AMD made great strides in January 2012 when it released its first GCN-based GPU. Codenamed 'Tahiti XT', the Radeon HD 7970 squeezed 4.3 billion transistors onto a 352 mm2 die, allowing for 2048 cores and a fill rate of 32 gigapixels per second.
After a long 22 month wait, AMD re-released the HD 7970 as the R9 260X (aka 'Tahiti XT2'), though the newer card actually offered less performance as it came with a slightly slower clock speed that ultimately reduced the fill rate from 32 to 27.2GP/s.
AMD showed its final Radeon 200 series graphics card a year later, but neither the price nor the performance of the R9 285 'Tonga Pro' was noteworthy. Instead, it was this GPU's adoption of the latest GCN 1.2 architecture that raised eyebrows, though truth be told, few eyebrows were actually raised. The only products that made the Radeon 200 series worth remembering were the Radeon R9 290 and R9 290X. They took the GCN 1.1 architecture and expanded the die to include more cores and a wider memory bus.
With GCN 1.2 providing no real performance edge over the first two versions, we've been wondering what AMD's next move would be. After all, it's now been three and a half years since GCN was first introduced.
Surprisingly – or perhaps unsurprisingly for the cynics among us – that next move is yet another round of rebadged Radeons.
Since these first Radeon 300 series GPUs are rebranded, AMD is ripping the band aid off quickly by releasing them all together versus trickling them out over the next few month. Today's launch brings the Radeon R9 390X, R9 390, R9 380, R7 370 and R7 360.
The R9 390X is the R9 290X with twice the VRAM (this seems unnecessary) that is clocked a bit higher alongside a slightly higher core frequency. On paper this means a 5% increase in GP/s and a 20% greater memory bandwidth.
The R9 390 is much the same, being a clone of the R9 290 with twice the VRAM as well as higher core and memory clock speeds.
The R9 380 is the R9 285 with a few minor upgrades, but performance-wise the R9 380 shouldn't be much faster than the 3.5 year old HD 7950 as they both have the same core configuration. The R9 380 has a core and memory clock speed advantage but it also suffers one major disadvantage, which we'll get to shortly.
We had lingering hopes for truly updated GPUs, but what's old is new again at AMD so say hello to a familiar family of Radeons.
HIS Radeon R9 390X
Comparing the R9 390X to the old R9 290X on paper, the most noticeable difference is the extra VRAM. Whereas the R9 290X was armed with 4GB of memory by default, the R9 390X gets 8GB. For what it's worth, there are 8GB versions of the R9 290X around and they only cost around $40 more than the standard 4GB models.
For a GPU with the same horsepower as the R9 290X, this much VRAM certainly seems to be overkill. The new GTX 980 Ti comes with 6GB of memory and is surely going to be faster.
Memory capacity aside, the R9 390X's biggest shot at exceeding the R9 290X's performance rests in memory clock speed as this has been boosted by 20% at 6Gbps.
Along with the memory upgrades, the only other change is a minor adjustment to the core clock speed. The default spec now reads 1050MHz, a 5% increase over the previous 1000MHz speed cap. That said, few board partners ran their custom cards at the reference clock speeds anyway.
For example, the HIS IceQ R9 290X Hybrid ran at 1100MHz with the same 6Gbps GDDR5 clock speed as the new R9 390X. That particular version from HIS was 12% faster than a standard R9 290X and just 4% slower than the GTX 980.
The new HIS R9 390X IceQ X² OC 8GB looks as though it's a special edition card with that huge gold fan shroud, but this is just an overclocked version and isn't part of a limited edition series.
Unlike cards based on the AMD reference design, this new HIS graphics card features a completely unique cooling solution that is considerably more efficient.
As the name suggests, the IceQ X² cooling solution has been employed, though it's a slightly modified version from what we have seen previously. Built with two super large dual 89mm dual axial fans, this dual-slot cooler features a massive heatsink along with three 6mm and two 8mm heatpipes linked directly to an extra-large 40 x 47.5mm copper base, where they help to improve heat extraction from the GPU.
Connected to the large copper base of the heatsink is a unique RAM heat spreader, which is designed to cool the GDDR5 modules. HIS claims that this cooler is quieter and more efficient than AMD's reference design.
HIS has modified the original IceQ X² design to include a rear back plate, though we should point out that none of the GDDR5 memory chips are located on the backside of the PCB. Also attached to the heat spreader on the front is an aluminum plate that extends up and over the heatpipes. Apart from looking nice, we're not sure what this does. HIS describes it as an easy means of carrying the graphics card, which seems strange.
To feed the card enough juice, HIS has gone with a 6-pin/8-pin PCI Express power connector configuration.
The R9 390X IceQ X² OC 8GB is designed to operate the core at up to 1070MHz with 8GB of GDDR5 memory clocked at 6Gbps. Although the AMD spec calls for a GPU clock speed of up to 1050MHz, many custom built versions will operate slightly higher than this.