And here's the pricing![]()
Here are reviews of PCIe SSDs with up to 3.3 GB/s...
http://www.storagereview.com/reviews/enterprise/application_accelerator
Note that most of these are not optimized for bandwidth, but for IOps and latency.
And here's the pricing![]()
Actually it is relevant and if you read what I wrote you will not that Apple's decision to only go with ATI reduces its market and makes it even more a niche product. This is not an attack on nMP that exploits Open CL, but rather raises the notion that if the nMP offered an NVidia option it might have been met with a less controversial response for those that were waiting for a newer Mac Pro.
The reality is that there are lots of software options that exploit GL and not CL. Apple has limited this new system to software that exploits CL. I speak of the nMP in a Prosumer and Professional system machine capacity.
I'll say again - those that get good mileage out of their nMP are lucky and I am glad for them. For the rest, they have to look elsewhere which is a shame.
I think the fact that Nvidia is not in the nMP probably has more to do with Nvidia than with Apple. I don't think Apple has anything against using Nvidia GPUs... they have used them extensively throughout their full line of products in the past (they were the default config in the 2009 Mac Pro) and used on many laptops, iMacs, the Mini, etc. Was it their lack of support for OpenCL or rather their conviction to CUDA? Their unwillingness to build custom cards? Or their unwillingness to offer better pricing than AMD? Or something else? It takes two to tango on this.![]()
It's unlikely that we will see a 3rd party solution until SandForce SF3700 ships sometime later this year. The Samsung drive is currently the only single-controller PCIe 2.0 x4 drive in the market (excluding enterprise drives) and that won't change until the SF3700 ships (Marvell has a x2 solution that is used in e.g. Plextor M6e).
Then what is Apple doing? What do they have access to presently that other companies don't?
Apple sources their drives straight from Samsung (and from Toshiba and SanDisk for some other Macs). SandForce drives are different in the sense that any company can license their controller, firmware and software stack and build an SSD, hence e.g. OWC could build a drive for the nMP with the SF3700 controller.
Interesting. How would you expect the SF3700 to compare to the Apple stock Samsung?
I would expect it to be faster, although the big questions is how much. SandForce has made some fairly big claims about the SF3700 and its performance but so far those have all been design targets. The advantage SandForce has is that they use on-the-fly compression, which gives them a performance advantage when dealing with compressible data (i.e. pretty much everything that isn't pre-compressed like video, audio and photos). We'll likely know more about the SF3700 and its state next month after Computex.
it's hard to image any situation where you wouldn't be better off adding more SSD space externally rather than internally, which means throwing away at least 256Gb of SSD.
Thanks, great update.
Add another reason for upgrading the internal Flash, updated or different controllers that are faster.
Apple sources their drives straight from Samsung (and from Toshiba and SanDisk for some other Macs). SandForce drives are different in the sense that any company can license their controller, firmware and software stack and build an SSD, hence e.g. OWC could build a drive for the nMP with the SF3700 controller.
it's hard to image any situation where you wouldn't be better off adding more SSD space externally rather than internally, which means throwing away at least 256Gb of SSD.
And here's the pricing![]()
I think this really demonstrates the incredible value of the nMP SSD... You get 80% of the performance of this enterprise drive for one-tenth the price.
There much cheaper FusionIO cards.
I'm curious to see some performance benchmarks between the two. FusionIO is all about low latency while most people tend to quote the least important performance metric. (Sequential)
I'll do some tests this week if I have time. Have a few Iofx cards here and an Mac Can.
Given that an internal SSD would have to connect to the Apple controller built into the motherboard, that's complete nonsense.
OWC could build a drive for the nMP with the SF3700 controller
I love it when people say things like "that's complete nonsense", thanks for that.
My understanding is that the controller is on the daughter card with the chips. Read that bit above by Hellhammer, who says he is the SSD editor for Anand.
I haven't seen any analysis yet for the nMP flash, but I believe these always consist of controller and actual flash. Indeed you'd want that design as there are different flash technologies.
Oh, here's another bit, I've heard reported that the 1TB flash is faster than the 256 due to more controllers in parallel. Apple certainly wouldn't put that on the motherboard, much more modular if you put it on the daughter card.
Oh, here's another bit, I've heard reported that the 1TB flash is faster than the 256 due to more controllers in parallel. Apple certainly wouldn't put that on the motherboard, much more modular if you put it on the daughter card.
You're absolutely right... the small daughter card that is the SSD contains both the Samsung controller (left) and NAND chips (right). This image is of the 256GB version. It's possible the 1TB version uses 8 chips and thus potentially has twice the NAND in parallel offering a performance boost.
![]()
fusionio is all about low latency while most people tend to quote the least important performance metric. (sequential).
Given that an internal SSD would have to connect to the Apple controller built into the motherboard, that's complete nonsense.
You're absolutely right... the small daughter card that is the SSD contains both the Samsung controller (left) and NAND chips (right). This image is of the 256GB version. It's possible the 1TB version uses 8 chips and thus potentially has twice the NAND in parallel offering a performance boost.
LSI's targets are far from humble as it's shooting for 1.8GB/s for the PCIe 2.0 x4 version, which is only 200MB/s short of the theoretical 2GB/s limit. LSI is expecting to be able to hit these performance targets, but it's not quite there yet. Based on LSI's own tests, they've been able to achieve 1.45GB/s sequential read speed with early firmware
...
There is a ton of potential in SF3700. If LSI is able to get the performance to the expected level, the SF3700 will likely be the fastest consumer-orientated PCIe solution we've seen.
Thanks VR. I was cautious in making pronouncements, even though I was entirely sure that the controller resided on the Flash daughter card. This is a beautiful example of how not to embarrassingly shoot your mouth off on the internet.
The usual approach to get a performance boost with Flash is by paralleling up the controllers, thus making a hardware RAID system. I'm expecting then that OWC is waiting for the SF3700 to release, and will then come out with a 512 and 1TB daughter cards that are both faster and cheaper than the Apple version.
Here's an article that "Hellhammer" above did for Anand on this controller
LSI Announces Sandforce SF3700
I thought that there was an artificial constraint that VR wouldn't discuss SSDs that used RAID-0 for bandwidth....![]()