aren't well received by many. Jen-Hsun from nVidia and Huddy from AMD come to mind. They are both polarizing personalities. While some of us might get some satisfaction by them being taken down a peg or two the overall picture is much larger than the two of them. If you care about yourself and PC gaming you should want them both to do well. But I'm sure you've heard that argument already, so I won't bore you.

The only point I was presenting with the post was that it's not "just" GPU's. There are real people behind these products who it is very important to. What's so hard to believe about that?
 
The Fury numbers have me disappointed. That means I attached some emotional value to what they were going to be. Yeah, that makes me an AMD fan. Look at my sig: 2 of my 3 gaming rigs have Nvidia cards. I need/want to upgrade an HD6870. The new card was going to be the cornerstone of giving that rig an upgrade into a 1440p frame-eating beast. Now, I have to rethink my approach.

I wanted to be able to put a Fury into it.

I will note that today's Newegg special shows a 4Gb 980 (vanilla, Asus Strix) for sale at $500.

I am pulling for AMD...but I am not giving money to a charity case. They've got to be competitive. If not in performance, then in price.

I may just toss a 380X (if we EVER get a review - HINT, HINT!) into that rig and call it a day and leave it as a 1080p.

Ken

(Edited to show the correct 980 price. I thought it was $450. I was wrong. It is $500.)
 
Relayer said:
So when the 980/70 were released if they would have compared them to O/C'd 290/X (which judging from the 390/X they would have meet/beat them) everyone would have been OK with that? Somehow, I think not. :p
Click to expand...

Yes if there wasnt any chance an overclocked card was coming out.
But that wasnt the case, unlike the Fury X.
There isnt a glimmer of an overclocked card on the horizon.

:p
 
Nenu said:
Yes if there wasnt any chance an overclocked card was coming out.
But that wasnt the case, unlike the Fury X.
There isnt a glimmer of an overclocked card on the horizon.

:p
Click to expand...

A new qualification. I see. :cool:
 
I'm sorry for poking the hornet's nest further, but has anyone seen this?
3v87mcvjenj.jpg


Supposedly the Fury X is taking the piss after 3.5GB of VRAM allocation, never mind the fact that bandwidth is far off from the claimed 512 GB/s at stock speed. Would be funny if this turned out to be true.
 
Relayer said:
The only point I was presenting with the post was that it's not "just" GPU's. There are real people behind these products who it is very important to. What's so hard to believe about that?
Click to expand...

There are people behind everything. Why should that matter to consumers or reviewers?
 
Armenius said:
I'm sorry for poking the hornet's nest further, but has anyone seen this?

Supposedly the Fury X is taking the piss after 3.5GB of VRAM allocation, never mind the fact that bandwidth is far off from the claimed 512 GB/s at stock speed. Would be funny if this turned out to be true.
Click to expand...

Fury X with potentially its own 3.5GB memorygate scandal ala 970? All the AMD Bros that always harp on evil Nvidia and the 970's memory better hope and pray to their lord and saviour that this isn't true, because this would put the nail in the coffin.
 
Last edited:
DPI said:
Fury X with potentially its own 3.5GB memorygate scandal ala 970? All the AMD Bros that always harp on evil Nvidia and the 970's memory better hope and pray to their lord and saviour that this isn't true, because this would put the nail in the coffin.
Click to expand...

haha is funny really, anyone remember this?

3a.jpg
 
DPI said:
Fury X with potentially its own 3.5GB memorygate scandal ala 970? All the AMD Bros that always harp on evil Nvidia and the 970's memory better hope and pray to their lord and saviour that this isn't true, because this would put the nail in the coffin.
Click to expand...

lol! if this turns out to be true, AMD went on a huge campaign to mock NV, so they will be eating their own words...
 
Armenius said:
I'm sorry for poking the hornet's nest further, but has anyone seen this?
3v87mcvjenj.jpg


Supposedly the Fury X is taking the piss after 3.5GB of VRAM allocation, never mind the fact that bandwidth is far off from the claimed 512 GB/s at stock speed. Would be funny if this turned out to be true.
Click to expand...

Aww yes. This situation definitely needed more drama, this would be perfect if it was true.
 
In fairness I don't believe Fury X actually has segmented memory like 970. After doing a bit of digging it looks like this may just be a software issue and mismagement of the way VRAM is allocated or managed.

Holding off on any indictment for now, need more data. This benchmark for all we know could also be flawed and requests VRAM differently than a game would.
 
Last edited:
It has to do with the fact that windows needs some space on the video card to do things like manage windows. The only way to get an accurate speed result is to run the card in headless mode apparently with aero disabled. E.g. run off the integrated video on the CPU without a monitor attached to the card. The original post:

http://forums.anandtech.com/showthread.php?p=37522172#post37522172

further down shows someone who did the test on a 980ti being used and it too shows the slowdown on the last 512MB, and that poster claims to have seen it do it on a Titan X as well. Basically windows reserves the last 512MB of a video cards buffer to do windows stuff.

Would have been hilarious if true though
 
-Dragon- said:
It has to do with the fact that windows needs some space on the video card to do things like manage windows. The only way to get an accurate speed result is to run the card in headless mode apparently with aero disabled. E.g. run off the integrated video on the CPU without a monitor attached to the card. The original post:

http://forums.anandtech.com/showthread.php?p=37522172#post37522172

further down shows someone who did the test on a 980ti being used and it too shows the slowdown on the last 512MB, and that poster claims to have seen it do it on a Titan X as well. Basically windows reserves the last 512MB of a video cards buffer to do windows stuff.

Would have been hilarious if true though
Click to expand...

So if the GTX970 reserves the last 512MB (of slow VRAM) for Windows stuff, why did people care and break out the torches and pitchforks?
 
ocellaris said:
So if the GTX970 reserves the last 512MB (of slow VRAM) for Windows stuff, why did people care and break out the torches and pitchforks?
Click to expand...

In the 970's case there was actually physical segmentation of the memory I really don't believe Fiji is doing same just based on a crude benchmark that results are getting misinterpreted on.
 
Last edited:
ocellaris said:
So if the GTX970 reserves the last 512MB (of slow VRAM) for Windows stuff, why did people care and break out the torches and pitchforks?
Click to expand...

Well the 970 is a physical issue. If you test the other gpus from a live cd or using onboard video they wouldn't have the issue but the 970 still would.

I'm not sure what kind of a performance difference software vs hardware allocation makes however.
 
ocellaris said:
So if the GTX970 reserves the last 512MB (of slow VRAM) for Windows stuff, why did people care and break out the torches and pitchforks?
Click to expand...

There shouldn't be any VRAM reserved when running games in fullscreen. In windowed mode yes it will behave like that app.
 
Kardonxt said:
Well the 970 is a physical issue. If you test the other gpus from a live cd or using onboard video they wouldn't have the issue but the 970 still would.

I'm not sure what kind of a performance difference software vs hardware allocation makes however.
Click to expand...


512 mb of Vram has always allocated to Windows by the graphics driver, since Vista, and WDDM 1.0. Bios also allocates 256 mb system memory for the OS. This is even done in exclusive mode.

Now the 970 does have physical limitation of of the way it was designed but this is why that last 512 mb doesn't effect performance in games.
 
razor1 said:
512 mb of Vram has always allocated to Windows by the graphics driver, since Vista, and WDDM 1.0. Bios also allocates 256 mb system memory for the OS. This is even done in exclusive mode.

Now the 970 does have physical limitation of of the way it was designed but this is why that last 512 mb doesn't effect performance in games.
Click to expand...

Except it does affect performance in games... When it's being used. nVIdia drivers do their best to avoid using that last 512MB but in cases where it needs to be used, it certainly affects performance. This has been documented and even video documented several times.
 
Relayer said:
In the same situation and circumstance, of course I would. Why wouldn't I? There might be certain personalities associated with these companies that aren't well received by many. Jen-Hsun from nVidia and Huddy from AMD come to mind. They are both polarizing personalities. While some of us might get some satisfaction by them being taken down a peg or two the overall picture is much larger than the two of them. If you care about yourself and PC gaming you should want them both to do well. But I'm sure you've heard that argument already, so I won't bore you.

The only point I was presenting with the post was that it's not "just" GPU's. There are real people behind these products who it is very important to. What's so hard to believe about that?
Click to expand...
No offense but why should any of us care? If your business puts out uncompetitive products and you aren't economically viable that's not my problem. The employees are still getting paid this whole time and if AMD was to fold they all work in tech so they'd find other jobs.

I'm not saying I want AMD to go under - quite the contrary - but don't bring emotional nonsense into an objective discussion of product quality.
 
RamonGTP said:
Except it does affect performance in games... When it's being used. nVIdia drivers do their best to avoid using that last 512MB but in cases where it needs to be used, it certainly affects performance. This has been documented and even video documented several times.
Click to expand...


Can you show me when that 512 mb is being used? Its actually reserved for windows.

It doesn't matter if the program is in a window or full screen, this even happens in exclusive mode.

Every card has to reserve that 512mb of memory for windows.

Yeah just double checked on this on MSDN.

Ok I did double check, the 512mb does affect the 970 as much as 5%...

Yeah the page file is pushed over to system memory when using over 3.5 gb.
 
Last edited:
Turns out HBM is indeed overclockable. Check out the big gains in 3DMark:

"Officially there is no option to adjust the memory clocks, but (probably due to a bug in the drivers) we did see a slider to adjust memory clock frequency after every other few reboots. Hence, we could also raise the memory clock.
We ended up being able to change the GPU speeds by 95MHz to 1145 MHz and the memory by 100MHz. The latter seems a bit on the low side, but it is actually 20% on top of the standard frequencies this card uses.
In the end we managed to get a 3DMark Fire Strike score of 16963 points on overclocked settings, a nice increase on the standard 14098 points we achieved"........

"So they achieved 20% increase in fs score with 9% OC on core and 20% OC on mem. Because this is AMD we are talking about, please note that the graphics score on above bench was 19321, compared to normal FuryX 16k score. It proves that overclocking HBM does impact the performance"........

http://www.overclock.net/t/1562593/hardware-info-hbm-on-furyx-can-be-overclocked-after-all
 
some were speculating that the low clocks were adding extra latency hence why the OC on memory made a difference, not usually seen with e\other existing cards with GDDR5. None felt it was anything to do with the extra bandwidth, as most here have already claimed with the existing >500Gb/s
 
After 31 pages they have only run synthetic tests, there are no gaming results.
We need to see how the overclock translates to real world use.
Are they avoiding the issue?
 
DracoNB said:
Who are you talking about? That info is from this review: http://uk.hardware.info/reviews/615...-flag-ship-graphics-card-overclocking-results
Click to expand...

Yes, all the cards sold out so there must be many owners who would like to prove its worth.
In 31 pages, no one verified its gaming performance.

This thread on Anandtech highlights there is a driver bug that allows anyone to enable memory overclocking
http://forums.anandtech.com/showthread.php?p=37521880
No gaming results in either thread though.
 
Nenu said:
Yes, all the cards sold out so there must be many owners who would like to prove its worth.
In 31 pages, no one verified its gaming performance.

This thread on Anandtech highlights there is a driver bug that allows anyone to enable memory overclocking
http://forums.anandtech.com/showthread.php?p=37521880
No gaming results in either thread though.
Click to expand...

lol, yes thought the same, no gaming tested but the most important thing, no valid result, hence the screenshot while: "validating score" also the 3dMark Graph at the right side is totally blank... I can easily make a photoshop'd image of my 280X scoring higher than a Fury X if i want =)..
 
I agree, I would like to see how those OCs would result in Gaming. But so far with the few synthetic benches all are showing performance increases. You have seen the 3DMark but another guy is claiming with 600mhz he goes from 49 to 57 in Kombusters 1080p test. So there seems to be some positives coming from OCing the memory, unlike what we have seen with GDDR5 and previous.
 
Thats the point of a synthetic test though, to show a difference in all a gfx cards elements.
This is why we dont use synthetic tests to gauge real world performance.
 
Nenu said:
Thats the point of a synthetic test though, to show a difference in all a gfx cards elements.
This is why we dont use synthetic tests to gauge real world performance.
Click to expand...

We who? Do you have some benchmarks to share?
 
You must log in or register to reply here.
Back
Top