Review: AMD Radeon R9 295X2

No-compromise card

The console vs. PC debate is often an amusing source for repartee and discourse amongst technology-loving friends. Thought about it objectively, the graphics quality of consoles is closest to PCs when there's a new launch, as there was at the tail end of last year with the Xbox One and PlayStation 4, but then the PC forges way, way ahead over the next few years.

A seismic shift in PC graphics ability usually occurs when either AMD or Nvidia 'double-up' on their latest GPU architectures by launching a card equipped with dual GPUs. Recent history is littered with such examples, with the GeForce GTX 590 and GeForce GTX 690 from Nvidia and Radeon HD 6990 and Radeon HD 7990 from AMD.

This time around, Nvidia has been first to ply two high-end present-generation GPUs on to one board. The monstrous GeForce GTX Titan Z was announced a few weeks ago and priced at an amazing $2,999. Only partial specifications are known, however, and Nvidia is marketing it more as a rackmounted workstation card than a gaming goliath. Perhaps the green team is holding back until AMD plays its hand?

It just so happens that AMD has been working on a new dual-GPU card of its own in the skunkworks division. This new card, Radeon R9 295X2, codenamed Vesuvius, takes two full-fat Radeon R9 290X GPUs, places them on one board, and then increases the peak core clock a touch further. Want to see how?

Specification

Analysis

We've lined up the biggest, baddest Radeon ever against the two best dual-GPU cards on the market - Titan Z is excepted because specifications are not fully disclosed - and also against three single-GPU cards that are the best each company has to offer. This, folks, is a lustworthy line-up.

The Radeon R9 295X2's numbers show that, for the first time, AMD's flagship dual-GPU card is more than two of its finest single-GPU cards lashed together. The per-GPU architecture is identical to the R9 290X's, but the higher peak core speed - 1,015MHz vs. 1,000MHz - means it should benchmark slightly faster than two 290Xs in CrossFire. There's a little more to it than that, which we'll get on to in a moment.

AMD is almost belligerent that this card is wasted when run on anything less than a 4K-resolution screen, such is its innate power. This is opportune insofar as the prices of 4K screens are falling quickly.

So knowing that the R9 290X can become hot and bothered and consume significant watts, how has AMD shoehorned two premium GPUs that make no compromise on frequency? The usual trick is to be conservative on speeds and hand-pick GPUs that run with low voltage.

This, however, is a no-holds-barred card whose performance potential dwarfs any other concern. This is why it's specified with a 500W TDP. Yes, that's right, this card guzzles a potential 500W when running at full chat, a figure that is 33 per cent higher than what we thought was the acceptable limit for a single card - 375W belonging to the HD 7990.

Playing fast and loose

Simple maths informs us that AMD is playing fast and loose with the PCIe specification. Two 8-pin connectors can pull 300W from the PSU, with another 75W coming from the slot. The up-to 125W shortfall is bridged by, when the occasion arises, the card pulling more from the 8-pin connectors than it technically should. AMD says your PSU should be able to supply a combined 50A over the two connectors, without this being shared between other components. That's a huge outlay of amperage; you'll need to ensure the PSU can deliver it and the cables can withstand such a load. It's no surprise that the Radeon team is touting 1,000W PSUs as advisable for a single card.

Throwing the TDP rule book out of the window has other ramifications, too. AMD knows that such energy consumption cannot be readily tamed by traditional air cooling so implements all-in-one, closed-loop cooling on the Radeon R9 295X2. Yup, it's worth repeating that the card ships with a pre-plumbed liquid cooler.

And understanding how AMD's PowerTune with Boost technology works is important when appreciating liquid cooling. The card opportunistically boosts to 1,015MHz on both GPUs when there's scope to do so, and this scope is widened if the GPUs are running at lower temperatures, which should be possible with quality cooling. Point is, lower temps should enable consistently higher clocks.

But it's not all great news. It's actually a tad disappointing that AMD hasn't pushed the memory clock further. As it is, it's exactly the same configuration - 512-bit pathway, 4GB framebuffer, 5,000MHz frequency - as on the standard R9 290X, albeit presented twice.

Geared for ultimate performance, let's now take a look at the beastie.