Note: This post is divided into two portions. The first covers the leak and my findings while the second is speculation.
Alright folks, something very very interesting popped up. Infact, I haven't had this level of excitement for a long while. Benchmarks for the 'Strato XT' GPU have popped up and they do not disappoint. The GPU is pretty powerful judging from the benchmarks and here is the real kicker: there is hard evidence of two more GPU cores of the same lineup: Strato PRO and Litho XT. With three completely new cores unaccounted for in the current AMD lineup, we are almost certainly looking at the R9 M300 Radeon Mobility Series.
AMD's Strato XT Mobility GPU gets benchmarks leaked - First look at Radeon Mobility R9 M300 Series
Lets start with a little refresh. The nomenclature first popped up on Zauba some few weeks back. I covered it then, but they just seemed like low powered MXM GPUs at that time. While I was doing some digging around my usual locations today, I spotted something eye popping: Benchmarks for the Strato XT GPU, one of the two dies previously leaked. And guess what, the chip isn't low performance at all. Infact, it beats the prototype of Carrizo (Carrizo-L I would guess) by a margin and then some. Before we go any further, take a look at the shipping entries below, which appeared the first time the R9 M300 series (something we now know) was spotted.
One shipping entry (Litho XT) is relatively recent whileas the other (Strato PRO) is relatively old. AMD's Strato PRO is a Type B MXM GPU with 4GB of GDDR5 memory. Similarly, Litho XT is a Type A MXM GPU but has just 2GB worth of GDDR5 memory. Ever since we discovered the benchmarks of the Carrizo prototype on GFXBench, I have been keeping a close eye on it for more leaks and today, I was not disappointed. Right there in plain sight, was a GPU by the ID "AMD Strato XT (6646)". The proceeding four digit number is a typical AMD prototype indicator and was actually used in the Carrizo leaks to trace it in the SiSoft Sandra database. Needless to say, the entire affair reeks 'legit', not to mention it is nearly impossible to fake benches on GFXBench. Anyways, here are the benchmarks of the Strato XT Mobility GPU:
Allright, this is the part where we leave the facts behind. If speculation isn't your cup of tea, this is where you stop reading. The avid reader will notice there is no mention of any indicator in the Zauba shipping entry that might hint at GPU power, so how exactly do I know which core is the more powerful variant? Well, AMD is using the nomenclature Litho and Strato which is almost certainly short for Lithosphere and Stratosphere. In Geography, the Lithosphere is part of the Mantle and is naturally below the Stratosphere, so if the rules of nomenclature hold any meaning to AMD, Strato should be the powerful variant. That would put the lineup, in ascending order of power as: Litho XT, Strato PRO, Strato XT. Three unique cores are more than enough to populate an entire lineup and this leads me to suspect that we are actually looking at the first hints of AMD's R9 M300 Mobility Series.
There is, however, one problem: Strato XT not being able to beat the Tonga Mobility GPU, namely R9 M295X. Basically, you are looking at the following approximate lineup for the Radeon Mobility R9 M300 Series:
- R9 M390X / R9 M380X equating to Tonga Mobility GPU.
- R9 M380X / R9 M370X equating to Strato XT GPU
- R9 M375X / R9 M370X/ R9 M365X equating to Strato PRO GPU
- R9 M370X / R9 M365X / R9 M360X equating to Litho XT GPU
The top spot in the R9 M300 series will either be a
Pirate islands Caribbean Islands chip or Tonga Mobility. The R9 M295X should be the perfect candidate for the R9 M380X considering how common re-branding is on the mobility side. That leaves the R9 M370X price-performance sweet spot to the Strato XT GPU. With the above lineup, AMD will be able to roll out a very fresh mobility lineup with minimum rebrands, and power cuts across the board. Ofcourse the question still remains what architecture this die is based on. I talked a bit with my colleague Khalid on this and he pointed out that these chips could be the 'refresh' and mobility graphics 'win', Mark Papermaster and Lisa Su were talking about. This actually makes it possible that these dies are really enhanced versions of existing dies, with better power efficiency, revised architecture and improved binning. Although, once again, this is a possibility that won't really surprise anyone with the branding mess that is the mobility side of things. Thats it for this write up, but I am pretty sure we will be hearing more about these GPUs soon enough.
— Usman Pirzada (@usmanpirzada) December 27, 2014