Nvidia GTX 990 Ti 24GB And AMD R9 395X2 16GB Benchmarked and Reviewed

Today’s a special day as we gracefully welcome two new flagships from Nvidia and AMD. The GTX 990 Ti 24GB and the R9 395X2 16GB, both reviewed and benchmarked !

We’ve anticipated this launch for many months now and they’re finally here. The fastest graphics cards to ever come out of Silicon Valley. Both featuring two massive GPU cores and elaborate cooling solutions to ensure that everything is operating under the appropriate thermal limits once you open the taps and run the latest games at maximum settings.

Nvidia GTX 990 Ti 24GB And AMD R9 395X2 16GB Benchmarked and Reviewed

So without any further delay let’s get straight into the juicy bits !

Nvidia GeForce GTX 990 Ti 24GB

First off we have the GTX 990 Ti, Nvidia’s next dual GPU flagship. Featuring two fully fledged GM200 GPUs with 3072 CUDA cores each for a total of 6144 CUDA cores. Twice that of the Titan X, featuring a 768bit wide bus or in other words 2x384bit and a whopping 24GB of 7Ghz GDDR5 VRAM, delivering 336GB/S of memory bandwidth per GPU.

NVIDIA GM200 Block Diagram

Both GM200 GPUs are clocked at 1216Mhz for the base and 1352Mhz for the boost, making this is by far the highest clocked graphics card Nvidia has ever introduced. And the most powerful as well, in fact it’s more than twice as powerful as the Titan X, completely eclipsing it in performance. And the best part, the card is all yours for an eye watering $1999.

	NVIDIA GeForce GTX Titan X	Nvidia GeForce GTX 990 Ti	NVIDIA GeForce GTX 980	NVIDIA GeForce GTX 970	NVIDIA GeForce GTX 960
GPU Architecture	Maxwell	Maxwell	Maxwell	Maxwell	Maxwell
GPU Name	GM200	2xGM200	GM204	GM204	GM206
Die Size	601mm2	2x601mm2	398mm2	398mm2	228mm2
CUDA Cores	3072	7144	2048	1664	1024
Texture Units	192	384	128	104	64
Raster Devices	96	192	64	64	32
Clock Speed	1002 MHz	1216 MHz	1126 MHz	1051 MHz	1126 MHz
Boost Clock	1089 MHz	1352 MHz	1216 MHz	1178 MHz	1178 MHz
VRAM	12 GB GDDR5	24 GB GDDR5	4 GB GDDR5	4 GB GDDR5	2 GB GDDR5
Memory Bus	384-bit	2×384-bit	256-bit	256-bit	128-bit
Bandwidth	336GB/S	336GB/S per GPU	224GB/S	224GB/S	112GB/S
Memory Clock	7.0 GHz	7.0 GHz	7.0 GHz	7.0 GHz	7.0 GHz

So let’s discuss the cooling solution that Nvidia has decided to use. We’re looking at two vapor chambers, each sitting on top of a GM200 GPU cooled by a 90mm central fan. The cooling solution you see above is how the card actually looks. Interestingly enough Nvidia decided that adding a metallic shroud similar to the Titan X or Titan Z would merely obstruct air flow elevating the thermal and acoustic footprints. According to Nvidia this new patented open air cooling design which Nvidia dubs ThermX™ is 23% more efficient and 13% quieter than traditional solutions.

This cooling solution also happens to be lighter so the card will exert less pressure on your motherboard and not exhibit any sag overtime. We like what Nvidia has done here, putting substance over style and functionality over aesthetics. Clearly targeting a more mature audience.

AMD Radeon R9 395X2 16GB

Moving on to the Red Team. This is the first time that we’ve ever got to see the R9 395X2. This is a dual GPU card just like the 990 Ti. Featuring two Fiji XT GPUs each rocking 4092 GCN cores for a total of 8184 GCN cores. Each GPU features a 4092bit wide IO memory interface feeding 8GB of stacked HBM1 for a total of 16GB of HBM running at 1.25Ghz. Resulting in a massive 640GB/s of bandwidth per GPU, twice that of the GTX 990 Ti. Each Fiji XT GPU is clocked at 1080Mhz which once again makes this the highest clocked graphics card ever produced by AMD. And easily the fastest, outperforming AMD’s previous R9 290X by nearly 300%.

	AMD Radeon R9 395X2	AMD Radeon R9 295X2	AMD Radeon R9 390X	AMD Radeon R9 290X
GPU Architecture	GCN 2.0	GCN 1.2	GCN 2.0	GCN 1.2
GPU Name	2x Fiji XT	2x Hawaii XT	Fiji XT	Hawaii XT
Die Size	2x557mm2	2x438mm2	557mm2	438mm2
Stream Processors	4092	2x 2816	4092	1664
Texture Units	2×256	2×176	256	176
Render Output Units	2×512	2×64	512	64
Clock Speed	1080 MHz	1000 MHz	1080 MHz	1000 MHz
VRAM	16GB HBM1	8GB GDDR5	8GB HBM1	4GB GDDR5
Memory Bus	2×4092-bit	2×512-bit	4092-bit	512-bit
Bandwidth	640GB/S per GPU	320GB/S per GPU	640GB/S	320GB/S
Memory Clock	1.25Ghz	5.0 GHz	1.25Ghz	5.0 GHz

So let’s take a quick look at the cooling solution AMD has chosen for the R9 395X2.

We can see that AMD has opted for a very similar cooling solution to the R9 295X2. Although the keen eyed among you will notice that the card is significantly shorter than its older sibling. That’s mainly thanks to the stacked HBM memory which doesn’t take any space on the PCB. In contrast, GDDR5 chips and the wiring necessary to connect them to the GPU occupy a significant portion of the PCB.

The card overall is also lighter than the previous R9 295X2 and the pump operates at a lower RPM thanks to the highly efficient Fiji XT GPU. In addition to a new technology AMD dubs FreeFlow™ which allows the cool air from the fan to travel along grooves inside the cooling shroud that work to tunnel air to cool the most thermally sensitive components. The result AMD says is more stable operation, 18% lower temperatures and 12% less noise. All of this is yours for another eye watering $1999.

Game Benchmarks

So let’s see how these cards perform running the latest and most demanding games at maximum settings and torturous resolutions.
Let’s start off with the ever more popular Minesweeper. Running the game with all the settings maxed out at the popular resolution of 360×240 the R9 395X2 averages just over three thousand FPS while the GTX 990 Ti averages just below three thousand FPS. Both graphics cards produce a very playable and smooth overall experience.

Solitaire is slightly less demanding than Minesweeper. With all the settings cranked up to the max running at 360×240 the game looks absolutely gorgeous. The 395×2 averaged 3720 FPS while the 990 Ti averaged 3816 FPS.  Both graphics cards produce an equally lovely gaming experience.

Now let’s look at how the cards perform when we present them with a real challenge. Cranking up the resolution to 32K or what’s otherwise known as 640×360. Minesweeper becomes an extremely demanding game with both GPUs averaging below two thousand FPS. I’d barely call this playable, while the game looks beautiful at 32K it isn’t as smooth as it was at 320×240. If you want to play at 32K we’d advise you to consider using two cards in a quad multi-GPU setup to attain the best possible gameplay experience.

Moving on to Solitaire at 32K, again the game is absolutely gorgeous at such a high resolution. You won’t be able to differentiate how the playing cards look in the game from actual playing cards. Again this is an extreme case scenario as not many people will play at a resolution this high. Both graphics cards worked really hard to output a very cinematic 2.4 and 2.5 thousand FPS.
For our thermal and acoustic testing we used a Corsair C70 case with two intake fans and one exhaust fan. We tested both graphics cards using the most intensive test we have which is Solitaire at 32K.

Solitaire at 32K is really demanding as we’ve established above, causing both cards to heat up to above 30c. Obviously we’re looking at an extreme scenario here. Not a whole lot of gamers will play Solitaire at 32K, nor will you experience temperatures as high as this while the GPU is idling. So I really wouldn’t worry too much about it.

Moving on to the noise test, again inside the same C70 case. The same case without either GPU installed produces 21dB of noise.

Amazingly, both cards actually reduce the total noise output from the same system without the graphics cards installed. A testament to the active noise cancelling effect of Nvidia’s ThermX™ and AMD’s FreeFlow™ technologies.

Conclusion

The final conclusion is that it’s pretty much a toss-up between both cards. Frankly you’ll be happy with either of them. And if you have $1999 to spend on a graphics card I’d not waste time trying to pick between them, flip a coin and enjoy your new game munching monster.