Just last month we put together our annual guide to the best graphics cards on offer at every price point. The key battles took place at $100, $150, $200 and $300, with top graphics cards such as the Radeon R9 390X and GeForce GTX 980 Ti taking us to $400 and beyond.
In the $100 to $200 range it was all AMD as we recommended the R7 360, R7 370 and R9 380. However, the jump from the $200 Radeon R9 380 to the $300 R9 390 left a void that normally we'd expect both AMD and Nvidia to fill with something for around $250.
We first caught wind of rumors about an upcoming Radeon R9 380X a few months ago, expecting it to be a new card and not a rebrand. As you might recall, the only cards in the R9 300 family to use AMD's current-generation Tonga GPU are the Radeon R9 285 and R9 380, meaning they are built on a 28nm manufacturing process and use AMD's GCN 1.2 architecture.
Today we put all R9 380X rumors to rest. AMD's latest graphics card starts at $230, sporting a Tonga GPU featuring 2048 SPUs, 128 TMUs and 32 ROPs. This is the same core configuration we've seen before with GCN 1.0 GPUs going as far back as 2012 with the Radeon HD 7970 and more recently with the R9 280X, however the memory subsystem is considerably different in the GCN 1.2 enabled R9 380X.
Whereas older GPUs were fed data by the way of a 384-bit wide memory bus, the 380X is limited to a 256-bit bus, as was the case with the Radeon R9 285 and R9 380 before it. However, unlike the 380, the 380X will come stock with at least 4GB GDDR5 and AMD tells us they are targeting gaming at 1080p and 1440p resolutions.
The default AMD specification calls for a core clock speed of 970MHz and a memory speed of 1425MHz, which provides a data rate of 5.7 Gbps. It should be noted that the 256-bit wide memory bus limits the 380X to a memory bandwidth of just 182GB/s, the same memory bandwidth of the 380, affording it a little over 50% less memory bandwidth than the R9 390.
Interestingly, the Radeon R9 380X has been rated with the same 190 watt TDP as the slower R9 380 in spite of its more complex core. Because they are based on the same architecture we can safely assume the 380X will consume more power and therefore generate more heat nonetheless.
Meet the New Radeon R9 380X
Although the R9 380X is based on the latest Graphics Core Next architecture, at its roots you will find a graphics card that is almost four years old now, the venerable Radeon HD 7970. Debuting back in January 2012, the 7970 ran for a cool $550 and was at the time AMD's flagship part.
Like the HD 7970 and R9 280X, the R9 380X features 1792 SPUs, 112 TAUs and 32 ROPs, while the TDP rating has been dropped considerably from 250w max to 190w, though the card still requires a pair of 6-pin PCIe power connectors. The clock speed has been set at up to 970MHz, 5% higher than the original HD 7970 and almost 15% higher than the R9 280X.
That all looks great for the R9 380X until you look at its memory subsystem, which sees the 384-bit memory bus severely downgraded to 256-bit.
The first Tonga graphics card we tested (Radeon R9 285) provided rather lacklustre performance. Then the R9 380 managed to considerably improve, and while not a pixel crushing monster, at $200 it provides an enjoyable 1080p gaming experience and is certainly preferable to Nvidia's GTX 960.
Still, despite the success of the R9 380, we have to wonder if a 2048 SPU Tonga R9 380X can bridge the gap between the R9 380 and R9 390? From where we're standing it seems far too limited by that 182GB/s memory bandwidth.
On hand for testing we have the Sapphire Nitro R9 380X 4GB OC graphics card which features a custom PCB and cooler design, along with some factory overclocking. Sapphire has boosted the core clock speed by 7% to 1040MHz, while the memory has been overclocked by just 5% to 1500MHz (6.0Gbps).
The Nitro R9 380X 4GB OC is a great looking graphics card featuring an extended height to accommodate two large 100mm fans and a nice big back plate.
Test System Specs
- Intel Core i5-4690K (3.5 - 3.9GHz)
- x2 4GB Crucial Ballistix Elite DDR3-2400
- Asrock Z97 Extreme6 (Intel Z97)
- Silverstone Strider Series (700w)
- Crucial MX200 1TB (SATA 6Gb/s)
- Radeon R9 390 (8192MB)
- Radeon R9 380X (4096MB)
- Radeon R9 380 (2048MB)
- Radeon R9 285 (2048MB)
- Radeon R9 280X (3072MB)
- Radeon R9 270 (2048MB)
- Radeon R7 370 (2048MB)
- Radeon R7 265 (2048MB)
- Radeon HD 7970 GHz (3072MB)
- GeForce GTX 970 (4096MB)
- GeForce GTX 960 (2048MB)
- GeForce GTX 950 (2048MB)
- GeForce GTX 760 (2048MB)
- GeForce GTX 750 Ti (2048MB)
- GeForce GTX 650 Ti Boost (2048MB)
- Microsoft Windows 10 Pro
- Nvidia GeForce 358.91
- AMD Catalyst 15.11.1 Beta
Benchmarks: Assassin's Creed, Battlefield
The Radeon R9 380X averaged just 48fps in Assassin's Creed Unity at 1080p using the ultra-quality settings with Nvidia's Percentage-Closer Soft Shadows (PCSS) turned off. This meant that the 380X was just 7% faster than the R9 380 and GTX 960, while it was 16% slower than the R9 390. Interestingly, the R9 380X was just 4% faster than the old Radeon HD 7970 GHz Edition and R9 280X.
This time the Radeon R9 380X was 2% slower than the R9 280X despite delivering a respectable 59fps. Although the R9 380X beat the R9 380 by a 9% margin, it was a whopping 29% slower than the R9 390.