Amd fury x review - Reviews 2024

A little before nVidia launched its GTX 980Ti and with the Titan X as the leader in monoGPU performance, AMD decided to launch what will be its flagship for the moment, a graphic with its own name instead of numbers, the Fury X. This graphic comes with some risky decisions, such as the inclusion of a liquid kit in the reference model (the first time in a non-dual graph) , and the incorporation of the most advanced memory technology available, being the first GPU to mount HBM memories compared to GDDR5, multiplying bandwidth. Let's see if it's enough to stand up to your arch enemy.

We thank the AMD Spain team for the loan of this graphics card for analysis:

Technical characteristics

Model	AMD Radeon R9 Fury X
Shaders	4096
Texture Units	256
ROPs	64
Boost frequency	1050MHz
Memory Frequency	1000Mhz (cash)
Memory Bus Width	4096 bits
VRAM	4GB
FP64 performance	1/16
TrueAudio	Y
Number of transistons	8, 900 million (approx.)
Typical consumption	275W
Fabrication process	TSMC 28nm
Architecture	GCN 1.2
GPU internal designation	Fiji
Release date	06/24/15
Launch Price	~ € 700

AMD Fury X

The graphic is presented in a luxurious box in the style of the 295X2, in which most of it is protective insulator to avoid failures due to insufficient care in transport

Already inside, the small size of the card stands out, and at the same time the generous size of the tubes and the radiator. We must note that both the supply and the tubes that go to the radiator and the fan are meshed and give a lot of quality to the whole. In the foreground, the two 6-pin connectors that power this graphics card

The design is an almost perfect mix of understated, elegant and modern. It is a graphic with a very, very small size, which will delight users of small boxes that have room for a 120mm radiator

The upper is finished in rubber with a perforated texture, a good choice to absorb noise from the pump, but somewhat dirty to the touch. As you can see from the photos, even with little use time like our review sample, you see an uneven texture that is difficult to remove completely.

Detail the back of the card, we see that they have included a backplate, discreet and with good quality, which improves the appearance and solidity of the set. In this case, without stickers that cancel our guarantee to the slightest carelessness

As we anticipated in the specifications, it is a chip with a lot of muscle, in fact, in gross power, it is the most powerful chip ever integrated into a consumer GPU, with a whopping 8601.6 GFLOPS (1050mhz x 4096 shaders x 2 operations / cycle) in simple precision, compared to the already impressive 6144 of the Titan X.

We must also say that it is a graphic with a moderate consumption but quite efficient, despite being built in the same manufacturing process at 28nm as its predecessors. It does not reach the performance levels per watt of the nVidia Maxwells, but it is really close, leaving behind graphics such as the 780Ti, which next to it look like a real oven. Much of this improvement is due to the HBM memory, which greatly reduces consumption compared to the usual GDDR5

The rest of its characteristics are those expected for a high-end GPU, except for the 4096 bits of the memory bus that undoubtedly surprise. We have 4GB of VRAM, memory frequency of 500mhz (1000 effective) and XDMA CrossFire support (without cables between the GPUs). The graphics chip has 4096 shaders running at 1050mhz boost.

The graphic is a closed block, which incidentally makes good use of the interior space to house the pump and at the same time cool the VRMs. The radiator is 120mm, with a very good quality fan manufactured by the well-known Nidec.

We note that the graphic is very quiet in charge compared to other tested models, of which more with the reference heatsink, with a light whine coil but very discreet. However, it is at rest when the graphics are somewhat annoying, since the noise of the pump, although soft, is quite sharp and penetrating, and will sound more than the rest of the PC in any minimally silent equipment. From what we have been able to see, AMD has reduced this noise at rest from the first review models in the models that have gone on sale, but at the moment we cannot confirm it since we have only had access to this unit.

The graphic incorporates LEDs that inform us of the consumption of the GPU (and therefore, approximately, of the workload that it has), as well as another green LED apart that lights up when the graphic enters ZeroCore mode to lower consumption near zero when not in multigpu setups.

Like all AMD GPUs of recent times, the graphics includes dual BIOS and a switch to select them.

The power is provided by two 8-pin pciexpress connectors, unlike the reference model of the 980Ti with 8 + 6, a very common configuration for high-end graphics, which we will also see in custom models of nVidia GPUs.

Unfortunately in this case the rear connections are abundant but not as varied as I would like to see, with the inexplicable absence of a DVI port to support the large number of users who still use these monitors. The connections that are included are sufficient, with 3 displayport ports and an HDMI. The HDMI port is revision 1.4a, so in case of opting for that port we are limited to 30Hz in 4K resolutions.

Luckily, the most complete connection for the future is Displayport, just the one we have 3, supports freesync, and in case you need a DVI we can take it out with adapters. Likewise, I find it very uncomfortable that a graphic in this range does not include DVI, or at least that HDMI is 2.0 to offer an alternative to DP.

About HBM memory

One of the riskiest decisions made in this graph is undoubtedly the inclusion of HBM memory, which will undoubtedly represent the future in terms of high-speed memory, and precisely GPUs are the devices that most appreciate a high memory bandwidth.

These are chips mounted in several layers, saving space on the PCB and reducing consumption and cooling needs.

We must note that the "fault" that this graphic only mounts 4GB of RAM is to bet on this technology, since the largest HBM memory chips allowed by current SK Hynix technology are 1GB. This graph mounts 4 of these chips, each with 1024bits of bus. Adding two more chips would have raised the complexity of the GPU and its cost to really absurd limits, and we are already dealing with a really complex chip, with 8300 million transistors (something more than the GM200 of the Titan X).

The restriction is only in size, as the memory bandwidth is simply brutal, with 512GB / s of bandwidth (1000 * 4096/8), compared to 336GB / s for the Titan X and the 980 Ti. The real contribution in performance is something that we will not be able to know, since there is no version of this graph with traditional GDDR5 memory to compare, nor any other GPU that uses this technology at the moment, but we can venture that without a doubt around 10 -20% extra than you would have done without it.

In my opinion, 4GB is more than enough today for resolutions up to 4K without filters, and if this choice was necessary to compete on equal terms with the 980Ti as this GPU manages to do, it has been a good decision for the market. Face the user, probably many will not care, but I personally would not have cared to wait a little longer and have a model with 8GB on a chip that does take advantage of it, unlike the new 390X, where 8GB does not make sense except in huge multigpu setups.

WE RECOMMEND AMD prepares a new driver for the Linux kernel

Test bench and performance tests

TESTING BENCH
Processor:	Intel i7 [email protected]
Base plate:	Asus Rampage V Extreme
Memory:	DDR4 Ripjaws4 4x4gb 2666MT / S CL15
Heatsink	RL Custom, EK Supremacy EVO
HDD	Samsung 850 EVO 1Tb
Graphic card	AMD Fury X
Power supply	Antec High Current Pro 850W

We will use the benchmarks of 3 games to evaluate the performance of this GPU. As we anticipated in the review of the 980Ti, the comparison between these two graphics is the most interesting, since they are close GPUs in benefits and with a very similar sale price.

In 3DMark we find a really good performance, with 14234 points, very close to a 980Ti of stock. In the tests we did with overclocking, reaching 1150mhz, they showed results higher than the 980Ti in stock, but our sample did not seem 100% stable at that frequency, so we finally decided not to include them in the comparison.

The results of the Metro: Last Light are equally good, in this case dropping below 60fps sporadically but with a fairly fluid and playable experience as well. We must say that this is a very demanding test with filters at a level much higher than usual. The 980Ti is above this graph again, but it's not a bad result at all, and this particular game usually favors nVidia. In the case of the Tomb Raider 2013 it usually happens the other way around…

… But surprisingly, this is not the case this time. It's funny how using TressFX technology to recreate hair was so favorable to AMD the last generation of graphics, with the Titan falling below the high-end AMD, while now with Maxwell it seems that nVidia has done its homework already. regained lost ground. So another victory, without much advantage of course, for the 980Ti.

Final words and conclusion

Probably not AMD's biggest triumph, but this graphics is certainly a milestone in the design of consumer GPUs, as it pioneered its memory, and that's something no GPU can say, not even the Titan X.

Unfortunately nVidia is also hitting this price point hard, and this Fury X has a few minor glitches that can tip the scales, without going any further, the noise of the pump.

The Fiji chip that mounts the Fury X is without a doubt a GPU with a lot of muscle, and I would not be surprised to see it above the 980Ti in the long term with titles that use DX12 and are demanding in terms of memory performance. Currently and at the same price, I can't help but see the 980 Ti as a slightly superior product on too many little things to recommend this GPU instead.

If it cost € 100 less, it would be a very clear victory for AMD, and his score would have been even better. We will see if the Fury really deserves this position. At the moment we have at least one other competitor of the Titan that costs a little more than half, which is not a little.

ADVANTAGE	DISADVANTAGES
+ MEMORANDUM TECHNOLOGY, THE FIRST GPU IN THE WORLD WITH HBM	- PUMP NOISE SLIGHTLY HIGH AT REST
+ DRASTICALLY IMPROVED CONSUMPTION FROM THE PREVIOUS GENERATION	- AMOUNT OF MEMORY (4GB) LESS THAN THAT OF ITS MAIN COMPETITOR
+ VERY REDUCED SIZE, IDEAL FOR LITTLE BOXES AVAILABLE	- HIGH PRICE, LITTLE VARIETY OF VIDEO OUTPUTS
+ ENOUGH SILENT ON LOAD
+ VERY GOOD PERFORMANCE, EVEN AT RESOLUTIONS ABOVE 1080P