⋮    ⋮    ⋮  

AMD Fiji XT R9 390X Leaked For The Third Time On Zauba – 20nm Looks Promising

Khalid Moammer
Posted Jan 5, 2015
111Shares
Share Tweet Submit

AMD’s elusive high-end GPU has been spotted for the third time in the Zauba shipping database. The flagship will allegedly power the R9 390X.

This time however unlike the previous two leaks not one, not two and not even three but four PCB assemblies for the GPU have been shipped. And for the first time the listing included a very peculiar suffix “(FOC)” which we believe stands for Full Operational Capacity. This signals that the card has exited the early prototyping stage as finalized samples are being shipped for end-stage testing and evaluation. The R9 390X moniker will only serve as a place holder for this article until we know for sure what AMD has decided to call it. However there’s strong indication that this is going to be the new flagship GPU from AMD to lead the charge on the 2015 graphics refresh.

AMD Fiji XT R9 390X Leaked For The Third Time On Zauba

If you remember back in November we had reported that there are four different Fiji XT prototype cooling units. With the leaked Hydra hybrid cooling unit being only one of them. The four Fiji XT units recently shipped could be one of each. Likely for the purposes of testing and making the final decision on which one is going to be used for the mass produced graphics cards. It may also just end up being pure coincidence.

This is indeed the third time that Fiji XT has made its way to the shipping database. And the second time it did so in disguise. The first time the GPU was leaked it had been frankly listed as Fiji XT but as we had discussed in our last article covering the second appearance of the GPU. Someone at AMD had actually realized this and proceeded to omit the Fiji XT nomenclature/code name from future listings. This occasion is no exception and as with the previous listing the code name has been omitted. However just like last time enough of the description (effectively all of it except “Fiji XT”) has been left out unchanged. So we can confidently conclude that this is indeed Fiji XT.

If the previous SiSoftware Sandra leak is anything to go by then we’re looking at a 4096 GCN core GPU with 4GB of stacked High Bandwidth Memory and a 4096bit wide I/O memory interface. All built on a cutting edge manufacturing process. I’ve discussed in great detail the numerous indications that AMD had given as to which process Fiji XT may be manufactured on. As things currently stand 20nm is shaping up to be the process of choice for a number of upcoming products from AMD.

NVIDIA GeForce GTX 1060 6 GB 3DMark Performance Results Leaked - Boosts Up To 1709 MHz, 8 GHz Memory Clocks
WCCFTech Fiji XT (R9 390X) GM200 (Titan II) GM204 (GTX 980) Hawaii (R9 290X)
CUDA/GCN Cores 4096 3072 2048 2816
Memory Capacity 4GB HBM 6GB GDDR5 4GB GDDR5 4GB GDDR5
Memory Clock Speed 1Ghz 7Ghz Effective 7Ghz Effective 5Ghz Effective
Memory Bandwidth 640GB/s 336GB/s 224GB/s 320GB/s
Boost Clock Speed 1.25Ghz ~1.2Ghz ~1.2Ghz 1Ghz
Manufacturing Process TSMC 28nm TSMC 28nm TSMC 28nm

TSMC’s 20 SoC has been described as being inappropriate for discrete graphics cards by some due to its low power nature. However TSMC’s 20nm SoC process has proven to be easily capable of hitting excellently high clock speeds. The process manages to sustain frequencies just as high as 28nm if not slightly higher, making it an ideal candidate for discrete GPUs.

We know this thanks to Nvidia’s just announced 20nm Tegra X1 SOC which features the Maxwell GPU architecture. Based on Nvidia’s official figures the integrated Maxwell GPU operates at a frequency of 1Ghz. That’s 50mhz above the integrated GPU of the Tegra K1 and merely ~200 mhz below the desktop GM204 based GTX 980 and GTX 970 graphics cards. 200mhz is the same frequency delta that stands between the 28nm Tegra K1 and the 28nm discrete Kepler GPUs. The Tegra SOCs (and notebook GPUs) are often clocked more conservatively than their desktop siblings for power saving purposes rather than due to a fault of the manufacturing process. So we’re looking at a 20nm process from TSMC that not only performs as well as the high performance versions of 28nm but also consumes significantly less power.

Share Tweet Submit