Having the fastest video card is a matter of prestige, and both NVIDIA and AMD know it all too well. Even though it’s more about publicity, rather than practical application, all companies want to be on top when it comes to performances. True, often that success comes with a price so large that many of the products were presented for newspapers and afterwards, and not as many were actually sold. Luckily, the latest generation of products that both companies made was intended to reach the top in a different way. AMD actually postponed the publication of its most modern GPU, which is at the same time the spiritual successor of the Radeon HD 6990. Therefore, we are now presenting the long-expected Radeon HD 7990, video card that strives to be the first when it comes to performances. True, much like the sport cars, there won’t be many people who have them, but we’re really interested to see it in action. We had the chance to drive the Ferrari of video cards, and it was an exciting ride.
Two engin-... I mean… GPUs
The basic idea remained the same. At the core of Radeon HD 7990 there are two Tahiti GPUs. Those are AMD’s most powerful graphic chips. They’re made in 28-nanometer lithography, and they’re based on GCN architecture. Aside from the fact that they support all of the current standards, like DirectX 11, ShaderModel 5.0, Eyefinity 2.0 etc., these chips are also extremely powerful pieces of hardware. Each one has 2048 Stream processors, 128 texture and 32 ROP units. When you double all of that, it’s clear what kind of “arsenal” this video card has. Since AMD always packed two graphic cards on one large circuit board, it wasn’t that hard to assume that it’d have two complete sets of memory units. Each GPU has its own 384-bit memory bus, which form twelve GDDR5 chips with 3 GB of capacity. Technically, because of them you can say that HD 7990 has 6 GB of local video memory! Often is the case that, due to implementation of the most powerful GPUs with maximum capacity of the memory, frequency has been lacking. Reason for that is thermal dissipation of the implemented components. Simply put, high frequencies with this hardware seem like a crazy idea, but AMD took everything into accounting. Namely, there were no compromises in any area. 1000 MHz clocks for two Tahitis and effective 6 GHz for memory seem a bit too much.
|Specification||AMD Radeon HD 7990|
|Manufacturing process||28 nm|
|GPU frequency||1000 MHz|
|Number of Stream processors||2x 2048|
|Number of OP / textural units||2x 32 / 2x 128|
|Memory||GDDR5 / 2x 3GB / 2x 384|
|Memory frequency / effective||1500 MHz / 6000 MHz|
Let us remind you, one HD 7970 has consumption of about 250 W, and since HD 7990 is, basically, two cards, total TDP can be up to 500 W! If you thought about treating yourselves to a card such as this one, you should probably by a new power unit, too. Jokes aside, for this card we really had doubts. Still, when you take a look at the cooler, you can assume why the manufacturer was so confident. Three 90 mm radius fans look like they can safely cool the entire PC, not only the video card. Namely, due to their specific design, the air flow is maximized, and noise was reduced. Up until now, cooling always relied on one fan and because of that fact, acoustic and thermal results always had a barely passing grade, rather than an excellent one. AMD realized that and we can only say one thing – finally! Still, what’s located underneath the casing of the three fans is also really important. There are two massive and separated cooling units. Ribs are densely packed, and four copper heat pipes are conducted through both parts. In this way, heat is evenly distributed throughout the whole surface, and each GPU has its own cooler, while they only share the fan power. Even PLX chip got its own cooler, which is somewhere in between. PLX was necessary to compensate the fact that these two logic cards share on PCI-Express 3.0 slot for connection to the rest of the PC.
Considering it’s the latest model, data transfer speed between the GPUs, according to AMD, reaches 96 GB/s. We should also mention the fact that the card has a CrossFireX connector. That means that you can pair up two Radeon HD 7990 cards and get even four Tahiti GPUs in one system. There is one DVI port and four mini DisplayPort connectors. Because the use of this card for gaming on only one screen doesn’t have a point, even with 1440p or 1600p resolution, the key is in Eyefinity technology. Namely, Radeon HD 7990 is used best in combination with at least three 1080p screens. Simply put, gaming will be a real pleasurable experience, and later on, you can return the invested money through the use of hardware acceleration. Many would think that his card will be ideal for use in Bitcoin mining, because there won’t be a faster card than this one in the near future. We should also mention two 8-pin molex connectors, which are necessary in order for HD 7990 to work right away. Huge consumption is held under control by using ZeroCore technology. As is the case with CrossFireX configurations, as soon as it’s not needed, the other GPU shuts down. In that way, the card will act as one HD 7970 for as long as the load isn’t huge, i.e. until you start playing games. So, despite the TDP of about 500 W during gaming, in reality, the card will require far less power.
Like we’ve already mentioned, using this card was a real pleasure. Not only from the aspect of flawless performance. We’re delighted to say that HD 7990 was silent during testing, as any other video card with a good cooling system. Maximum GPU temperature was about 75 degrees, and even in prolonged sessions of gaming and testing it was pretty quiet. True, it’s noisier than a middle-class card, but in a quality tower, it will be hard to differentiate its sound from the rest of the components’. That was impossible until now, and that’s why the use of three fans was a definite improvement. We need not say anything more about performances. What you would have got with two Radeon HD 7970 cards, you get with one HD 7990. That’s something, right? Everything that was impossible with one GPU is possible with two. From playing games in 1440p and 1600p resolution, to excellent performances even on multiple screens.
|1920x1080 / 2560x1440, 4xAA 16xAF||AMD Radeon HD 7990||NVIDIA GeForce GTX 670|
|Futuremark 3DMark (Fire Strike) GPU Score||11245 / 6886||5112 / 3082|
|Futuremark 3DMark 11 (Extreme preset) GPU Score||5395 / 3290||2910 / 1722|
|3DMark Vantage (Extreme preset) GPU Score||34879 / 22848||19796 / 11812|
|Unigine Valley 1.0 (DX11, Ultra, tesselation off) [fps]||88.8 / 54.5||50.1 / 30.8|
|Unigine Heaven 4.0 (DX11, Ultra, extreme tesselation) [fps]||74.8 / 47.3||55.5 / 38.5|
|Crysis Warhead (DX10, Enthusiast) [fps]||87.4 / 73.9||63.2 / 56.1|
|Crysis 2 (DX11, Ultra, HD Textures) [fps]||135,3 / 89,3||70.5 / 40.6|
|Metro 2033 (DX11, very high, tesselation) [fps]||85.6 / 60.1||45.9 / 30.6|
|AvP DX11 (DX11, max) [fps]||129.5 / 79.4||55.7 / 34.6|
|Hitman Absolution (DX11, Ultra) [fps]||66.1 / 62.3||46.4 / 30.2|
|Dirt Showdown (DX11, Ultra) [fps]||123.9 / 112.2||63.1 / 43.1|
|Sniper Elite V2 (DX11, Ultra) [fps]||60.2 / 35.2||26.5 / 15.6|
|Sniper Elite V2 (DX11, High) [fps]||99.4 / 61.6||44.1 / 26.9|
|Sleeping Dogs (DX11, Ultra) [fps]||85.7 / 50.5||40.8 / 23.3|
|Just Cause 2 (DX11, max) [fps]||194.2 / 182.7||91.8 / 61.9|
|Potrošnja (Min/Max) [W]||103 / 494||99 / 344|
|Test configuration: Core i5 2500K @ 4.5GHz, ASUS P8Z77-V, 2x4GB AMD Multimedia Edition DDR3 1600MHz, Seagate Barracuda 1TB, Windows 7 64bit, ForceWare 314.22, Catalyst for HD7990 BETA|
It’s hard to say whether this model is faster than the GTX Titan or than the GTX 690, which we had a chance to test with beta drivers. Still, we can definitely say that this card is the best one available, not only AMD’s, but also when compared to the rest of the competition. There are small variations, which aren’t that important, because the conclusion is simple. For everyone who thinks that they need an uncompromising solution in terms of a video card, Radeon HD 7990 is the right choice. Although, you probably figured it out for yourself about halfway through this article.