Comments on: Benchmarking the World’s Fastest Client SSD

By: Justin Sykes

Justin Sykes — Tue, 08 Dec 2009 21:26:00 +0000

Thanks for the post, JMEDC.
1. Yes, firmware is upgradeable.
2. Micron won’t be handling retail sales for the C300, so we can’t promise what will be included (although a 3.5” adapter is a pretty standard offering). Watch for announcements of a distribution partner early next year.
3. Our SSD group won’t be at CES this year, but we will be at Storage Visions (Jan. 5-6) in Vegas. We’ll be hosting a keynote that explains why in-depth NAND expertise is going to be critical in the next generation of SSDs.

By: JMEDC

JMEDC — Sat, 05 Dec 2009 03:32:57 +0000

Thank you for the product demos. The performance of the RealSSD C300 is most impressive. Look forward to their arrival in the retail channel. A few questions:

1. Is the controller firmware upgradeable?

2. Will the drives be sold with adapter brackets to enable their installation in 3.5 inch drive bays?

2. Will Micron be demonstrating these drives at the January 2010 CES?

Thank You.

By: GullLars

GullLars — Fri, 04 Dec 2009 05:35:51 +0000

Thanks for the answer.

I would have thought you would go for more than 8 channels. It seems to me you could deliver more performance pr $ by having 1 NAND chip pr channel at the lowest capasity, and then increase capacity by adding more chips serially at the same channels. Assuming you use 32Gbit chips for 8GB capacity pr unit, the smallest C300 has 2 chips pr channel. Maybe this is because of controller complexity as the number of channels grow. On the other hand, the competing “drive x”‘s 10-channel controller has been on the market for over a year now, and another competitors products wich launch in Q1 2010 will have 16 channels.

Assuming performance scaling by # channels, a 16 channel drive with a controller that could cope on a SATA 6Gbps interface would saturate the interface in sequential reads, deliver 100.000 random read 4KB IOPS, 430MB/s sequential writes, and 80.000 random write 4KB IOPS.
By looking at competing products’ sustained write, i would guess you manage to write to 16 NAND chips at a time (or more at higher capacity) since the transfer-time is lower than array-time for writes, and as such write performance wouldn’t scale noticably by doubling the channels on the 128GB C300, but could double for 256GB.

Does the C200 also use 8 channels?

Looking forward to the random IOPS benchmark numbers

By: Justin Sykes

Justin Sykes — Thu, 03 Dec 2009 22:15:28 +0000

Thanks for the comment, Richard. Calling out competitors directly can sometimes invite unnecessary grief–besides, it’s the numbers that matter. We’ll let the tech bloggers do the labeled head-to-heads when the C300 is out in a few months.

Note: We didn’t have a chance to include a random/IOP test this time, but we’re planning to. Stay tuned–tomorrow, our random numbers won’t disappoint

By: Justin Sykes

Justin Sykes — Thu, 03 Dec 2009 22:14:35 +0000

Hey GullLars, You’ll find most of the answers to your questions on our web site: http://www.micron.com/products/real_ssd/ssd/client/index. It’s ONFI 2.1 NAND on 8 channels, and yes, we do support TRIM.

Sorry, we don’t reveal future product plans. PCIe does provide some very interesting performance potential, but the consumer market isn’t where these devices will take off.

By: GullLars

GullLars — Thu, 03 Dec 2009 16:04:41 +0000

@SF, ATTO does NOT test random packets, it is sequential. This can easily be seen by testing a harddrive.

@Richard, the benchmark numbers shows the X drive in this test is Intel x25-M gen2 160GB with TRIM firmware. It delivers 35-40K 4KB random read IOPS and 15K 4KB random write IOPS at a queue depth of 32, and read IOPS scales well from QD 1-10 at wich point it gets deminishing returns.

I still hope to hear if C300 supports AHCI and makes use of NCQ, the number of parallell flash channels, wich ONFI standard (2.1?), TRIM support, spare area, and if 40K/50K r/w 4KB random IOPS is sustained. If you can’t share that’s OK too, but a reply would be nice.

By: Justin Sykes

Justin Sykes — Thu, 03 Dec 2009 15:43:28 +0000

SF,
The NAND you are referring to is eNAND, and the C300 is not using it. For consumer/client SSDs, with our advanced wear-management, eNAND is overkill. However, this drive is not using run-of-the-mill consumer 2-bit per cell MLC NAND, either. Being a NAND manufacturer gives us flexibility in this area.

You can expect the C300 to be very competitively priced.

By: FS

FS — Thu, 03 Dec 2009 11:05:07 +0000

Hey,

I was wondering, what model is the other SSD that you are comparing the C300 to?

By: GullLars

GullLars — Thu, 03 Dec 2009 09:35:19 +0000

I also have another partially related question.
It’s regarding the PCI express cards you posted a demo of last november that did 200.000 IOPS random read at 2KB and 800MB/s bandwitdh.
Are you considering releasing PCIe based SSDs in the comming year?

This seems to me the optimal solution, since you can avoid the latency of HBA’s designed for rotating media, and also the bandwidth limitations these often have. There are already (MLC) PCIe SSDs in the market that beat your new SATA 6Gbit SSD in both bandwith and IOPS. Even the lower end PCIe SSD can do 600-1000 MB/s reads and 300-400MB/s writes, though these have mediocre IOPS.

By: GullLars

GullLars — Thu, 03 Dec 2009 08:31:16 +0000

This seems like a great drive. I just have a few questions left after this video.

1. What ONFI standard is this drive using?

2. How many internal parallell flash-channels does this unit have? (brand x in the video has 10)

3. Does this unit fully support ACHI and make use of NCQ?

4. Will it support TRIM?

5. How much spare area is reserved for the controller?

6. What is the SUSTAINED sequential write and SUSTAINED 4KB random write numbers?

7: Can you share a rough, non-binding, estimate of the prices of these drives? (say nearest $10-50)

Also, do you have the opportunity to run IOmeter and make following graphs?
A: 4KB 100% random 100% read, Queue Depth scaling at X axis 2^n n=0-8. (1, 2, 4, 8, 16, 32…)
B: 4KB 100% random 100% write, Queue Depth scaling at X axis 2^n n=0-8.
And C: 8KB 80% random 80% read, Queue Depth scaling at X axis 2^n n=0-8.
This will provide very interresting information for all us benchmarkers.

The IOPS info you reply with to the first post is about the same as the enterprise version of brand x, wich is really nice if the IOPS figures you quote are sustained numbers for an MLC device.

I currently own 2x Mtron Pro SSDs and 2x OCZ Vertex, but if the prices on these drives become reasonable i think i might just buy a few of these too