Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

scottrichardson

macrumors 6502a
Original poster
Jul 10, 2007
718
297
Ulladulla, NSW Australia
Hi all,

I recently added a 960GB OWC Accelsior SSD RAID PCI card to my 09' Mac Pro. I wasn't 100% happy with the results so I decided to also buy and test a Sonnet Tempo SSD Pro PCI Card with 2 x Samsung SSD 840 Pro 512GB Drives in RAID 0.

The benchmark results are in and the Sonnet wins. I was able to hit over 1040MB a second and across the board, it is faster at every block size for both read and write. To get the best results for small and large files, I set the RAID 0 block size to 16KB. While this yielded slightly lower large transfer speeds over a block size of 32K, it gave better overall speeds for smaller files 4K, 8K, 16K etc. With 16K block sizes on the RAID, I am getting just over 1000MB a sec with the largest transfers.

I am pasting the results comparison between the Accelsior and the Sonnet below. Only in a very small area does the Accelsior beat the Sonnet/Samsung combo.

Enjoy :)
 

Attachments

  • sonnet-accelsior-benchmarks.jpg
    sonnet-accelsior-benchmarks.jpg
    308.6 KB · Views: 4,239
I'm probably blind but what stripe size did you use for your RAID0?

I used 16K.

I tried with 32K, but found the 4K and 8K transfers were slower than Accelsior. 32K DID yield slightly faster results for large transfers, but we're talking 2-3% difference - and for me, I care more about the smaller blocks being faster.
 
Why is the test volume size and free space different in the 1st comparison test?

Because I tested the Accelsior while I was using it as my boot drive - so it had data on it already :) But the speed tests were essentially the same when the drive was fresh.

----------

I can also confirm that the Sonnet Tempo SSD Pro Boots just fine into Mac OS X with the two Samsings in RAID 0. It also booted fine with just a single SSD on it.
 
Thanks for your follow-up. I gather this is bootable? Any issues with wake from sleep? If it's all good, this is a great solution that is better than a pair of Apricorn Solo X2 in RAID0 for only $100 more... and a MUCH better solution in terms of performance, flexibility, and price compared to the Accelsior.

To summarize, if you want 1TB of bootable SSD, these are basically your options (from lowest cost to highest):
1. Pair of SSD's on the SATA2 backplane (max STR of ~550MB/s); Cost: $0+Drives
2. Pair of SSD's on a pair of Solo X2 cards (max STR of ~600MB/s); Cost: $200+Drives
3. Pair of SSD's on a Sonnet Tempo (max STR of ~1000MB/s); Cost: $300+Drives
4. OWC Accelsior (max STR of ~650MB/s); Cost $1400 incl. drives
 
Because I tested the Accelsior while I was using it as my boot drive - so it had data on it already :) But the speed tests were essentially the same when the drive was fresh.

----------

I can also confirm that the Sonnet Tempo SSD Pro Boots just fine into Mac OS X with the two Samsings in RAID 0. It also booted fine with just a single SSD on it.

I see, Thanks!
 
Thanks for your follow-up. I gather this is bootable? Any issues with wake from sleep? If it's all good, this is a great solution that is better than a pair of Apricorn Solo X2 in RAID0 for only $100 more... and a MUCH better solution in terms of performance, flexibility, and price compared to the Accelsior.

To summarize, if you want 1TB of bootable SSD, these are basically your options (from lowest cost to highest):
1. Pair of SSD's on the SATA2 backplane (max STR of ~550MB/s); Cost: $0+Drives
2. Pair of SSD's on a pair of Solo X2 cards (max STR of ~600MB/s); Cost: $200+Drives
3. Pair of SSD's on a Sonnet Tempo (max STR of ~1000MB/s); Cost: $300+Drives
4. OWC Accelsior (max STR of ~650MB/s); Cost $1400 incl. drives

Correct mate. I have tested all of the above except option 2.

Yes this is bootable 100%. But alas I cannot advise about wake from sleep right now, as my Mac doesn't wake from full system sleep due to me running a Gigabyte GeForce 660 Ti GPU which seems not to wake my displays after a full sleep (any advice or tips on this is appreciated... anyone?).

I am extremely impressed with the performance.

As an example... today I duplicated a 1.68GB photoshop file, and it went so fast that the Finder didn't even have time to display the progress bar. Seriously cool stuff!

I'll be packing up the Accelsior and returning it tomorrow.

Scott
 
Wow, with this "Sonnet Tempo SSD Pro PCI Card" and a new 670 or 680 graphics card, the current mac pro could stand with the best of them today. The sonnet is bootable? has a recovery partition?
 
Wow, with this "Sonnet Tempo SSD Pro PCI Card" and a new 670 or 680 graphics card, the current mac pro could stand with the best of them today. The sonnet is bootable? has a recovery partition?

Bootable yes - but NO RECOVERY PARTITION when running 2 x SSD as a RAID 0 stripe, sorry. You will need a spare SSD or standard drive for that purpose. Which is why I'm keeping my old 240GB OWC Mercury Extreme Pro SSD with Windows 8 on a bootcamp partition and 10.8 on another partition with a recovery partition to go along with it.
 
Bootable yes - but NO RECOVERY PARTITION when running 2 x SSD as a RAID 0 stripe, sorry. You will need a spare SSD or standard drive for that purpose. Which is why I'm keeping my old 240GB OWC Mercury Extreme Pro SSD with Windows 8 on a bootcamp partition and 10.8 on another partition with a recovery partition to go along with it.

Cool, thanks.
 
What's the difference between the "Pro" and regular versions of the Tempo SSD card, and is that likely to manifest in a noticeably different experience?
 
What's the difference between the "Pro" and regular versions of the Tempo SSD card, and is that likely to manifest in a noticeably different experience?

The 'Pro' card is capable of much faster overall speeds due to it using a more powerful controller chip with greater bandwidth.

You're looking at around 650MB/sec versus 1000MB+/sec on the Pro. (each card with 2 drives in RAID 0)
 
Sonnet Tempo SSD Pro vs OWC Accelsior (benchmarks)

Scott, thanks for all of this great info! Unfortunately, I read it a all bit too late. I just ordered a 960gb accelsior and a 480gb accelsior from OWC. They will likely be in here in the next day or two. Did you have to pay the 15% restocking fee to OWC?

Also, do you think that the Accelsior may have been slower than the Sonnet Tempo config because it was busy running your OS as well as a disk test?
 
Last edited:
I used 16K... I tried with 32K, but found the 4K and 8K transfers were slower than Accelsior.

That's to say that the Accelsior card actually performed faster than the Tempo solution depending on the block size. This of course leads the question; what is the optimal block size to test with, and which one most realistically reflects an OSX boot disc, with apps., vs. a data scratch disc?
 
That's to say that the Accelsior card actually performed faster than the Tempo solution depending on the block size. This of course leads the question; what is the optimal block size to test with, and which one most realistically reflects an OSX boot disc, with apps., vs. a data scratch disc?

Well the accelsior was only slightly faster in just a small portion of the benchmarks. As soon as you were up over 16K the sonnet solution streaked ahead. When I dropped the block size from 32 to 16K the sonnet + Samsung raid was substantially better as you can see in the screen shot.

Is there a way to determine or edit the block size on the accelsior?!
 
That's to say that the Accelsior card actually performed faster than the Tempo solution depending on the block size. This of course leads the question; what is the optimal block size to test with, and which one most realistically reflects an OSX boot disc, with apps., vs. a data scratch disc?

From what I've read and the little testing I've done personally, optimal RAID0 stripe sizes for SSDs vary from drive to drive, rather than by storage I/O profile. Back in HD days, optimal stripe size was a function of I/O profile... small stripes for small random file access patterns and large stripes for large sequential files access patterns. This made sense given the moving head/platter arrangements of HDs. With SSDs things are different, there are no moving parts, but instead, the way the NAND blocks are organized and how the controller works is more important... and unfortunately since SSD vendors don't provide any recommendations, one must test different RAID0 stripe sizes with any given set of drives to find the optimal performance. For example, my old first-gen Intel drives loved large stripe sizes like 128K because the NAND blocks were big and any writes smaller than that would result in a nasty write amplification penalty. On my new Vertex 4 drives, they seem to perform best with very small stripe sizes (16K or even smaller).

Another more important factor in selecting an SSD is the controller. For example, SandForce controllers use an innovative on-the-fly compression of data to improve write performance and reduce write amplification. The problem is, this only works with data that is highly compressible. So it works fine with normal executable files and office files, etc. but SandForce SSDs actually perform worse than most others with incompressible data like images and video (which are already compressed). So using a SandForce drive for OS/Apps can be a great thing, but it's not the best choice for a media storage device.

The Vertex 4 uses another innovative feature to improve write speeds. In fact, there's no MLC based drive I'm aware of that writes incompressible data faster. There's an interesting and cool reason for this. As you may know, SLC NAND is single-bit per cell technology and it's super fast to write data to although it's very expensive. MCL NAND on the other hand, is dual-bit per cell so it's much more affordable but it takes a lot longer to write data to, hence write speeds suffer. This (and the endurance benefits) is why you see SLC NAND typically used in enterprise drives. It seem OCZ took advantage of the fact that you can treat MLC NAND somewhat like SLC NAND if you're only writing one bit per cell at any given time, and thus gain some write performance. Of course, this will only work up to 50% of the drives capacity. After which, the garbage collection needs to reorganize the data from single bit per cell to double bit per cell to free up NAND. This is pretty effective if you never write more than 50% of your free capacity at any given time. You effectively get SLC write speeds using MLC NAND when writes are not more than 50% of available free space.

Anyway, the bottom line is that it's important to choose the right SSD for job and test your SSDs performance with different stripe sizes if your planning to run in RAID0 if you really want to squeeze every last bit of performance out of them. :)
 
  • Like
Reactions: MrAverigeUser
(2) Sonnet Tempo SSD Pro + (4) Samsung 440 Pro's Benchmarks

Well the accelsior was only slightly faster in just a small portion of the benchmarks. As soon as you were up over 16K the sonnet solution streaked ahead. When I dropped the block size from 32 to 16K the sonnet + Samsung raid was substantially better as you can see in the screen shot.

Is there a way to determine or edit the block size on the accelsior?!

From what I've read and the little testing I've done personally, optimal RAID0 stripe sizes for SSDs vary from drive to drive, rather than by storage I/O profile. Back in HD days, optimal stripe size was a function of I/O profile... small stripes for small random file access patterns and large stripes for large sequential files access patterns. This made sense given the moving head/platter arrangements of HDs. With SSDs things are different, there are no moving parts, but instead, the way the NAND blocks are organized and how the controller works is more important... and unfortunately since SSD vendors don't provide any recommendations, one must test different RAID0 stripe sizes with any given set of drives to find the optimal performance. For example, my old first-gen Intel drives loved large stripe sizes like 128K because the NAND blocks were big and any writes smaller than that would result in a nasty write amplification penalty. On my new Vertex 4 drives, they seem to perform best with very small stripe sizes (16K or even smaller).

Another more important factor in selecting an SSD is the controller. For example, SandForce controllers use an innovative on-the-fly compression of data to improve write performance and reduce write amplification. The problem is, this only works with data that is highly compressible. So it works fine with normal executable files and office files, etc. but SandForce SSDs actually perform worse than most others with incompressible data like images and video (which are already compressed). So using a SandForce drive for OS/Apps can be a great thing, but it's not the best choice for a media storage device.

The Vertex 4 uses another innovative feature to improve write speeds. In fact, there's no MLC based drive I'm aware of that writes incompressible data faster. There's an interesting and cool reason for this. As you may know, SLC NAND is single-bit per cell technology and it's super fast to write data to although it's very expensive. MCL NAND on the other hand, is dual-bit per cell so it's much more affordable but it takes a lot longer to write data to, hence write speeds suffer. This (and the endurance benefits) is why you see SLC NAND typically used in enterprise drives. It seem OCZ took advantage of the fact that you can treat MLC NAND somewhat like SLC NAND if you're only writing one bit per cell at any given time, and thus gain some write performance. Of course, this will only work up to 50% of the drives capacity. After which, the garbage collection needs to reorganize the data from single bit per cell to double bit per cell to free up NAND. This is pretty effective if you never write more than 50% of your free capacity at any given time. You effectively get SLC write speeds using MLC NAND when writes are not more than 50% of available free space.

Anyway, the bottom line is that it's important to choose the right SSD for job and test your SSDs performance with different stripe sizes if your planning to run in RAID0 if you really want to squeeze every last bit of performance out of them. :)

VirtualRain and ScottRichardson thank you for giving me inspiration from your advice and benchmarks. I had 2 Accelsior's in my hand but returned them for the setup below. Check out these benchmarks! Almost 1500mb/s reads w/3 Samsung's 840 Pro's in Raid 0. Pretty sweet.

For anyone interested, this is what two Sonnet Tempo Pro cards and (3) Samsung 840 Pro's in a Raid 0 will do.
(plus a separate Samsung 840 pro 512GB for the OS also mounted on the Sonnet card)

I used the top two slots in my mac pro because a USB 3.0 card is in the 16 lane slot. No signs of bottlenecking.

(2) Sonnet Tempo Pro PCI-E SSD
(3) Samsung 840 Pro 512 GB SSD in Raid 0 (scratch)
(1) Samsung 840 Pro 512GB SSD (OS and Apps)

OS: 10.6.8
RAM 48GB
12 x 2.93GZ Mac Pro

Stripe Size 64k (32k was actually slower in small and large reads in most cases)
 

Attachments

  • 64ksmall.jpg
    64ksmall.jpg
    144.7 KB · Views: 1,160
  • 64klarge.jpg
    64klarge.jpg
    119.8 KB · Views: 1,022
VirtualRain and ScottRichardson thank you for giving me inspiration from your advice and benchmarks. I had 2 Accelsior's in my hand but returned them for the setup below. Check out these benchmarks! Almost 1500mb/s reads w/3 Samsung's 840 Pro's in Raid 0. Pretty sweet.

For anyone interested, this is what two Sonnet Tempo Pro cards and (3) Samsung 840 Pro's in a Raid 0 will do.
(plus a separate Samsung 840 pro 512GB for the OS also mounted on the Sonnet card)

I used the top two slots in my mac pro because a USB 3.0 card is in the 16 lane slot. No signs of bottlenecking.

(2) Sonnet Tempo Pro PCI-E SSD
(3) Samsung 840 Pro 512 GB SSD in Raid 0 (scratch)
(1) Samsung 840 Pro 512GB SSD (OS and Apps)

OS: 10.6.8
RAM 48GB
12 x 2.93GZ Mac Pro

Stripe Size 64k (32k was actually slower in small and large reads in most cases)

Awesome... Nice scaling.

I found my own three drive array on my Highpoint card bottlenecked by something associated with the top slots... It topped out at 1GB/s in the 3rd slot but would hit 1.3GB/s in the 2nd slot. Perhaps that's resolved in your case by splitting the array across both top slots whereas my single card solution only used one slot.

At any rate, that Sonnet Tempo appears to be the best choice for PCIe SSD storage... Bar none.
 
I need some help ...

I installed a Sonnet Tempo SSD Pro in a new Mac Pro with 2 Crucial M4 512GB disks configured as RAID-0 and installed the current OS X on it. I also have the original 1TB disk in there with the same OS X installation on it. I was able to do the "Option" key on startup and see the bootable choices icons as expected.

I then added a Velocity 2 PCIe card with a Samsung 840 Pro on it and used WinClone to put a image of Windows 8 on it. I set the Startup Disk to the Windows installation and rebooted into Windows with no problems. I ran some Windows updates which rebooted with no issues.

However, now when I do the "Option" key startup to select the boot drive, all I get is a mouse pointer with no boot options. I have reinstalled both OS X images, and re-partitioned and formatted the Windows SSD with no changes to the boot options. I have done SMC and PRAM resets to no avail.

Any idea how to get my boot manager working properly again?



Thanks,
-howard
 
I need some help ...

I installed a Sonnet Tempo SSD Pro in a new Mac Pro with 2 Crucial M4 512GB disks configured as RAID-0 and installed the current OS X on it. I also have the original 1TB disk in there with the same OS X installation on it. I was able to do the "Option" key on startup and see the bootable choices icons as expected.

I then added a Velocity 2 PCIe card with a Samsung 840 Pro on it and used WinClone to put a image of Windows 8 on it. I set the Startup Disk to the Windows installation and rebooted into Windows with no problems. I ran some Windows updates which rebooted with no issues.

However, now when I do the "Option" key startup to select the boot drive, all I get is a mouse pointer with no boot options. I have reinstalled both OS X images, and re-partitioned and formatted the Windows SSD with no changes to the boot options. I have done SMC and PRAM resets to no avail.

Any idea how to get my boot manager working properly again?



Thanks,
-howard

It's possible you may have data corruption depending on how you installed the OS's. I can confirm Scottrichardson's previous post about the sonnet tempo card having issues when writing an OS to drives mounted directly to it AND in raid 0.

This card reads data off the drives fine in a raid 0 setup but it can sometimes crash when trying to write complicated directories or complex folder systems such as a large iTunes Media folder or OS (only in raid 0) to its drives.

Therefore, large complex directories such as an OS should be installed in a software raid 0 directly on the sata backplane first...and then transfer the drives over to the sonnet card afterwards.

I don't experience a total data loss when it crashes though. A simple restart fixes the issue and the raid is back up and running 100%. For me it's not an issue, but can be a headache when writing or copying large, complicated folder systems (we're talking directories over 10GB in size).

Possibly that's what happened to you? Did you install Windows or a Mac OS to the drives while directly mounted to the Sonnet card? If so, it's likely you may have may have data corruption. Try installing the OS's without the Sonnet card in a software Raid 0 directly to the mac (via SATA backplane). Then move the drives over to the card. Things should work great by then.
 
It's possible you may have data corruption depending on how you installed the OS's. I can confirm Scottrichardson's previous post about the sonnet tempo card having issues when writing an OS to drives mounted directly to it AND in raid 0.

This card reads data off the drives fine in a raid 0 setup but it can sometimes crash when trying to write complicated directories or complex folder systems such as a large iTunes Media folder or OS (only in raid 0) to its drives.

Therefore, large complex directories such as an OS should be installed in a software raid 0 directly on the sata backplane first...and then transfer the drives over to the sonnet card afterwards.

I don't experience a total data loss when it crashes though. A simple restart fixes the issue and the raid is back up and running 100%. For me it's not an issue, but can be a headache when writing or copying large, complicated folder systems (we're talking directories over 10GB in size).

Possibly that's what happened to you? Did you install Windows or a Mac OS to the drives while directly mounted to the Sonnet card? If so, it's likely you may have may have data corruption. Try installing the OS's without the Sonnet card in a software Raid 0 directly to the mac (via SATA backplane). Then move the drives over to the card. Things should work great by then.

Thanks for the information. :)

I did originally perform a clean install of OS X 10.8.3 followed by a data migration from a system disk to the RAID-0 SSDs on the PCIe card. I currently have the SSDs removed from the Sonnet card and installed in the normal drive bays, and it is running fine and the startup boot-selection screen is again working.

At your suggestion, I will now repeat the clean install and migration while the SSDs are mounted in the standard MP disk trays, and then transplant the drives back to the Sonnet PCIe card and see if that solves the problem.

Would you know if it is possible, while booting/running OS X from the Sonnet RAID-0 based PCIe (external) system, to install bootable Windows using any of the popular methods on a separate SSD mounted to preferably a PCIe card (Velocity x2 - seen as external), or a tray mounted SSD, and have it seen by the boot-manager? Here again, it may make a difference if I do the installation locally from backplane mounted drives and then transplant the drives to the PCIe cards.

Just to clarify ... the RAID-0 array works great either on the Sonnet card or on the Mac Pro backplane (of course it is much faster on the Sonnet card). It is only the startup boot-manager which can't be seen with the Sonnet card, but which works normally when installed to the normal blackplane.


Thanks for your help,
-howard
 
Last edited:
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.