AMD has been teasing their Radeon R9 295X2 codenamed Vesuvius for quite some time now, including a lengthy ‘viral’ campaign that involved secret agents, semi-ambiguous packages of Volcanic Island water and chips as well as creepy photos of yours truly. Now that the secret is out of the case, we can finally tell you about AMD’s new card and exactly what it is intended to do. First and foremost, this card’s sole purpose is to deliver a single graphics card 4K gaming experience. Something that is currently impossible even with the latest crop of AMD’s Radeon R9 290X and Nvidia’s GeForce GTX 780 Ti. In this review, we’ll be seeing whether or not AMD’s latest and greatest graphics card is capable of delivering true 4K gaming in a single card.
The name of the AMD Radeon R9 295X2 itself doesn’t quite make sense to me in terms of naming when you consider that it is really just two R9 290X’s in a single dual GPU graphics card. I would’ve called it the R9 290X2, but perhaps I’m missing the point here. But when you take into consideration that this card is effective two R9 290X’s strapped onto a single graphics card and you remember all of the thermal issues they had with a single R9 290X it was obvious they needed to go liquid cooling. And they did, so they tapped Asetek for a custom cooling solution to cool both Hawaii GPUs in this Vesuvius card.
The Radeon R9 290X2 has both an LED lit fan as well as an LED lit logo between the two liquid cooling tubes.
The R9 295X2 came to us in a briefcase, which we originally teased in an article, even though we’re not actually sure what the retail packaging for this card will be since these were press samples. Judging by ASUS’ own packaging, it doesn’t look like it’ll be coming in a briefcase like ours did, what a shame.
Looking the card itself in detail, you are getting a dual Hawaii ASIC graphics card that allows for a maximum capacity of 8GB GDDR5 memory. When you compare it side to side against the R9 290X and the R9 290 you can see exactly what you’re getting in terms of performance and raw support.
Furthermore, the compute performance is actually more than 2x the 5.6 TFLOPS that the R9 290X is rated at at 11.5 TFLOPS instead of the expected 11.2 TFLOPS as the GPUs actually work up to 1018 MHz rather than 1 GHz on the R9 290X (the reason for the 295X2 naming?). Higher clocks were enabled thanks to the cooling and improved binning more than anything else, which we’ll talk about next. The overall compute capability of 11.5 TFLOPS is a result of the fact that it has 5,632 cores/stream processors enabling a texture fill rate up to 358.3 GT/s.
In order to combat the issues that AMD had with the R9 290X and overheating causing thermal throttling of the GPU, AMD approached Asetek to help them water cool the entire graphics card. Asetek is very good at making self-contained liquid cooling systems and they’re capable of scaling in the thousands if not hundreds of thousands. So, as a result they helped AMD engineer a solution that would help them not only cool a single Hawaii GPU but two slightly higher clocked Hawaii GPUs, a pretty tall order.
If you look at the card itself, there are a few major parts to the whole system. Primarily the metal backplate, the PCB with the GPUS and memory, the Asetek liquid cooling blocks and full-length cooling plate, and the metal cover and fan. There is also a heat exchanger (radiator) attached to each of the Asetek waterblock/pumps and each block and pump is connected to the other. If the GPUs are already water cooled why have a fan on the cover of the card, some of you might say? Well, that’s because the fan’s purpose is to keep the memory chips cool as well as the voltage regulators in the middle. Remember, voltage regulators typically are the first failing points of most dual GPU cards.
As a result, you get a very very powerful graphics card that is fully contained and self-sufficient, as long as you give it 500 watts of power. That means that this card has two PCI Express 8-pin power connectors and it also means that you need to have the right PSU with enough current on the +12v rail in order to power it safely and in a stable manner. For this purpose, we will be using a Thermaltake Toughpower XT 1475W power supply. Obviously you don’t need that much unless you plan to run two of these cards in a single system.
The back of the card is also populated with four MiniDP and one Dual-Link DVI connector, netting you a total of five monitors, at up to 2560×1600 resolution if you use all 5. However, you could theoretically drive three 4K monitors off of this display – realistically you’d want to have two of these GPUs in order to drive a triple-4K display setup smoothly. Gaming in 25 million pixels? Everything is possible.
That takes us to the GPU’s overall performance and evaluating its capability as a 4K gaming card and whether or not it is actually worth the $1,499 pricetag. Not just that, but how does it utilize the 8GB of GDDR5 and does it run anywhere nearly as hot as the R9 290X with the new cooling solution?
For our testing, we tested synthetic benchmarks, OpenCL benchmarks and game benchmarks. All of these benchmarks were tested with both Nvidia’s and AMD’s latest graphics drivers and AMD’s Mantle was enabled where possible (Battlefield 4). The system we tested the cards on was powered by a Core i7-4960X with 16GB of DDR3-1600 and on an X79 board in an x16 PCIe slot. The PSU, as mentioned earlier, was a Thermaltake Toughpower XT 1475W and we were using a SHARP Sharp PN-K321 monitor to test 4K. We did not test in any other resolution other than 4K because it seems ridiculous to spend $1,500 on a graphics card and use it for anything other than 4K, seriously. Theoretically, you could dedicate it towards compute, but in that case we’ve got you covered.
The tests that we ran in our benchmarking were 3DMark Fire Strike Extreme, Kishonti Compubench (CLBenchmark), Luxmark v2.0, Battlefield 4, Crysis 3 and Counter Strike: Global Offensive.
In our testing, we tested the R9 295X2, R9 290 and the GTX 780 Ti. We had some issues getting the last gen Radeon HD 7990 to work properly with our 4K monitor in the game tests so we had to throw it out and will likely have to figure out a solution to get it work in the future. Unfortunately, that issue presented AMD’s drivers at its worst. However, we do have some non-real-gaming HD 7990 benchmarks in our 3DMark and OpenCL benchmarks to give an idea of performance over the HD 7990.
In 3DMark, which was the first test we ran, we were able to see that the R9 295X2 pretty much came in as the fastest single card that we’ve ever tested by a huge margin. The R9 295X2 scored a 3DMark Fire Strike Extreme score of 8403 in our testing, which bested our GTX Titan SLI setup with 7391 and the HD 7990 which scored 4639. It also beat the R9 290’s 4631 by almost twice, even though the 290X would have scored higher so the R9 295X2 isn’t quite double the performance of the Hawaii GPU in 3DMark. But even so, it is still the fastest multi-GPU setup we’ve tested in 3DMark.
In Kishonti’s CompuBench, the new name for their CLBenchmark (since they’ve added renderscript on mobile) we wanted to see how each card handled compute as Nvidia has traditionally struggled against AMD and here we found some really interesting results. In terms of OpenCL compute, AMD’s Radeon R9 295X2 actually outperformed three of Nvidia’s GTX Titans and obviously the short lived Radeon HD 7990. This card scored 686,475 points which is higher than the three Titans’ 646,233 and the Radeon HD 7990’s 560,695. What is interesting, though is that when compared to the R9 290’s 333,444 score, the R9 295X2 scored more than double, meaning that the R9 295X2 likely scores exactly double that of the R9 290X.
The story was a little different in Luxmark v2.0 however, in the end the R9 295X2 was still much faster than all of the competitive GPU solutions and AMD actually took the top three slots even against three of Nvidia’s GTX Titans. The R9 295X2 is a really powerful card for compute and that is quite evident here with the score of 5350 over the 7970 GHz editions in CrossFireX (4679) and the HD 7990 (4475).
Real World Game Performance
In Battlefield 4 we wanted to get a playable scenario for all of the card that we tested so we set the settings to the high preset rather than the highest preset, which I believe is Ultra. We also opted for 4x MSAA instead of 8x MSAA. As a result, all of the card provided playable settings with the R9 295X2 delivering an average performance of 81 FPS compared to the R9 290 and 780 Ti which both got an average of 40 FPS. Now, considering that the R9 295X2 had an average of 81 FPS we would likely be able to crank the settings up to the ultra preset which would enable us to still have playable graphics at near maximum settings. Thanks to Mantle AMD’s R9 295X2 is able to deliver an impressive amount of performance in BF4 in 4K. During our benchmarking the R9 295X used a maximum of 6.7 GB out of the GPU’s available 8GB of memory, indicating some wiggle room for graphics settings bump, especially with Mantle, which would probably bring the card closer to 8GB.
With Crysis 3 we just wanted to make the graphics cards cry and to see what would happen if we ran the game at nearly the maximum settings possible while still allowing the R9 295X2 to be playable. This was achieved by setting the game settings to Very High and dialing in the MSAA at 4x instead of 8x. As a result, we got the R9 295X2 to perform at an average of 29 FPS while the 780 Ti performed at 20 FPS and the R9 290 at 18 FPS. Even though the game looked absolutely stunning, at these settings in 4K the game was only playable on the R9 295X2 and would’ve required two 780 Tis or two R9 290s to get anywhere near similar performance. What was the most interesting thing was that this game far surpassed our memory utilization on the GPU than Battlefield 4 did. At its peak, the R9 295X2 used 7.5GB of 8 GB of memory, indicating that we had effectively maxxed out the card’s capabilities at 4K.
Thus, if you think that 8GB of video memory is an overkill for a graphics card, think again.
In terms of testing Counter Strike: Global Offensive, we knew that we weren’t going to be getting anything crazy in terms of graphics, but we did know that we would be able to easily play the game at 4K on all of our cards and to get high enough FPS to be able to differentiate between the cards more effectively. Not to mention the fact that this would be a Source Engine game while the others were Frostbite and CryEngine. Personally, I play a lot of 4K so this interested me a lot and since Counter Strike: Global Offensive is on its way to be one of the most played games on Steam, it only makes sense to test it out in 4K and see what kind of performance it would get. We also set the game at its maximum possible settings in order to try to lower the FPS as much as possible and improve the overall visual quality at 4K.
No surprise, at max settings all of these cards managed over 150 FPS. The R9 295X2 managed to deliver an average of 250 FPS (300 is the max set by the Source Engine) while the R9 290 managed 163 FPS and the 780 Ti 148 FPS. So, with the R9 295X2 you’re basically getting almost 100 FPS over the single GPUs in 4K. And in this scenario the memory utilization of the R9 295X2 was expectedly much lower than the other two games with it only utilizing 3.6 GB of the 8GB of memory.
That wraps up our performance testing. But before we move on to thermals we wanted to talk about the reason why we mentioned memory utilization. We wanted to illustrate the purpose of going with an 8GB card rather than a 16GB card, 8GB is pretty much enough in almost any scenario unless you believe that you are going to be running Crysis 3 at 4K. Anyone trying to sell you a card with more than 8GB is probably selling you more memory than you need and if you’re not running 4K you probably don’t need more than 3 or 4GB. Just keep that in mind as add-in board partners try to charge more for memory that you’ll never use.
We also wanted to note that even with AMD’s latest drivers and a custom CrossFire profile, certain benchmarks simply would not run dual GPUs. We also had some problems with configuring the 4K monitor with the R9 295X2, however, we were able to work those out. To us, it seemed that this card’s drivers still needed some work, but were playable in the three games that we tested them in. Also, AMD’s frame pacing appeared to work well as we didn’t notice any stuttering or nearly as much tearing as we had with previous AMD dual GPU setups.
Now, getting back to the card, we wanted to talk about the thermals. For me, when I first heard about this card and its possible existence my first concern was cooling and temperatures. AMD really struggled a lot with the R9 290X in terms of perception because the card topped out at water-boiling 95 Celsius under extreme loads and as a result actually throttled performance. Of course, their add-in board partners eventually released better cooling solutions that alleviated this issue to a degree, but that concern came back again once I heard that AMD would be making what was supposed to be a dual Hawaii GPU. To be honest, my first thought about a dual Hawaii card was that it had to be dual 290s and that it would have to be water cooled.
When I found out that this card would be two slightly higher clocked 290X’s I was a little concerned about how it would perform thermally and if they could handle the temperatures generated by these GPUs. Even when I saw the cooling solution I wasn’t sure that the 120mm Asetek radiator would be able to handle a maximum of 500W of GPU power dissipation. After all, most CPU coolers that are cooling 130W CPUs also have 120mm radiators and are effectively cooling half the TDP.
Upon doing all of my testing, I was able to conclude that the first GPU at idle was running at a normal temperature of around 36C which is pretty good for a normal graphics card, but fairly expected for a water cooled graphics card. Now, what really caught me by surprise was the temperature of the first GPU (the most heavily used GPU) under maximum load. The GPU never got hotter than 67C, even when the radiator and tubes were warm to the touch. Because of that, this card is capable of delivering a very stable level of performance and to do it consistently time and time again. I don’t know exactly what the guys at Asetek and AMD did, but they have engineered a pretty fantastic cooling solution that I frankly thought might be underpowered for these GPUs.
The Radeon R9 295X2 is a $1,500 card. There’s no getting around that. In fact, it makes a lot of sense if you think about it. The R9 290X should sell for around $550, even though there’s still a fair amount of gouging as a result of last year’s cryptocurrency rush. However, Newegg is selling a Visiontek R9 290X for just under $600 with a non-reference cooler. Now, if you take the $550 price and assume it is multiplied by two, you’ve already arrived at an $1,100 card purely by the assumption of 2X the price. However, this card is also water cooled, which normally carries a $100-$200 premium on any graphics card, bringing the expected retail price up to about $1,300. However, most dual GPU cards rarely sell for 2X the price of their base-model single GPU parents and Nvidia announced the Titan Z was going to be selling for $3000. This gave AMD some wiggle room on price and I suspected that they would be selling it somewhere betwen $1,300 and $1,500. And here we are, at $1,500.
The card itself is when compared to buying two R9 290Xs simply does not provide more value than owning two cards. However, it is significantly quieter and cooler than running two R9 290Xs and it provides a pretty high level of gaming performance in 4K with a single graphics card. However, in order to properly make good use of either this GPU or any dual 290X setup, you should really be looking at a 4K display. So, this card makes sense if you want the absolute fastest or if you want to be able to cram four Hawaii GPUs into a board with only two PCIe slots or if your case simply isn’t big enough to fit four GPUs. Not to mention, cooling four Hawaii GPUs in any case on air would be significantly more difficult than two of these Vesuvius cards. In fact, you would be spending around $2,300 on four 290Xs and would probably have a much hotter and louder setup than if you spent the extra $700 and went with two 295X2s. So, there is some sense in going with these cards if you absolutely have to have the fastest and the best and use 4K.
In terms of value, just have one thing in mind: for the price of a single GeForce GTX Titan Z, you can purchase a R9 295X2 and not one, but TWO 28″ 4K 60Hz panels from Samsung. If you dislike the TN panel, you can still afford a 24″ IPS 4K Panel from Dell. Talk about a value deal.
I have to say that I was personally quite impressed with this card itself, especially on the performance and thermals side of things. If you want the absolute latest and greatest, not to mention fastest, fastest in the world, you have to buy the R9 295X2. And if you want to buy just one card for 4K gaming, this is it. And because of that, we’re awarding this card our editor’s choice award.