256GB/512GB/1TB flash speed differences?

[userjohn]

Does anyone know whether the 3 choices of PCIe flash size for the nMP would differ in speed? I've been unable to find any discussion of that issue.

I'm asking the above question because I've read that in the latest Retina MacBook Pro, the 1TB PCIe flash is substantially faster (at both reading and writing) than the 512GB flash. It seems the reason may be that the 1TB flash is 4-channel (whereas the smaller capacity is not). Apple apparently only advertises the slower flash speeds for the rMBP and based its testing on the 512GB model. In contrast, for the nMP, Apple's testing refers to the 1TB model.

On a related point, I am wondering how performance would compare for PCIe flash of the same capacity on the rMBP and the nMP.

 

[tuxon86]

Does anyone know whether the 3 choices of PCIe flash size for the nMP would differ in speed? I've been unable to find any discussion of that issue.

I'm asking the above question because I've read that in the latest Retina MacBook Pro, the 1TB PCIe flash is substantially faster (at both reading and writing) than the 512GB flash. It seems the reason may be that the 1TB flash is 4-channel (whereas the smaller capacity is not). Apple apparently only advertises the slower flash speeds for the rMBP and based its testing on the 512GB model. In contrast, for the nMP, Apple's testing refers to the 1TB model.

On a related point, I am wondering how performance would compare for PCIe flash of the same capacity on the rMBP and the nMP.

The real question is: will you be able to tell the difference?

There are so many factor in determining if an application or a file transfer seem to be slower... Is it really the SSD speed that is impacting performance or is it a matter of an OS services stealling ressources or another app doing the same...

All current SSD leaves HDD in the dust...

 

[VirtualRain]

Does anyone know whether the 3 choices of PCIe flash size for the nMP would differ in speed? I've been unable to find any discussion of that issue.

I'm asking the above question because I've read that in the latest Retina MacBook Pro, the 1TB PCIe flash is substantially faster (at both reading and writing) than the 512GB flash. It seems the reason may be that the 1TB flash is 4-channel (whereas the smaller capacity is not). Apple apparently only advertises the slower flash speeds for the rMBP and based its testing on the 512GB model. In contrast, for the nMP, Apple's testing refers to the 1TB model.

On a related point, I am wondering how performance would compare for PCIe flash of the same capacity on the rMBP and the nMP.

This wouldn't surprise me but it's not necessarily the case.

I saw a BlackMagic result of a 512GB drive from the nMP in a video yesterday and it's performance was around 900MB/s... which is fast, but not as fast as the advertised 1.2GB/s on Apple's promo material. But until we see further benchmarks, it's too soon to jump to the conclusion that the lower capacity SSDs will run slower. However, if this is what was done on the MacBooks, it's safe to assume the same applies here.

As an FYI, When building a set of different capacity SSDs, vendors can basically choose one of two approaches...
1. Max out the channels on the controller at each capacity by using different size NAND modules (e.g. 8x64=512, 8x128=1024), OR
2. Use the same size NAND modules but use more of them across more channels in the higher capacities, and less modules across less channels in the lower capacity models (e.g. 4x128=512, 8x128=1024).

Since SSDs basically treat channels and NAND chips like drives in a RAID0 array, ideally, they use all the channels. However, it's often easier and cheaper to just populate less channels for the lower capacity models. It's not a best practise though, so hopefully Apple/Samsung are not doing this.

Having said all this, I'm not sure I'd worry much about the difference between 900MB/s and 1200MB/s... both are insanely fast!

 

[Bear]

Currently in general for SSDs there is a big performance jump from 128GB to 256GB and small to none from 256GB to 512GB.

I would look at how big a system disk you would need so you don't have to micromanage disk space on it. For most people that will be 256GB and for some it will be 512GB. And some the 1TB. This of course presumes external data disks. For me it would depend if any of the external disks would be SSDs or all HDs. I would need the 512GB SSD to be able to just use HDs externally.

 

[handheldgames]

Currently in general for SSDs there is a big performance jump from 128GB to 256GB and small to none from 256GB to 512GB.

I would look at how big a system disk you would need so you don't have to micromanage disk space on it. For most people that will be 256GB and for some it will be 512GB. And some the 1TB. This of course presumes external data disks. For me it would depend if any of the external disks would be SSDs or all HDs. I would need the 512GB SSD to be able to just use HDs externally.

A 128GB version of the PCIE SSD used in the nMP is definitely crippled in write speeds compared to the 512GB version as I'm running a XP941 variant in a cMP as a boot disk. AFAIK, there hasn't been a test of the 256GB part to date.

 

[deconstruct60]

Does anyone know whether the 3 choices of PCIe flash size for the nMP would differ in speed? I've been unable to find any discussion of that issue.

Most higher end SSD vary in performance due to capacity. In part this is because flash is slow at writing. Nothing major has changed there. The primary changes have been in getting better at hiding the slowness.

Generally with more Flash chips/dies to target the better the "slow work-arounds" tend to work. More dies means more capacity. So there is a correlation.

Is it worth buying capacity you don't need just to get somewhat faster writes ... probably not if you are killing off enabling some other aspect of your system ( smaller , fewer cores , back up drives , etc. )

It seems the reason may be that the 1TB flash is 4-channel (whereas the smaller capacity is not).

It is something actually sitting on the channel that is the issue. The other issue is that for some other Mac configurations Apple is flipping from controller X on some capacities and controller Y on others.

In contrast, for the nMP, Apple's testing refers to the 1TB model.

Most of the bragging "up to ..." limits is all about maxing out the BTO config.

On a related point, I am wondering how performance would compare for PCIe flash of the same capacity on the rMBP and the nMP.

I'm not sure they are going to compete. The nMP's SSDs are quite likely x4 PCI-e v2 designs. The laptop's x2 ones. At the lower capacities there is probably a much smaller gap and expanding as can invoke more RAID like accesses to a higher number of flash dies in the larger capacities. Long term the potential replacements in a Mac Pro are probably going to do better at gaping the laptop versions.

 

[hugodrax]

Damn! 900Mb blows SSD drives out of the water, they top out at 560Mb.

I thing the 1.2 is for the 1TB model.

 

[handheldgames]

Damn! 900Mb blows SSD drives out of the water, they top out at 560Mb.

I thing the 1.2 is for the 1TB model.

This is from the 128GB Samsung XP941 on the cMP. Crazy single drive numbers in many respects, performing beyond 900Mbs/sec in many sequential read operations. It just looses steam against the 256GB and larger variants by the number of cells it works with and it's onboard cache.