Marvell has been very quiet of late, too quiet, and most people wrote them off long ago. This isn’t to say they haven’t been up to something, in fact, SemiAccurate’s sources say that Marvell is up to something really big. As the TV news report interviews a shocked looking neighbor; it is always the quiet ones…….
Back to the silicon side of things, what is Marvell cooking up? The company has an architectural license for ARM, so their core will not be a black box like all the A9s out there. Marvell is doing what Qualcomm did with Snapdragon and the upcoming Krait CPUs, basically making a ground up new chip that implements the ISA in question, but does some things better.
Better means whatever the chipmaker wants it to, lower power, higher performance, more bandwidth, or whatever else they choose. Instead of having to implement a middle of the road core as designed by ARM(1), a company is free to put the sliders for cost, performance, power and anything else where they want, not where things suit the majority. As you can see with the Qualcomm Snapdragon line, it is a very effective way to go.
SemiAccurate’s moles say that Marvell is cooking up an A9/A15 hybrid part, philosophically closer to the A9 on the core, but with a very A15-like memory controller. This likely means a 3-pipe design rather than a 4-pipe, but with >32-bit memory access and virtualization instructions. The new chip should implement the full A15 ISA, so there should be no software compatibility issues, but has less raw grunt than a full A15. At least on paper.
There are two ways that this could end up. The first is that Marvell did this to save die space, power, or both, while getting the benefits of the A15 ISA. This way would mean a killer high volume smartphone chip, basically a “90% of the performance for 50% of the cost” type part. We think this isn’t what they are up to though.
The more likely scenario is that Marvell is going for very high performance, and is doing it in some other way. Since they are clean sheet implementing the A15 ISA, they could do anything they want with very few restrictions. The rumor mill is full of next-gen quad and more core Armada chips aimed at servers and similar cloud markets.
In any case, multiple sources in the know say that the key players in the ARM game are all quite nervous about Marvell’s offering, more so than even Krait. Specifics are few, but interest is very high among CPU watchers and system builders. If this chip pans out like many are saying, it could be one of the, if not the fastest ‘next gen’ ARM core. Marvell is not going to be the ‘quiet type’ for much longer, stay tuned.S|A
(1) This is not to say that the ARM cores are bad, on the contrary, they are very good, they are just aimed at a the middle of the intended market. Changing that is a lot of work, usually more than the potential benefits, so very few companies put their development money there.
Latest posts by Charlie Demerjian (see all)
- More on Intel’s 10nm process problems - Sep 17, 2018
- Intel puts out another 14nm 2020 server platform - Sep 11, 2018
- Why Can’t Intel Supply Enough 14nm Xeons? - Sep 10, 2018
- Intel can’t supply 14nm Xeons, HPE directly recommends AMD Epyc - Sep 7, 2018
- AMD reintroduces the Athlon name with two CPUs - Sep 6, 2018