Fury X Benchmarked @ 4K (via AMD)

On the consumer side, whatever results the Fury X gets, enjoy your "cheap" watercooled technical marvel. (Not being facetious here, I am amazed at what these guys can do with silicon and tiny bits of metal.)

On the business side, is it enough to sway a Greenie over to the Reds and gain market share? Is there enough margin built into that $649 part to recoup R&D costs considering ONLY AMD will be making the WCE version? Remember, since AMD is "making" this they have to bear the brunt of the supply line costs and delays rather than just selling a chip to a partner and saying good luck. Anyone think they put in an order for 2 million of those WCE units? Yeah, probably not.

Is there going to be enough stock of the Fury X or are most going to have to settle for the AIB Fury versions? How is stock going to affect pricing as the X will be the only Fury available for a month until the non-X. Does this put the bulk of AMD's reputation on 3rd parties to deliver PCB/cooling solutions that are cost and performance competitive?

Are they staggering the releases so they can report the costs in different financial quarters instead of a(nother) big bomb in one quarter or is it purely technical/manufacturing reasons?

Will I be blackballed from receiving review samples for asking these questions? *grin*
 
I'm actually more curious to see what the air cooled version looks like. I like the WC version, but I really want to see if they applied their engineering chops to the air cooler too. All of their past air coolers have been bollocks.

I also hope that the air cooled version is not a cut-down GPU part.
 
I don't know if anyone has even seen a reference air cooled version. I think they may have left them totally up to the partners to follow the ref. design or go with their own.
 
Osjur said:
Can we please stop spouting this bullshit about 4GB not being enough for 4K. Someone show me a game benchmark where it really shows that 4GB is not enough because I haven't seen one. Not here and elsewhere.

Atm you can't max out games at 4K with single Fury X, TX, you name it so the VRAM concerns become moot. It might hold some candle if you are driving 4x TX against 4x Furys and can MAX the games at 4K but we will see it when someone decides to do that test.

Yes its a bitch thing to not have HDMI 2.0, though I don't really care and I've wanted DP only card for a while now, too bad they just decided to use full size DP's and not mDP
Click to expand...

http://www.hardocp.com/article/2015/04/20/grand_theft_auto_v_video_card_performance_preview/3
What we have learned from these tests are that 4GB video cards are going to suffer at the highest in-game settings in this game due to the limited framebuffers. We are going to have to adjust settings to alleviate that bottleneck for these to truly shine. In our full-evaluation we will find out what settings are actually playable once we figure out which setting reduces the VRAM bottleneck on them. As it is right now, GTX TITAN X takes the cake as the best experience.
Click to expand...
 
DracoNB said:
You realize that the 295x2 had higher avg frames than the Titan right?

Maybe everyone should just chill out and see what the actual performance is instead of guessing?
Click to expand...

What's the fun in that?

Did the guy on reddit eat a sock?
 
DracoNB said:
You realize that the 295x2 had higher avg frames than the Titan right?

Maybe everyone should just chill out and see what the actual performance is instead of guessing?
Click to expand...

You did read my post?
 
Nenu said:
You did read my post?
Click to expand...

Did you read the review? The 295x2 was getting higher avg frames, but they were choppy which is why they went with the titan, besides it being 15-20% slower.

We don't know if that was due to drivers, max ram, ram bandwidth or what exactly caused it.
 
I hope to see tests like that repeated on the Fury X, although it 'might not' be a VRAM problem it should be easy to compare it with the FX's results since it's a single GPU.

It would be super awkward if the Fury X is marketed as a 4K card and yet it can't even handle GTA V without sporadic dips.
 
DracoNB said:
Did you read the review? The 295x2 was getting higher avg frames, but they were choppy which is why they went with the titan, besides it being 15-20% slower.

We don't know if that was due to drivers, max ram, ram bandwidth or what exactly caused it.
Click to expand...
Thats why they wrote a small editorial which I quoted, [H] are in the best seat to provide answers.

More proof that 4GB can be maxed out
http://www.hardocp.com/article/2015/06/15/nvidia_geforce_gtx_980_ti_video_card_gpu_review/10
Dying Light maxes out 4GB at 1440p.
GTA V maxes it out at 4K and thats without using MSAA !

Shadows of Mordor maxed out my 290x at 1080p !!
 
Dual chip cards from either camp have been real dogs since they were introduced. Consumers should pretty much consider them Chernobyl/Fukushima grade toxic crap and avoid em.

The warnings never work though. As soon as they tease a new dual card everyone nerdgasms all over themselves despite still being sore from the ass pounding experience had with the last dual card.
 
Nenu said:
Thats why they wrote a small editorial which I quoted, [H] are in the best seat to provide answers.

More proof that 4GB can be maxed out
http://www.hardocp.com/article/2015/06/15/nvidia_geforce_gtx_980_ti_video_card_gpu_review/10
Dying Light maxes out 4GB at 1440p.
GTA V maxes it out at 4K and thats without using MSAA !

Shadows of Mordor maxed out my 290x at 1080p !!
Click to expand...
Problem is, the fastest GPUs on the market right now are the 980 Ti and TX both of which have 6 GB or in excess, the only way to compare 4 GB performance-wise is by relying on SLi and CrossFire.
You can't find issues with multi-GPU setups and blame it on VRAM.
 
TaintedSquirrel said:
Problem is, the fastest GPUs on the market right now are the 980 Ti and TX both of which have 6 GB or in excess, the only way to compare 4 GB performance-wise is by relying on SLi and CrossFire.
You can't find issues with multi-GPU setups and blame it on VRAM.
Click to expand...

Dying Light @ 1440p and Shadows of Mordor @ 1080p can max 4GB with playable settings on single card.
There will be more new games pushing the boundary for sure.

Its not hard to establish whether the bottleneck is memory, when you can see stutters and you know you are maxing 4GB thats a good indication to start with.
The 980ti can be clocked down to the same performance as the 980 to fully confirm that it is just memory size.

I trust [H] opinion on this, its something I have watched closely for a long time.
They do a fair amount of memory committed testing now.
 
I don't trust utilization numbers but I do trust performance skips, IE: frame times / stutters.

Don't care what a number in MSI Afterburner says.
I'm going to put this in my sig, tired of saying it... over and over...

"6000 MB" in a chart tells me nothing about real-world use.

This single line is more useful than an entire chart of VRAM numbers:

During actual gameplay it did feel stuttery, and not as smooth as the performance would indicate. In fact, you can kind of see this in the framerate which again isn't a smooth line by any means, but rather a lot of up and downs in framerate.
Click to expand...

Repeat that issue in a single GPU setup and you have me convinced. Nothing else will sway me in the slightest.
No utilization tables, no multi-GPUs. Tell me the Fury X, with its 4GB HBM, stutters at 4K (and the 980 Ti doesn't) and we're done here.
 
AVogel said:
Dual chip cards from either camp have been real dogs since they were introduced. Consumers should pretty much consider them Chernobyl/Fukushima grade toxic crap and avoid em.

The warnings never work though. As soon as they tease a new dual card everyone nerdgasms all over themselves despite still being sore from the ass pounding experience had with the last dual card.
Click to expand...

Agreed, it's kind of like Right Wing political thinking. Seems great in principle, but never actually works out in practice.


(fanning:D)
 
TaintedSquirrel said:
I don't trust utilization numbers but I do trust performance skips, IE: frame times / stutters.

Don't care what a number in MSI Afterburner says.
I'm going to put this in my sig, tired of saying it... over and over...

"6000 MB" in a chart tells me nothing about real-world use.
Click to expand...
They described the experience and the cause, it wasnt just charts.

This single line is more useful than an entire chart of VRAM numbers:

Repeat that issue in a single GPU setup and you have me convinced. Nothing else will sway me in the slightest.
No utilization tables, no multi-GPUs. Tell me the Fury X, with its 4GB HBM, stutters at 4K (and the 980 Ti doesn't) and we're done here.
Click to expand...
They were single GPU results !!
If the FuryX stutters due to lack of vram, the 980 will as well and vice versa.


Some more single card info here
http://www.hardocp.com/article/2015/03/10/dying_light_video_card_performance_review/10
The issue to focus on however is that all of these video cards are hitting the maximum capacities in Dying Light. This game demands high VRAM capacities to alleviate stutter. We found that in VRAM limited situations the game would stutter when being grabbed by zombies, or having to combat zombies. There would be a hang, or pause, or freeze if the resolution was too high for the video card.

For example, at 4K this happened on the GTX 980 and R9 290X, the 4GB of VRAM wasn't enough at 4K. On the 2GB video cards trying to run them at high setting sat 1080p also caused stuttering. That's right, even 2GB video cards had limitations at 1080p when trying to run higher settings.

Our recommendations, based on our experiences, lead us to recommend these VRAM capacities for the best enjoyment of Dying Light.

1080p - 3GB VRAM Video Cards

1440p - 4GB VRAM Video Cards

4K - 6GB+ VRAM Video Cards

This is a huge open world, there is a lot of detail in Dying Light. Running "High" textures, and "Very High" Shadow Map with "Maximum" View Distance will eat into your VRAM big-time.
Click to expand...
 
Nenu said:
If the FuryX stutters due to lack of vram, the 980 will as well and vice versa.
Click to expand...
It's a lot of extrapolating for a card nobody has touched before.
I don't really care how the 980 and its 256-bit bus, 1/2 bandwidth of the Fury X, performs. Nvidia has a long history of gimping their bandwidth compared to AMD, I would never attempt to compare their poor memory performance with anything AMD releases.

I'm more interested in HBM and AMD's miracle driver.
 
TaintedSquirrel said:
It's a lot of extrapolating for a card nobody has touched before.
I don't really care how the 980 and its 256-bit bus, 1/2 bandwidth of the Fury X, performs. Nvidia has a long history of gimping their bandwidth compared to AMD, I would never attempt to compare their poor memory performance with anything AMD releases.

I'm more interested in HBM and AMD's miracle driver.
Click to expand...

Its common sense.
If you run out of memory, frame time increases as it frees up memory and loads more data in.
 
TaintedSquirrel said:
It's a lot of extrapolating for a card nobody has touched before.
I don't really care how the 980 and its 256-bit bus, 1/2 bandwidth of the Fury X, performs. Nvidia has a long history of gimping their bandwidth compared to AMD, I would never attempt to compare their poor memory performance with anything AMD releases.

I'm more interested in HBM and AMD's miracle driver.
Click to expand...

doesn't matter the total amount of bandwidth you have, its how its used, just like any other component in the gpu.

Look at the 6600gt and had half the bandwidth of the 5900 fx, but it kicked its ass.

AMD has to have this miracle driver for memory management if there is one, before reviews come out, not after.
 
What I would like to know is if those games actually need that much VRAM or if the engines are written with such horrible memory control and garbage collection that unused crap just piles up by the GB.

If games (haha fat chance) or drivers can be written to more aggressively clean up untouched assets, you can probably squeeze a lot more useful data into the 4GB space. Yeah, these 8kx8k llama textures have been sitting in memory for 20 minutes and haven't been called since the opening cutscene, dump them.
 
We’ve seen AMD and NVIDIA showing off 12K gaming setups at trade shows all the way back to 2013. Running three 4K Ultra HD displays side-by-side is still impressive looking now was it was then as you are looking at around 1.5 billion pixels a second of gaming goodness (running at 60 FPS). That is a ridiculous amount of pixels!

12K Gaming
Click to expand...

ibhVkm.jpg


The most impressive part about the 12K AMD Eyefinity demo shown this week by AMD running the game title Dirt Rally is that they needed just one video card in the PC to push all those pixels at a playable frame rate, the AMD Radeon R9 Fury X. The frame rate was pushing 60FPS, which isn’t too bad considering what the single $649 graphics card is doing behind the scenes to push out all those pixels at an acceptable rate. AMD informed us that two Radeon R9 290X’s or a Radeon R9 295×2 get around 45-50 FPS on this exact setup. AMD used three Radeon R9 290X 4GB video cards back in 2013 when the Hawaii GPU launched to run Dirt 3 on a similar demo setup.
Click to expand...

http://www.legitreviews.com/12k-gaming-with-one-amd-radeon-r9-fury-x-graphics-card_166585
 
AVogel said:
What I would like to know is if those games actually need that much VRAM or if the engines are written with such horrible memory control and garbage collection that unused crap just piles up by the GB.

If games (haha fat chance) or drivers can be written to more aggressively clean up untouched assets, you can probably squeeze a lot more useful data into the 4GB space. Yeah, these 8kx8k llama textures have been sitting in memory for 20 minutes and haven't been called since the opening cutscene, dump them.
Click to expand...


The thing is even with loading things in the background which many engines do now, you still need certain amount of texture and meshes always loaded up.

I did this before 1 8k texture takes up 64 mb of ram with an alpha channel compressed, a full texture set for an 8k texture can run up to 256 mb and can be more if we are talking about detail textures and what not. Now not all assets will use 8k, usually level assets or floor and ceiling tiles are less. But they add up quickly.
 
TaintedSquirrel said:
I assume that would be 15.15 which came out on the 16th.
Click to expand...

15.15 is what was used in the reviewers guide and it shows far cry 4 in there and we know that game is close to the 4gb limit, without AA Fury X seemed fine but with x2 AA which by now should be used by any hardware, takes a 20% dive.
 
15.15 is only for the 300s though isn't it? Not integrated into the full main release...

Supposedly the magical 15.30 will be the key. May even be WHQL, what a novelty.
 
HBM doesn't mean squat if your card is memory starved. Higher memory bandwidth has seldom had a significant effect on gaming compared to sheer pixel throughput.
On paper FuryX's shader throughput is far more impressive than it's substantially less memory.
It's as if one is saying that 4GB of ultra fast memory is sufficient for my operating system compared to 16GB of decently fast memory.
The available buffer size is always great to have. And with the direction the new games are taking, it is more important to ship more vram for a high end product.

We should be happy that FuryX is a great piece of engineering on the GPU core and memory interface. But HBM2 is where capacity and overclockability of the memory lies.
 
Did you guys not see the min fps on their charts? Hold opinions and conjecture some are assuming as fact until the furys release.
 
I like how they didn't use Mantle for BF4 and barely managed a .4 fps squeaker. If they had use Mantle they probably would have got stomped.

Realistically though, I put that on DICE more so than AMD, and this is one of the reasons D3D12 kind of scares me.
 
razor1 said:
http://www.forbes.com/sites/jasonev...o-card-edges-out-nvidia-gtx-980-in-4k-gaming/

Food for thought the 390 beating the 980 on AMD reviewers guide and we know that didn't happen with independent tests.

Its all about what settings are used.
Click to expand...

yup this is starting to look like the difference will be very slim. I don't see fury being much faster than 980 ti. Yes it is cheaper but it isn't going to force nvidia to drop prices or worry. If it was a lot faster I would think the AMD internal benchmarks would show much faster performance than 9%. But oh well I'll wait out the NDA and see what people say about it on the 24th (not very far away).
 
While the R9 290X with Z97 and DDR3 RAM might have an issue, maybe the Radeon Fury with Z170 and DDR4 RAM won't be having this memory swapping issue ? If the stutter is "minimal" (compared to running out of system RAM), maybe a higher system memory bandwidth and GPU memory bandwidth will solve this ?

I'm not saying it is, but it seems logical to me that that huge pipe going from the GPU to the CPU to the memory should be good for something.
 
Phuncz said:
While the R9 290X with Z97 and DDR3 RAM might have an issue, maybe the Radeon Fury with Z170 and DDR4 RAM won't be having this memory swapping issue ? If the stutter is "minimal" (compared to running out of system RAM), maybe a higher system memory bandwidth and GPU memory bandwidth will solve this ?

I'm not saying it is, but it seems logical to me that that huge pipe going from the GPU to the CPU to the memory should be good for something.
Click to expand...

I think the bottleneck was in the pcie connection. doesn't matter how fast the computer memory is if the bottleneck is somewhere else.
 
If the bottleneck was already in the PCIe connection, than it would matter more than a few % when going from PCIe 3.0 x16 to PCIe 3.0 x8 for discrete GPUs: http://www.anandtech.com/show/8045/asrock-z97-extreme6-review-ultra-m2-x4-tested-with-xp941/11 <-- scroll to "Effect on Average FPS..."

Let's not forget PCIe 3.0 x16 is capable of about 16GB/s, which DDR3 can only achieve with DDR3-2133 or higher. For DDR3 that's high-end memory, for DDR4 that's low-end. 2133 MHz is also where Skylake's supported speed is, while Haswell is specced at DDR3-1600.
 
Last edited:
Phuncz said:
If the bottleneck was already in the PCIe connection, than it would matter more than a few % when going from PCIe 3.0 x16 to PCIe 3.0 x8 for discrete GPUs: http://www.anandtech.com/show/8045/asrock-z97-extreme6-review-ultra-m2-x4-tested-with-xp941/11 <-- scroll to "Effect on Average FPS..."
Click to expand...

None of the games there require >4GB of vram.

So they are testing diffrent scenario

To make a proper test you would need to
1) find games and settings that choke on 4 GB GPU but work on higher amount
2) test them at various PCI-express speeds to see impact.
 
Very true but I do not know of any such test and I don't have the hardware to test it myself.
Mind you that those games were tested not on a 4GB card but a 3GB card, which shouldn't be a problem on 1080p at which the test was run.
 
verylostindeed said:
I sort of doubt they'd go liquid cooling if it wasn't pretty much required though? It doesn't make much sense otherwise.
Click to expand...

it's their work around for the 275w TDP spec standard they have to stay within, but everyone knows high-end motherboards/PSU's and GPU's can support well above that. so you release the card at 275, water cool it and tell people to have at it with overclocking.

can't wait to see what the overclock numbers are though.
 
sirmonkey1985 said:
it's their work around for the 275w TDP spec standard they have to stay within, but everyone knows high-end motherboards/PSU's and GPU's can support well above that. so you release the card at 275, water cool it and tell people to have at it with overclocking.

can't wait to see what the overclock numbers are though.
Click to expand...

so with all that being said, what do you predict the OC'ability will be? like hawaii? or tahiti?

i have this weird feeling that it'll be like hawaii but i can't imagine AMD actually coming out stating that the card overclocks "like a beast" and then come out with another 100mhz tops

we all know that hawaii was pretty much overclocked from factory
 
Reality said:
so with all that being said, what do you predict the OC'ability will be? like hawaii? or tahiti?

i have this weird feeling that it'll be like hawaii but i can't imagine AMD actually coming out stating that the card overclocks "like a beast" and then come out with another 100mhz tops

we all know that hawaii was pretty much overclocked from factory
Click to expand...

1300mhz should be attainable.
 
You must log in or register to reply here.
Back
Top