Update: As our reader Fister has pointed out, its also possible that this benchmark is not actually that of a combined MCM that Kyle and Fudzilla have claimed to exist but rather a new Vega variant, benchmarked with a Kaby Lake processor, that has 23 CUs. If this is true than that would mean you are looking at a cut-down Vega die with 1472 SPs. I have featured his comment for anyone wishing to read this hypothesis. This article retains its rumor tag (until such point where Kyle's and Fud's claims are debunked) and the original text can be seen below:
Kaby Lake processor with MCM AMD Radeon graphics spotted on SiSoft Sandra and benchmarked
Before we begin, keep in mind that while we have not seen any successful attempts of SiSoft database manipulations, it might still be possible, so what I said in my original article still stands, treat everything with a pinch of salt till confirmation from Intel or AMD. The entry listed at SiSoft Sandra lists a Kaby Lake processor that has a very interesting integrated graphics core.
Basically, it is a 'Gen9' Graphics core that has 1720 SP. Now as you know, Intel iGPUs does not have SPs (they have EUs but that's a topic for another article). The 1720 SP core count would make this a decently powerful GPU at that although the clock speed would shave off some of that power. The clock speed is slated to be 1 GHz. To give you some perspective, the RX 580 with 2340 SP clocked at 1340 MHz nets about 6.1 TFLOps while this will net around 3.4 TFLOPs (Core Count * 2 * Clock Speed). For even more context, the PS4 Pro nets around 4.2 TFLOPs. Not bad for an integrated processor.
AMD's Kaby Lake MCM graphics core vs Intel Iris Pro 580
If a processor is benchmarked at SiSoft then it is a fair bet that it will also have an entry in GFXBench and Compubench and sure enough, we were able to find the same processor in their database. The GFXBench score of the Kaby Lake/AMD processor is given below and I also took the liberty of doing a comparison between the AMD iGPU and Intel's highest end Iris Pro 580 iGPU:
The AMD graphics core has pretty decent performance and easily manages to outstrip the Iris Pro 580 by a factor of 2-3x depending on which benchmark you are going with. Keep in mind that this is OpenGL and early drivers so this performance will only increase once the finalized software kicks in.
Compared to the Iris Pro 580's 40.9 FPS in the Manhattan benchmark, the AMD graphics core scores 52.9 FPS. Considering that the fact that the onscreen FPS is actually lower than Intel's 580 makes us believe that the driver for this particular portion of the benchmark was not optimized at the time of testing. We also have the T-Rex benchmark after that in which the AMD counterpart scores roughly 2.55x the Iris Pro 580 with around 195.3 FPS compared to the Iris Pro's 76.5 FPS.
Intel MCM designed for Apple with Kaby Lake processors and AMD graphics, Codenamed: Palo Alto?
So what exactly does this benchmark mean? Well, as to that we can't really say. Kyle over at HardOCP and Fudzilla both have the same guess however: that this is a custom chip that is designed specifically for Apple. This means that we are not looking at an across-the-board change from Intel, rather a custom product for a client with the deepest pockets in the world: Apple.
This is something that could easily be true and would fit perfectly with Apple's philosophy of customized silicon so far. These processors would help give its underpowered configurations (that only have an iGPU) some much needed boost without eating into the TDP envelope like a dGPU would. If this rumor is true then we should hear an announcement in due course from either Apple or Intel.
Also to those who are wondering, it would appear that no cross-licensing agreement between NVIDIA and Intel is necessary since they have apparently granted Intel permission to use its IP (filed up to March 2017) in perpetuity. This means that Intel does not need to get into another cross licensing deal with anyone and can simply continue producing its current products as well as innovating new ones as long as they only utilize existing IP (as of March 2017). So while Fudzilla and Kyle are confident that this is an Apple only product, there is nothing stopping it from being a mainstream product as well.
