TITAN X vs. 290X CrossFire vs. GTX 980 SLI

We are using 33 games to benchmark the performance of three top single-GPU video cards which represent the fastest from AMD and from Nvidia – the GTX TITAN X, the GTX 980, and the R9 290X.  We have already tested them before at stock and overclocked, but never before using TITAN X versus 290X CrossFire and against GTX 980 SLI. The TITAN X is in a class by itself as it is the only card with 12GB of vRAM which allows it to play with some AA and maxed-out settings, even up to 4K resolutions. In our earlier testing, the GTX 980 won decisively over the 290X, but we only tested up to 2560×1600 resolution.  Now we’d like to also see how GTX 980 SLI and non-throttling R9 290X CrossFire performance compare with each other, and also with the stock and the overclocked GTX TITAN X.

Since we have two reference versions of the 290X – a very capable PowerColor overclocked PCS+ card and a recent VisionTek 290X – we are going to need to run them each with their fans running at 100% to prevent throttling when they are in CrossFire.  The GTX 980 and 980 SLI are tested at stock, while the GTX TITAN X is benched at both stock and also at moderately overclocked speeds.

The GTX TITAN X is the undisputed king of single-GPU performance.  However, performance comes at a premium as it costs $1000.  What we want to find out is if a $600-$700 pair of R9 290Xes in CrossFire (or even GTX 980 SLI for $1100), can deliver the same or a similar experience as a single TITAN X.

AMD’s R9 290X are generally priced beginning from $299 to well over $400, sometimes with a rebate, and often bundled with DiRT Rally at Newegg.com.  The overclocked versions are generally clocked from 1030MHz to 1050MHz on the core, up 30MHz to 50MHz over reference, similar to our PowerColor-clocked PCS+ 290X.  At the time of writing, not one reference version of the 290X is available at Newegg.  The GTX 980 and GTX 970 are both now bundled at participating etailers with the Witcher 3: Wild Hunt and with Batman: Arkham Knight.

The Myth of the Throttling Reference 290X & the lack of a proper CrossFire Solution

The reference versions are evidently quite unpopular, originally getting a bad reputation for running hot and throttling well below AMD’s “up to 1000MHz” on the core when it first launched.  We were quite fortunate to get a PowerColor PCS+ overclocked reference version of the 290X at launch from retail that is clocked at 1030MHz on the core, and it does not throttle in cool ambient temperatures.  And in testing our recently purchased VisionTek reference version which also pegs 1000MHz, we find it does not throttle either at stock Uber (55%) fan profile. Neither card throttles at all – either separately or in CrossFire with a space between them – when their cooling fans are set to a nearly unbearably noisy 100%.

Evidently the early Hawaii GPUs at their launch barely met specifications as the reference cooler was unable to cool them properly without throttling the clock speeds.  At Hawaii’s launch, AMD was determined to beat the original TITAN’s performance even if it meant running the 290X on the edge with the Uber clocks and uber noise.

Newer Hawaii GPUs evidently have no such issues with the reference cooler as long as the gamer is willing to put up with what most consider excessive noise from a fan that spins up to 55% when the GPU hits 94C.  This is a rather big issue that CrossFire users face when picking a 290/X video card – either use reference and put up with extreme noise from two cards, or use an open design for two hot-running cards and roast the interior of your case and the PC hardware inside.  Of course, the ideal solution is to use water to cool two hot-running Hawaii GPUs, but even a highly overclocked watercooled R9 290X will not catch a less expensive aftermarket air cooled GTX 980 in performance.

The R9 290X vs. the GTX 980 & vs. the GTX TITAN X

The VisionTek reference version of the R9 290X is clocked up to 1000MHz while the PowerColor 290X PCS+ Edition is clocked up to 1030MHz .  Memory clocks are 1250MHz at stock GDDR5 5000 speeds.  When we ran them together in CrossFire, both of the GPU speeds are normalized so that they both peg at 1000MHz.  Also, PowerDraw is set to its maximum +50%, and the fans are set to 100% using the latest version of MSI’s Afterburner.   We could tell that there was no throttling with the fans screaming at 100% since we had very uniform results across all of the runs over long benching sessions.

The reference GTX 980 and TITAN X were left at stock clocks and fan profile although we maxed out Power and set the thermals to 94C using EVGA’s PrecisionX.

What about overclocking?

Both of our evaluation samples of the reference GTX 980 overclock well on stock fan profile and stock voltage so that they performs significantly faster than our mildly overclocked PowerColor R9 290X which can only reach 1050MHz together with the VisionTek 290X in CrossFire.   Hawaii does not overclock particularly well on air, so we only overclocked our TITAN X.  We are able to add +200MHz offset to the core and +200MHz to the memory.

Since we do not want any chance of our CPU bottlenecking our graphics, we are testing all of our graphics cards by using our Intel Core i7-4790K at 4.0GHz/4.4GHz Turbo, 16 GB Kingston “Beast” HyperX DDR3 at 2133MHz, and an ASUS Z97E SLI/Crossfire motherboard. The Core i7-4790K which turbos all 4 cores to 4.4GHz is fast enough to differentiate even high-end video cards at high resolution and at high detail settings.

Before we look at our test bed and run benchmarks, let’s check out test configuration.