A little while back, AMD announced the Radeon R9 285, a mid-range card based on the Tonga GPU sporting either 2GB or 4GB of VRAM, with enough performance to best the outgoing R9 280 and Geforce GTX 670, while at the same time offering more features, and better power management. Intended as an update to the Tahiti family of GPUs, it was faster and cheaper than the Radeon HD 7970. But AMD only released the Tonga Pro variant of the chip to consumers, choosing to send Tonga XT, a more powerful version, to Apple in an exclusive deal for the new 27-inch Retina 5K iMac desktop. Today, over 13 months later, we finally get Tonga XT in the form of the Radeon R9 380X, and it’s expected to be available in retail stores in the coming weeks for a price approaching R3299. Let’s check it out.
Nothing changes architecturally for this release, so that’s something to take note of before you start thinking of smashing open those piggy banks – Tonga/Antigua XT is really just the full-fat version of the chip without some of the units disabled. It is close in design to the Fiji architecture, though it isn’t nearly as dense as Fiji in terms of shader count. It doesn’t have any new features that aren’t currently present in most of their current lineup, with the exception of the rebadged Radeon R7-300 series, which is based on Pitcairn (now three years old). So that keeps this release announcement rather short and sweet.
During AMD’s press call, they didn’t spend too much time on the R9 380X, but it does seem like it’s going to be beefy. With 32 compute units inside, that works out to 2048 shader cores, the same number as the R9 280X, though these are tweaked for higher performance and better efficiency. It has a base clock of 970MHz, though AMD didn’t specify what the boost clock would be. Boost clocks for both AMD and NVIDIA’s products are temperature-based at this point, so better cooling will inevitably result in higher boost clocks so long as the GPU is capable of running that fast.
As far as feature support goes, it’ll run Mantle, DirectX 12, or Vulkan-based games, and it works with AMD FreeSync monitors. AMD said nothing about the video ports, though, which is either a sign that they’re sticking to HDMI 1.4a, or they’ve snuck on HDMI 2.0 and aren’t going to say anything until the review embargo lifts. During the press call, AMD did slip in something about ongoing developments to get variable refresh rates working on HDMI, so if it comes first to this card that would be one reason to upgrade.
The Tonga, Hawaii, and Fiji architectures also benefit from technologies that AMD designed like colour compression, which eases up on the restrictions to memory bandwidth and makes running games a more efficient task for the GPU. The improvement is always theoretical because every game treats the tech differently, but under ideal scenarios there’s roughly a 15% reduction in overhead for the memory subsystem, freeing up more bandwidth for other tasks.
|Comparions to contemporary GPUs|
|R9 390||R9 380X||GTX 960||R9 380||R7 370|
|VRAM default size||8GB||4GB||2GB||2GB||2GB|
|Memory bus size||512-bit||256-bit||128-bit||256-bit||256-bit|
|FP performance||5.1 TFLOPS||3.9 TFLOPS||2.3 TFLOPS||3.4 TFLOPS||1.9 TFLOPS|
|Estimated price (ZAR)||R5999||N/A||R3999||R3999||R2999|
Looking at what’s available today, if the R9 380X is priced around R4999, it’ll be unmatched by any of NVIDIA’s solutions in that price gap, and there’s a chance that when overclocked, it’ll match the stock performance of the R9 390. Compared to the GTX 960, it does look like quite a guzzler though, with a higher thermal design point (TDP) and twice the number of dedicated hardware units on the GPU itself, except for the ROP counts. At 32, the R9 380X is probably meant to be matched up against cards at 1080p and 1440p, although gaming at UltraHD 4K is doable so long as you drop the settings to medium and lose the MSAA to bring up the frame rate.
Who is AMD targeting with this launch? They’re looking at owners of mid-range cards from 2012, like the Radeon HD 7850 or the Geforce GTX 660, and the R9 380X promises close to a 2x increase in frame rates for modern games. It’s interesting that AMD doesn’t show the Radeon HD 7970 GHz edition here, one of their most popular sellers in 2012. That card would probably be very close to the R9 380X in terms of performance, so if you own one of those cards, you’re not the target market for the R9 380X. AMD would prefer that you jump up to the R9 390 or R9 Fury to see a decent performance boost, instead.
AMD’s pre-supplied benchmarks suggest that the R9 380X will be overall faster than the GTX 960 than the GTX 960, although Star Wars Battlefront isn’t a great indicator of overall performance anyway. The real battle will be addressing performance in games that prominently feature NVIDIA’s Gameworks features, like Metal Gear Solid V: The Phantom Pain. AMD’s upcoming release of Radeon Software Crimson looks set to address that particular issue, with a more aggressive rollout of patches and performance improvements planned for the future. I find this slide downplays the potential performance of the R9 380X as well, because those fps numbers are almost certainly only calling out the scores for the games that had the lowest frame rates.
Looking at the footnotes AMD supplies at the end of their slides, footnotes #8 and #9 list the performance results of the games tested. If you had to cut out the scores for Civilization: Beyond Earth and Dragon Age: Inquisition, the new minimum score jumps up to 94fps. I think the company is trying not to hype up the launch of the R9 380X too much, but it would help if they listed individual fps scores on the previous slide so that readers would have more context with how the card will generally perform. A 71fps average for Beyond Earth is lower than my test runs for the Sapphire R9 380, so I’m stuck with guessing as to how they got those scores to begin with (I’m going with clock speed limitations, because the Core i7-5960X has a base clock of 3.0GHz).
Finally, on launch day AMD expects to have R9 380X cards available from nine of their biggest partners. None of the cards will sport a stock cooler design, although they’ll all be reference designs underneath initially, simplifying things for anyone planning to put this card under a water cooler. There’s no mention of a mini version of this card for ITX builds, and it’s going to require a 6-pin and an 8-pin PEG power connector, so I’d look only at power supples starting at 500 watts and higher to run a system with this card inside. There’s no official word of when these cards will be available locally, but they should start showing up online in the first two weeks of December 2015.
And that’s it, really. Tonga/Antigua XT has taken ages to reach the general consumer market and I’m eager to see how it performs against the competition. I also asked AMD a couple of questions about the R9 380X, and they replied today, just a few hours before the NDA lifted.
Q: Is the GPU that powers the Radeon R9 380X the same one that was found in the Apple iMac desktops recently?
AMD: The GPU that powers the Radeon R9-380X is the Antigua XT.
Q: When Raja was doing a presentation earlier this year in India, he showed a slide that showed a fully enabled Tonga chip with a 384-bit memory bus. Why did you decide to stick to 256-bit instead?
AMD: While the Radeon R9 380 and 380X products are packed with state-of-the-art features, they’re also attacking very specific sweet spots in the market. Starting at $199 – $249 range, they offer killer performance at very competitive price point, which are unrivaled in their categories. To achieve the above mentioned value proposition as well as deliver 4GB of memory which this segment demands, the Radeon R9 380 Series takes advantage of its fame buffer color compression and delivers amazing performance while configured with a 256-bit memory interface.