SATA Express meets the '09 Mac Pro - Bootable NGFF PCIE SSD

[OCDMacGeek]

About 5-10% slower than in a cMP.



Sadly my sluggish D300s are each getting 16 lanes of PCIE 3.0 while the SP951 is only getting 4 lanes of PCIE 2.0.



Oh well, at least it's shiny.



Gonna slap it back in my 4,1 where I can use a GTX980 and get some real performance.

I've got a 1tb blade in a 6,1 as well, but with D500s. Great speeds when tested with QuickBench, ridiculously bad speeds when tested with Blackmagic (200 MB/s). Did you experience anything like this? Doesn't seem like the cMP crowd ever had this issue.

If you wouldn't mind, where exactly in the system profiler is the info on number of lanes and PCIE 2.0 vs 3.0?

Edit: On closer reading, I guess you must have your blade installed in a PCI adaptor connected to the nMP, not directly in the primary PCIE slot of the nMP?

 

[OCDMacGeek]

Does it make sense that one of these blades would perform poorly when tested with small files (like the 4KB files some tests use) but then perform much better the larger the file?

The one I have isn't immediately fast, but seems to have a bit of "warm up" time before the speeds kick up.

 

[Synchro3]

Does it make sense that one of these blades would perform poorly when tested with small files (like the 4KB files some tests use) but then perform much better the larger the file?

The one I have isn't immediately fast, but seems to have a bit of "warm up" time before the speeds kick up.

Yes. Did you set the file size to 5 GB in Blackmagic?

 

[OCDMacGeek]

Yes. Did you set the file size to 5 GB in Blackmagic?

I did run Blackmagic with a 5 Gb test file. In Blackmagic, for whatever reason, the speeds remain about 200 MB/s regardless of the file size. However, in QuickBench the speeds increase to where they should be when the file size is larger.

I downloaded several different benchmarking apps for windows, and ran them in Boot Camp (HD Tune, ATTO). When testing larger block sizes, the speeds are where they should be as well. All except CrystalDiskMark, which acts like Blackmagic and finds slow speeds.

Ultimately, this may come down to some kind of difference in testing algorithms, but since others here are not having the same problem with Blackmagic, I am confused about what is happening.

 

[Synchro3]

I did run Blackmagic with a 5 Gb test file. In Blackmagic, for whatever reason, the speeds remain about 200 MB/s regardless of the file size. However, in QuickBench the speeds increase to where they should be when the file size is larger.

I downloaded several different benchmarking apps for windows, and ran them in Boot Camp (HD Tune, ATTO). When testing larger block sizes, the speeds are where they should be as well. All except CrystalDiskMark, which acts like Blackmagic and finds slow speeds.

Ultimately, this may come down to some kind of difference in testing algorithms, but since others here are not having the same problem with Blackmagic, I am confused about what is happening.

Try AJA System test (File size 8 GB / Video frame size 4096*2160), and Quick Bench (Extended test 20-100 MB).

If all other benchmarks except Blackmagic are good, i wouldn't pay attention to the Black Magic bench.

 

[TzunamiOSX]

because it uses 4 PCIe lanes and all the smaller SSDs from Apple are only using 2 lanes AFAIK.

And that is my problem! How can i identify the Apple 4x Blades?

 

[OCDMacGeek]

Try AJA System test (File size 8 GB / Video frame size 4096*2160), and Quick Bench (Extended test 20-100 MB).

If all other benchmarks except Blackmagic are good, i wouldn't pay attention to the Black Magic bench.

Here is the bizarre thing: My 1 TB PCIE Blade utterly flunks the AJA System test described above. It only gets 100 MB/s write, and 250 MB/s read. But on the QuickBench test you describe, it passes with flying colors: 900+ MB/s read/write.

What could possibly account for this disparity in the tests?

Update:

As stated above, I had been getting great tests using Blackmagic at first. I tried to figure out what I had done before the Blackmagic tests plummeted. I had installed a Boot Camp partition. I deleted the Boot Camp partition but that did not increase speed with Blackmagic.

Around the same time the speed dropped, I had also moved my 650 GB iPhoto library onto the SSD. I deleted the large iPhoto library, and viola -- the tests using both Blackmagic and AJA System Test boosted right back up to where they should be (~900+ MB/s) like everyone else is getting.

I have heard talk about how SSDs can slow down when approaching being full. My 1 TB SSD had about 150 GB free when the Blackmagic tests tanked. But then I tried going halfway -- leaving 470 GB free, and the Blackmagic 5 GB stress tests still dropped to 400 MB/s r/w. (Not as bad as when 150 GB free, but almost.)

So, it seems that Blackmagic testing gets proportionally worse the more filled the drive is. But, as stated above, Quickbench and several Windows utilities always showed fast speeds even when the SSD was full.

Any advice on determining which one is real? Is Blackmagic correct that when the drive gets more full its speeds plummet and Quickbench is being to optimistic? Or is it vice versa?

It occurs to me that both AJA and Blackmagic are tests that seem to relate to video rendering performance, and those are the ones that tank when my SSD is full. QuickBench doesn't seem to measure things in terms of video file resolution. I don't know what that means, though.

I really would like to use my SSD to almost its full 1 TB (that was part of the point of a 1 TB blade!).

For what its worth, it "feels" fast, but thats so subjective.

 

[Loa]

Hello,

Bit of advice regarding the benchmarks. When in doubt, do a real world test. Create a RAM disk (download small app here) and put a big file there. Let's say you put a 5 GB file. Copy it to and from your RAM disk to your SSD and time it. Divide the size in MB (here 5000) by the number of seconds. Voilà. You'll get the read and write speeds.

If your workflow usually consists of many small files transfers, dump a huge bunch in there until you have 5GB. Divide that 5000MB by the number of seconds. Boom.

It's a bit more trouble, but this way you get the actual numbers for your systems with no fudging by the benchmark app. Clearly the benchmarks are a lot more complicated than that, and since they don't work consistently, this real-world method is the only way to know the actual speeds.

Loa

 

[OCDMacGeek]

Hello,

Bit of advice regarding the benchmarks. When in doubt, do a real world test. Create a RAM disk (download small app here) and put a big file there. Let's say you put a 5 GB file. Copy it to and from your RAM disk to your SSD and time it. Divide the size in MB (here 5000) by the number of seconds. Voilà. You'll get the read and write speeds.

If your workflow usually consists of many small files transfers, dump a huge bunch in there until you have 5GB. Divide that 5000MB by the number of seconds. Boom.

It's a bit more trouble, but this way you get the actual numbers for your systems with no fudging by the benchmark app. Clearly the benchmarks are a lot more complicated than that, and since they don't work consistently, this real-world method is the only way to know the actual speeds.

Loa

I did the RAM disk test you suggested, and it seemed to bear out the slow testing I was getting in Blackmagic when my SSD gets more full. There may be something wrong with this SSD.

I copied 9.86 GB of photos to the RAM disk. It completed at a nice speed of about ~900 MB/s. I copied the same folder back from the RAM disk to my SSD, and it did so at a horrible rate of 118 MB/s. I copied it right back to the RAM disk from where I dropped it, and it did so at 440 MB/s. This pattern could be repeated over and over. It seems to make a difference what folder on my SSD I am copying to or from.

When I copy to the RAM disk from some folders it is quite fast. When I copy to the RAM disk from my OS X desktop, its slow. When I copy from the RAM disk to anywhere, its slow.

I did more testing, and here is what I know: The SSD speeds get worse the more you fill up the blade. With up to 417 GB free, it was testing fine. With 216 GB free, the speeds went down to about 400 MB/s r/w. Then, with 150 GB free the speeds tank at ~200 MB/s r/w.

Questions:

(1) Has anyone else with a 1 TB PCIE blade tested what happens to their speeds when they only have 150-200 GB free?
If others are experiencing this, then perhaps its a phenomenon that just happens when these blades get full. If I'm the only one, then that's not the issue.

(2) Given that these blades seem to have a few memory chips that together form the 1 TB of data, would my results indicate that one "chip" is bad? I.e. everything functions just fine until I'm trying to read/write to a certain "chip" of the SSD, which gets accessed once I reach 200 GB free, and then things go to hell.

(3) Would the fact that my speeds tank as the SSD gets more full be indicative of anything else you can think of?

 

[Keter]

(2) Given that these blades seem to have a few memory chips that together form the 1 TB of data, would my results indicate that one "chip" is bad? I.e. everything functions just fine until I'm trying to read/write to a certain "chip" of the SSD, which gets accessed once I reach 200 GB free, and then things go to hell.

(3) Would the fact that my speeds tank as the SSD gets more full be indicative of anything else you can think of?

(2)I doubt, but possible.

(3) Yes, and no. SSD are designed with an excess of storage for Garbage Collection (GC).

What I would suggest is make a backup of your drive, if possible. Reboot to recovery mode and run a secure erase of your SSD, or use a tool for a secure erase. This guide is an excellent piece of info: http://www.pcworld.com/article/2088341/how-to-restore-your-ssd-to-peak-performance.html

Restore your data from your backup or complete a clean install and run each of your tests again. If the performance is low you will know your SSD is faulty.

 

[OCDMacGeek]

(2)I doubt, but possible.



(3) Yes, and no. SSD are designed with an excess of storage for Garbage Collection (GC).



What I would suggest is make a backup of your drive, if possible. Reboot to recovery mode and run a secure erase of your SSD, or use a tool for a secure erase. This guide is an excellent piece of info: http://www.pcworld.com/article/2088341/how-to-restore-your-ssd-to-peak-performance.html



Restore your data from your backup or complete a clean install and run each of your tests again. If the performance is low you will know your SSD is faulty.

Thanks for your suggestion. The article mentions Windows tools like Parted Magic. But, you suggest booting to recovery mode and doing a secure erase. Does that truly perform the same expunging of "garbage collection" on an SSD?

 

[Keter]

Thanks for your suggestion. The article mentions Windows tools like Parted Magic. But, you suggest booting to recovery mode and doing a secure erase. Does that truly perform the same expunging of "garbage collection" on an SSD?

You can use Disk Utility (may be disabled GUI now) but you can use the terminal command (CLI) or other tools like DiskWarrior or TechTool Pro. Basically you want the drive to write 0's or random data to every cell, preferable resetting the logical page map.

Samsung has their tool foo their drives. They perform for the most part the same exact functions.

Another option is a GParted boot CD and run the secure erase function from there.

You have a SSD, it shouldn't take too long to backup and reinstall a vanilla copy of OS X with a thumb drive.

 

[IowaLynn]

". As an SSD is filled with more and more data, there will naturally be fewer free blocks readily available. The SSD is then forced to actively consolidate valid data and prepare free blocks in order to write new data and perform maintenance. This process of moving and consolidating data takes time, which is perceived as decreased performance, and requires free space. This is why Over Provisioning, which guarantees a certain amount of free swap space to use for Garbage Collection and other maintenance activities, is so important for SSD performance"

Samsung SSD White Paper: Understanding SSDs

Set aside 12% for over provisioning, boot from another system device that has TRIM and run DU to invoke TRIM command or a special OS X maintenance system that is not used for anything else and has TRIM enabled.

I did not think you want to write zeros.

 

[handheldgames]

I picked up a MZ-JPV512R/A02, 512GB SSUBX x4 version of the Apple flash. 1500MB/S reads and writes! Performance appears to match the 1TB version.


Putting performance in persecutive. The 512 and 1TB SSUBX are 46% faster write and 72% faster Read performance compared to the x4 1TB SSUAX. Another great find on eBay. Synchro tipped me to a vendor who had a 1TB version on auction(many thanks again). There was a 512Gb version set to buy it now. Search and you may find.

 

[Synchro3]

I picked up a MZ-JPV512R/A02, 512GB SSBUX x4 version of the Apple flash. 1500MB/S reads and writes! Performance appears to match the 1TB version.

You're welcome. And cool SSD!

Could you also benchmark with AJA System Test? Just to compare graphically with my MZ-KPC1T00/0A2 (~SM 953)

Attached the results of the MZ-KPC1T00/0A2. (file size 8 GB / Video Frame Size 4K)

 

[handheldgames]

You're welcome. And cool SSD!

Could you also benchmark with AJA System Test? Just to compare graphically with my MZ-KPC1T00/0A2 (~SM 953)

Attached the results of the MZ-KPC1T00/0A2. (file size 8 GB / Video Frame Size 4K)

Absolutely. MZ-JPV512R/A02, 512GB SSUBX <---> MZ-KPC1T00/0A2

The Test Setup:
2009 Mac Pro 4,1. 2.66 base: Upgrades: 5,1 Firmware, 32GB GSkill ram, i7 990x, mPCI-e ->ExpressCard->FrescoLogic USB 3.0 SuperSpeed

PCIe1 - 680GTX
PCIe2 - XP941 128 w/Lycom adapter* (disconnects on deep sleep in PCIe2)
PCIe3 - 512GB SSUBX x4 PCI(499.55 free) w/sintech adapter* (requires SMC reboot on cold start?)
PCIe4 - 1TBGB SSUAX (System Boot 726GB free) w/sintech adapter

Notes from the Workbench
Getting two of the apple SSD's 100% functional on the cMP was more plug-and-pray than plug-and-play. On the path to optimal performance, I dealt with multiple instances of the cMP hung in the boot process attempting to mount the SSD's or booting into a less than desired performance envelope.

Sintech PCIe adapter - A Blessing and a Curse
The Sintech PCIe adapter is a bit difficult beast to work with and requires the most finesse to install that I've ever had to deal with. As far as I can tell, the distance of the PCB from the PCIe back plane is about 1mm+/- greater than it should be. So when the card is in place, a small amount of pressure is required on the back of the card, allowing it to slide into the PCIe slot.

PCIe2 recognizes as x4 but at 2.5 GT/s, limiting speeds under 900MB/s.
PCIe3 is problematic, but can be recognized at x4 5.0 GT/s. I'll explain below.

Scottie.. We need more power
Moving the card to PCIe3 with the 1TB in PCIe4, the system powers up but may not reach the boot sequence where the cMP site on a grey screen waiting for the Apple Logo. Switching adapters had limited success and ultimately found that resetting the SMC fixed the boot issue and allowed both Apple SSD's to share the single x4 connection for PCIe slots 3 & 4.

Update
The SSUBX is not behaving well in my cMP and appears to requires the SMC to be reset for detection to occur. Hopefully the correct combination and placement in the cMP will be easy to track down. As a fail safe, I'll test the SSD in my rMBP to make sure it's aOK.

 

[Synchro3]

Thanks. Going to the limits of PCIe 2.0 (4 lanes). Would be interesting to watch these new ones connected to PCI Express 3.0.

 

[TheStork]

I want to acquire an Apple 512GB or 1TB PCIe SSD, and I've been following the posts in this thread.

However, I can't find the model #s of the "blades" you all have been posting about on eBay.

Can you recommend an eBay seller and the model # (or #s) of the SSDs I should be considering. TIA.

 

[handheldgames]

Thanks. Going to the limits of PCIe 2.0 (4 lanes). Would be interesting to watch these new ones connected to PCI Express 3.0.

Here is the SM951 on PCIe 3.0 @x4 5.0 GTS on a windows box (from Ananadtech). At most PCIe 3.0 brings about a 100-200MB/s ~ 5-10% increase in speed.

I want to acquire an Apple 512GB or 1TB PCIe SSD, and I've been following the posts in this thread.

However, I can't find the model #s of the "blades" you all have been posting about on eBay.

Can you recommend an eBay seller and the model # (or #s) of the SSDs I should be considering. TIA.

There isn't one seller I'm aware of that has these SSD's in stock. Availability seems to be a luck of the draw. You may want to set multiple watches on Ebay to locate one when it's posted.

 

[Synchro3]

~1700 MB/s Read and ~1600 MB/s Write. Nice.

Anyway, cMac Pro users can just live with the 1500 MB/s. (nMp 1400 MB/s)

 

[handheldgames]

~1700 MB/s Read and ~1600 MB/s Write. Nice.

Anyway, cMac Pro users can just live with the 1500 MB/s. (nMp 1400 MB/s)

I have an Akitio Thunder2 PCIe Box on the way to test with a late 2013 rMBP's implementation of 20Gb/s Thunderbolt 2 with the 512R SSUBX. I'm looking forward to see how it performs against the Maximum theoretical bandwidth of 2,500 MB/s. With that in mind, It should be able to match the MacPro's x4 slot, unless my research was off in left field. (or missed something. which happens to all of us..)

 

[OCDMacGeek]

I have an OWC Helios 2 and a PCIE adaptor that fits my nMP's original 256 gb blade -- I don't know if it's a Sintech because I got it off EBay from a US based seller who wasn't sure. (I replaced my original 256 gb blade with a 1 TB, which is why I repurposed it). I'm only getting about 700 MB/s read and write in that set up, even though I recall the speeds being a bit higher when it was the primary drive in my nMP. This could be the fault of my adaptor. I need to get a Sintech one to be sure.

 

[Loa]

I'm only getting about 700 MB/s read and write in that set up

Sounds like the card is a 2X, as that's the speed limit for 2X cards. You need a 4X card.

Loa

 

[flehman]

The SSBUX is not behaving well in my cMP and appears to requires the SMC to be reset for detection to occur

Just to clarify, do you think that the problems you were having with the MZ-JPV512R/0A2 512GB (Apple part 655-1859B) were attributable to the fact that you were trying to get 2 of the Sintech adapters to play nice together in a single case? You don't think that the 512GB stick itself is more finicky than the 1TB? I wonder if it was by itself in the computer (i.e. only 1 Sintech adapter) it would be as reliable as either of the super-fast 1TB sticks (MZ-KPV1T00/0A2 or MZ-KPV1T0R/0A2)?

 

[handheldgames]

I have an OWC Helios 2 and a PCIE adaptor that fits my nMP's original 256 gb blade -- I don't know if it's a Sintech because I got it off EBay from a US based seller who wasn't sure. (I replaced my original 256 gb blade with a 1 TB, which is why I repurposed it). I'm only getting about 700 MB/s read and write in that set up, even though I recall the speeds being a bit higher when it was the primary drive in my nMP. This could be the fault of my adaptor. I need to get a Sintech one to be sure.

The only manufacturer of PCIe x4 Apple PCIe SSD adapters is Sintech, while there is a 2nd revision with a heat spreader, we're unsure if there were any changes to the PCB. On that front, you should be good. Regarding the PCIe expansion from OWC, lets just say they try hard, execute poorly and are known for great RAM warranties.

Maximum TB2 speed?
I just found this one on the net: "The maximum real world performance of this Thunderbolt 2 device is 1375MB/s". Over at Barefeats 2 XP941's in a Hardware Raid 0 are only hitting 1374.

Whereas in the cMP... The same SSD's in PCIe slots take the lead...

Just to clarify, do you think that the problems you were having with the MZ-JPV512R/0A2 512GB (Apple part 655-1859B) were attributable to the fact that you were trying to get 2 of the Sintech adapters to play nice together in a single case? You don't think that the 512GB stick itself is more finicky than the 1TB? I wonder if it was by itself in the computer (i.e. only 1 Sintech adapter) it would be as reliable as either of the super-fast 1TB sticks (MZ-KPV1T00/0A2 or MZ-KPV1T0R/0A2)?

The jury is still out on this one. I'm a bit stumped and have pulled out an Apple keyboard to reset the PRAM next since my circa '86 PS/2 Model-M Extended keyboard lacks the Command/Windows key. Will report back...

Reboot update: Zapping the PRAM, booting into 10.10, shutting down and cold boot into 10.10 worked! The grey-screen-of-death is more akin to a temperamental child that's refusing to eat it's oatmeal. Once the cMP forgets there is an issue with liking the 512GB SSBUX, further cold boots are not an issue. Ha!

SSUBX is the defining mark on all New Apple/Samsung SSD's
As a programmer, we're always looking for patterns to help explain what we're dealing with in code and in our lives. The one commonality I've seen on Apple/Samsung SSD's is the 5 digit Alpha code next to the size on the lower left corner of the SSD's label. Late 2013 Technology is tagged with SSUAX. Late 2014 Technology(the fast ones) are tagged with SSBUX, which seems to define the Platform where as Model specifies the Model and Revision. My .02 FWIW.