AMD released their R9 Fury back in July 2015, and our review at the time was fairly positive with concerns mainly leveled over the pricing. Today we look at the Sapphire Nitro R9 Fury 4GB – a new triple fan custom solution featuring enhanced out of the box clock speeds.
Sapphire have earned a reputation for creating some of the finest AMD gaming cards. In their supporting literature they claim the Nitro R9 Fury is ‘The Fastest and Quietest on the market'. Partners are keen to focus on noise levels after well documented concerns over related coil whine problems.
During the write up for this article, Sapphire detailed that the Fury card prices will be getting more competitive. We were told that the Nitro R9 Fury OC will be available shortly at around the £430 mark in the UK. This puts the card into the same price zone as the high end custom Nvidia GTX980's (HERE).
GPU | R9 390X | R9 290X | R9 390 | R9 290 | R9 380X | R9 380 | R9 285 | Fury X | Fury |
Launch | June 2015 | Oct 2013 | June 2015 | Nov 2013 | Nov 2015 | June 2015 | Sep 2014 | June 2015 | June 2015 |
DX Support | 12 | 12 | 12 | 12 | 12 | 12 | 12 | 12 | 12 |
Process (nm) | 28 | 28 | 28 | 28 | 28 | 28 | 28 | 28 | 28 |
Processors | 2816 | 2816 | 2560 | 2560 | 2048 | 1792 | 1792 | 4096 | 3584 |
Texture Units | 176 | 176 | 160 | 160 | 128 | 112 | 112 | 256 | 224 |
ROP’s | 64 | 64 | 64 | 64 | 32 | 32 | 32 | 64 | 64 |
Boost CPU Clock | 1050 | 1000 | 1000 | 947 | 970 | 970 | 918 | 1050 | 1000 |
Memory Clock | 6000 | 5000 | 6000 | 5000 | 5700 | 5700 | 5500 | 500 | 500 |
Memory Bus (bits) | 512 | 512 | 512 | 512 | 256 | 256 | 256 | 4096 | 4096 |
Max Bandwidth (GB/s) | 384 | 320 | 384 | 320 | 182.4 | 182.4 | 176 | 512 | 512 |
Memory Size (MB) | 8192 | 4096 | 8192 | 4096 | 4096 | 4096 | 2048 | 4096 | 4096 |
Transistors (mn) | 6200 | 6200 | 6200 | 6200 | 5000 | 5000 | 5000 | 8900 | 8900 |
TDP (watts) | 275 | 290 | 275 | 275 | 190 | 190 | 190 | 275 | 275 |
The AMD R9 Fury GPU is built on the 28nm process and is equipped with 64 ROPS, 224 texture units and 3,584 shaders. The more expensive Fury X has 64 ROPS, 256 Texture units and 4,096 shaders. They both have 4GB of HBM (Hynix) memory across a super wide 4096 bit memory interface. Bandwidth rating for both cards is 512 GB/s.
The utter power of the AiB 980Ti’s still stuns me to this day! nVidia did a stunning job on their final 28nm flagship imo…
With a price drop this card will be a damn good deal. Not as powerful as a 980ti, but cheaper.
I will agree, but i like what amd are doing too, i think both companies are releasing good products, although a lot of problems exist with both, they are finally making major investment,
there are 8gb versions in the pipeline, so crossfire those, and you have a sub £1k monster, able to hit 4k rez at 60fps.
Can you cite your source for that?
I’m not trying to be argumentative, I’d just like to read where you hear that there are 8GB Fury cards coming, as I’m under the impression that the Fiji memory controllers and the physical structure of HBM1 together impose a 4GB hard limit.
Heyyo, tbh sounds more likely that he is talking about the Fury X2 which is 8GB total… But that’s split between GPU Cores, so it’s 4GB per core. The only time it will act as a 8GB solution is games that will take advantage of explicit multi-adapter… From how hard it seems to implement according to Oxide Games and their use of it in Ashes of the Singularity? Odds are it will be a rare occurrence… So I wouldn’t bank on that 8GB of total VRAM.
Heyyo, WHOA! Dat Ashes of the Singularity though! Super impressive that it ties the Fury X and GTX 980 Ti reference card… Now I just wish there was Fable Legends also to benchmark… Hmm I wonder why Microsoft never released that benchmark to the public??? Sad face I am making…
Good enough that I bought one. EVGA does a super job with their factory overclocked cards that are only $10 more for a 15% o/c. Cool, quiet, fast, reliable, with excellent drivers. What more could I ask for?
AMD gave a great effort with the Fury X/Fury line but it really didn’t help their business all that much. AMD took a major loss when the GTX980Ti was released at a price level $150 less than what AMD was going to charge for the Fury X. AMD had to match the price and the AiB were furious and did not make many cards available. Between that, the hassle of a water cooler, and the coil and pump whine for about the same benchmarks as what NVidia was offering made folks shun the Fury and embrace the 980Ti. AMD had a real chance here too.
The GTXs suck with async compute. If you test AoS with AA (async compute) at 4k, you will understand why Nvidia asked reviewers not to use AA in benchmarks. Check out the Titan X and the r9 390 scores at 4k, crazy settings and Msaa x4. Same settings except the Titan X uses a newer version of AoS, the R9 uses an older version of AoS.
https://www.youtube.com/watch?v=JS0rhMYmehE
https://www.youtube.com/watch?v=PqPj4N9H77A
Hi there,
I have the Nitro card but somehow I cannot access the OC+ Bios, help!
hbm 2 fury
Heyyo, that information isn’t correct dude. It’s not that NVIDIA “sucks” at async compute… it’s apparently not implemented yet despite what their drivers looked like. http://www.overclock.net/t/1569897/various-ashes-of-the-singularity-dx12-benchmarks/2130#post_24379702
The situation hasn’t changed yet either. NVIDIA still don’t have an async compute driver out and so far it’s unclear if it’ll mean native async compute performance or something similar to dynamic recompiling as we saw with DirectX 9 and Shader Code 24bit to 16bit (NVIDIA’s FX 5000 series at the time didn’t support 24bit natively, only 16bit, thus lost rendering time dynamically recompiling and lower framerates) which was evident when Half-Life 2 released. You can find an article about it if you Google Maximum PC shader code NVIDIA. The following generation of cards? Yes, the NVIDIA Geforce 6 series did natively support shader code 24bit.
With that said? I wouldn’t buy a Maxwell GPU as odds are since nothing has come forward yet about async compute working on Maxwell? It’s not worth the risk to get it. NVIDIA Pascal might have native async compute support… but same thing, I wouldn’t buy one until there was definitive proof from multiple review sources. Same idea as the beginning of the Dx9 era… I wouldn’t risk it unless I knew for certain it was working.
http://www.3dmark.com/fs/7490341 I7-6700K 4.6Ghz Kraken X61 DDR4 16GB 4400 Mhz XFX Fury x 1080 Mhz Sabertooth mark 1 z170
http://www.3dmark.com/sd/3810052
AMD’s expensive Fury is Not a good deal when compared with the many factory OCed versions of the performance per watt champion Maxwell GTX980Ti. HBM1’s bandwidth is wasted on 28nm GPUs, and with only 4GB it’s a 4K lame game gimped.
VERDICT: NO thank you AMD, I recommend a factory OCed GTX980Ti which is by far the biggest bang for your GPU buck.
It is mentioned in the reviews, but no one says what position is what:
1. the default BIOS of TDP 260w and a target T of 75C,
2. the second BIOS of TDP 300w and a target T of 80C!
Namely, light on is what (pressed), light off (unpressed) is what??
In my testing in both positions temperature under stress stopped at exactly 79C?? And that left me puzzled in the dark…?
Those aren’t launching until some time 2017. The landscape will most likely look completely different by then.
That score is a bit low, easily deserves a 9 imho. Fantastic cooler and performances, at the right price would’ve been a 9.5-10.