In this article, I would like to provide the very first estimates of the performance of Intel's (INTC) upcoming "Broxton" mobile processor based on the new "Goldmont" CPU core and Gen. 9 GPU. Indeed, the majority of my work is done by fleshing out the broad estimates laid out at its Investor Meeting.
From these claims, let's get to it.
For CPU performance, I will be using Geekbench which, while far from perfect, seems pretty representative of raw, best-case CPU performance (i.e. in code that's not particularly complex/branchy). I pull my data for the NVIDIA (NVDA) Tegra 4, the Snapdragon 800, the Atom Z3770, and the Atom Z2760 from the Geekbench results browser, taking care to pick results that tend to be in-line with other similar chips.
Without further ado, here are my CPU performance estimates (both single threaded and multi-threaded):
Essentially, if Intel delivers to the promise of "5x" the CPU performance of the Atom Z2760 (and I am assuming that they mean 5x the multithreaded performance and that Broxton is a quad core with roughly 80% coherence), the CPU performance should look very much like what is shown here.
From the chart above, we can draw the following inferences:
- Apple could mildly tweak its "Cyclone" CPU core and drop two more of them onto the same die on 20nm/16nm FinFET process and end up having a part that is more than competitive with Broxton on CPU performance
- As far as multithreaded performance goes, Broxton will offer greater performance than today's premium Ultrabooks in a smartphone/tablet form factor
- Intel needs to advance its "big core" chips much faster in order to be able to continue to command a premium over Atom class products and to differentiate the notebook form factor over smartphones/tablets
In Intel's estimates, the company claimed that in GLBench 22.214.171.124, the "Broxton" chip would offer roughly 15x the performance of the Z2760 chip (which, let's face it, was uncompetitive). So, scouring the internet for GLBench 2.5 results for Z2760 led me to the following chart from AnandTech:
Using these results here, we can come up with estimates for Broxton:
Now, what's interesting is that compared to the Snapdragon 805 (based on 1.3x Snapdragon 800), Broxton doesn't really provide a whole lot of performance uplift (although it is likely to be able to run at full tilt for longer inside of a smartphone/tablet). Also, when we're dealing with "15x", the choice of starting number is pretty critical. If the Z2760 result had been, say 8 or 9fps instead of 7fps, then the performance delta over the Snapdragon 805 (PROJ) would be much higher. But I do think that Intel is finally gunning for "leadership" in graphics, something that it has traditionally lagged in.
"Broxton" will be out in "mid-2015" according to Intel, which means that it will need to face the 20-nanometer generation of products from Qualcomm/NVIDIA/Apple and there's a small chance that it goes up against 16-nanometer FinFET products from Apple/Qualcomm (depending on how well those design teams execute and how well TSMC does tackling its FinFET process development).
From these estimates, Broxton looks like it will be a solid performer, but it's not tough to see the ARMy come up with products that offer similar performance (particularly Apple on the CPU side, and any of ImgTec's (OTCPK:IGNMF) licensees on the GPU side given what we know about Series 6XT). I also expect NVIDIA to continue to push hard here, as will Qualcomm (QCOM).
The nice thing about Broxton, though, is that it will sport a die size under 60mm^2 which should make it rather economical. If Intel can offer the best performance/watt/$, then it has a good shot of winning some serious share/business with this part. The thing to keep in mind, though, is that the estimates presented here are rough and that final performance is subject to change (for example, thanks to Intel's new claimed "SoC methodology", it would be able to stick in a more powerful GPU block if the competition moves faster than expected at some cost to die-size/margins).