with SSDs, the bigger the drive, the faster the drive. it's because the larger the size of the storage, the more channels the controller has (in order to send data to all those chips). To make an SSD bigger, they put more NAND chips. So a 1 TB drive has twice as many chips as a 512 GB drive. So those channels can be used in parallel so the more channels, the faster the data transfer.
All Surface Pro 4 machines have PCIe 3.0 based NVMe which is a huge jump over the SATA based SSD in Surface Pro 3.
The new Surface Pro machines also are PCIe NVMe and use the Samsung PM971 which is significantly smaller than any M.2 SSD of most ultrabooks, especially the 2280 versions that are the most common. The drive is also most likely more energy efficient than the fastest SSDs available today, which should help with battery life.
But the downside to the PM971 BGA SSD is that it's not as fast as the fastest NVMe drives out there. It's still much faster than a SATA SSD, but the PM971 is limited to a PCIe 3.0 x2 link, which is going to impact the maximum read and write speeds. You have to sacrifice something in order to fit something as small as the Surface Pro.
PM971 actually outperforms the same size PM951 drive which was easily the most popular SSD found in Ultrabooks until recently. The PM971, despite its "PM" name, is MLC NAND, so write speeds are reasonable, and other than the lowered maximums created by the PCIe link being x2 instead of x4, the performance is still quite good.
I was reading about the practical everyday usefulness of SSD speeds today. People always hype the sequential read. However its the small numbers of random reads and writes than actually matter for most useage. Sequential is only really for back-ups, file copies and so on. And for those, you'll need both disks to run the same speed - ie, if you do a file copy from disk to disk, the copy will go the speed of the slowest disk.
For example, to take advantage of an NVME internal disk copy, you'd need a USB 3.1 raid dual msata usb stick. Not exactly an every day scenario.
So in practice the NVME thing won't make much difference, nor will sata3. Its the random read-write speeds (small numbers of such), that's the number to pay attention to.
Was a really interesting little bit of research because it shows really, how for the most part, these broad bandswidths and high benchmarks are useless. The numbers to watch for performance are the random ones, without many iterations. Outside of that, only specialised applications will see any benefit, and only if there is no bottleneck in between.
For this of course, usually larger disks are fractionally higher. But I am not sure it would make any meaningful difference, if those random read/writes ("4K" in crystal mark), are fairly similar.
TLDR - for actual use, it doesn't matter that much. It's mainly random read/write, of small numbers of iterations (like say 4), that directly effect most applications. Even for things like file copies, you'd need the same sequential read/write at both ends to take advantage, which for NVME would require a usb 3.1 dual msata based raid 0 device (which you can get, but no one has them) or an external usb 3.1 ssd drive that can actually use the full 10 gb/s