Cherryville - SSD 520

[johnw]

The latest Samsung SM825 review on storagereview explains it pretty well. They call it simply "Cache Power Protection" which essentially describes the functional purpose, however I do think power-loss protection is a nicer term for it. BTW, comparing the size and number of the capacitors with that on the Intel 320 is rather telling

The Samsung uses a large DRAM cache (256 MiB), and so needs a larger total capacitance than the Intel 320 series which has only a small DRAM cache (64 MiB) and so does not need as much capacitance to power the board until the cache can be flushed to flash. Also, I suspect the Samsung's controller uses more power than the Intel 320 SSD's controller.

Note that SR uses the term "ultra capacitor" to refer to the capacitors in the SM825. That means the same thing as "super capacitor" -- they are both names for an electric double layer capacitor.

Anyway, you can bet that the 4 capacitors on the Samsung SM825 board have a lot higher total capacitance than the 6 capacitors on the Intel 320 board, and they surely cost a lot more, too.

[minpayne]

The Samsung uses a large DRAM cache (256 MiB), and so needs a larger total capacitance than the Intel 320 series which has only a small DRAM cache (64 MiB) and so does not need as much capacitance to power the board until the cache can be flushed to flash. Also, I suspect the Samsung's controller uses more power than the Intel 320 SSD's controller.

Note that SR uses the term "ultra capacitor" to refer to the capacitors in the SM825. That means the same thing as "super capacitor" -- they are both names for an electric double layer capacitor.

Anyway, you can bet that the 4 capacitors on the Samsung SM825 board have a lot higher total capacitance than the 6 capacitors on the Intel 320 board, and they surely cost a lot more, too.

Anandtech says the Intel 320 Series does not store user data into the DRAM cache, but instead, in the SRAM cache. Anyway, the capacitors may be responsible for the safety of the FTL table or whatever mapping tables working in the DRAM cache.

I can confirm that the Samsung 830's controller uses A LOT MORE power than the Intel 320 Series', when bombing them with 1QD 4k random write full span. The temperature difference is very easy to tell!

[Ao1]

Moved to a new thread

[johnw]

Anandtech says the Intel 320 Series does not store user data into the DRAM cache, but instead, in the SRAM cache. Anyway, the capacitors may be responsible for the safety of the FTL table or whatever mapping tables working in the DRAM cache.

That is what I have always assumed, also. I don't see how an SSD could be both fast and reliable without storing its LBA-to-page mapping table in some sort of volatile cache. An SSD is useless without that table, so ideally it would be stored to non-volatile flash, but if it wrote the table to flash frequently enough to avoid any chance of data loss from an unexpected power outage, then it would wear out the flash quickly (if it used its own dedicated flash) or it would hurt performance (if it had to write to the same flash that user data is being written to). So that table must be kept in some sort of DRAM cache, but periodically flushed to flash, presumably when there is not a great deal of user writes going on, although the cache should obviously be flushed more frequently just after a lot of user writes have occurred. I think this sort of thing is key to the performance and reliability of an SSD, but we know very little about the details of the techniques used in the various SSDs -- trade secrets, I suppose.

[johnw]

Why oh why did I call it a power cap. I think this issue is interesting however as the approach taken by different vendors varies considerably.

Ha! It did bring up some interesting discussion. And thank you for posting the comparison of caching among the different SSDs, that is a good summary.

Sandforce drives

No DRAM, the NAND itself is used for transaction and cache buffer space.

I have never seen a number (size), but I am nearly certain that there is a significant SDRAM cache on the Sandforce controller itself. I guess at least 16MiB. As I said in my other post, I do not see how an SSD could have both good performance and reliability without caching the LBA-to-page mapping table in a volatile cache.

[DooRules]

Been using the drives for OS last few days now in R0. First pic is with the Enterprise drivers.



This is with the 11 series drivers.

[Computurd]

looking good. i am curious as to the performance after a long period of time with these drives. me thinks there may be tuning for steady state performance to remain rather high.........

[dbw09]

The retail kits (A3K5) are now available from Amazon and are eligible for prime even! Get them while they are hot... I picked up one of each just to test them all!

[Anvil]

Reviews are starting to appear

TheSSDReview
LegitReviews does 2R0
(haven't read any of them yet, will do asap)

Looks like the NDA is lifted, prices are a bit higher than the competition so the question is what do we get vs the already existing SF based SSDs.

[Ao1]

Toms:

http://www.tomshardware.co.uk/ssd-52...iew-32373.html

Anandtech

http://www.anandtech.com/show/5508/i...y-to-sandforce

[Smartidiot89]

From us @ NordicHardware:
http://www.nordichardware.se/lagring...orce-test.html

What's so special about this drive is that Intel have been working in close collaboration with SandForce for over a year on the firmware. Will be interesting to see how reliable their SSD's actually are, and if Intel can deliver what's promised (in terms of reliability) it will be worth the price premium imo

[johnw]

Comparing the AS-SSD benchmarks for the 256GB Corsair Performance Pro with the 240GB Intel 520

They are similar except in three respects, the 520 is a bit faster (5%) at sequential read while the Corsair destroys the 520 at sequential write -- 40% faster. The Corsair has much lower access times for 512B IOs (mostly only important for unaligned partitions).

As for pricing, it looks like the 240GB Intel 520 will cost at least $100 more than the 256GB Corsair Performance Pro.

Hardware heaven has a good review where you can compare the Corsair Performance Pro, the Samsung 830, and the Intel 520. Similar performance, except the Corsair does very well in the BlackMagic video test for writing video data, and the Samsung 830 falls down in Microsoft Outlook PST repair test.

http://www.hardwareheaven.com/review...ead-write.html
http://www.hardwareheaven.com/review...intenance.html

[Anvil]

I think I'm settling for the 180GB for testing.

It does look to be very little separating it from the competition though, the price is on high side but that is expected for a new drive.

A bit surprising that R.A.I.S.E is gone in favor of O.P.

[Gilgamesh]

On Newegg:
Intel 520 240GB - 550$
http://www.newegg.com/Product/Produc...82E16820167086

Samsung 830 256GB - 360$
http://www.newegg.com/Product/Produc...82E16820147135

Crucial Performance Pro 256GB - 400$
http://www.newegg.com/Product/Produc...%20pro%20256gb

Plextor PX-256M3 - 330$ (5 years warranty)
http://www.newegg.com/Product/Produc...82E16820249015

Corsair Force GT 240GB - 375$
http://www.newegg.com/Product/Produc...e%20gt%20240gb

Intel 520's price is crazy high for an "hacked firmware" and 5 years warranty

[Computurd]

they are saying that the RAISE is only for lower quality NAND. with Intel using high-binned NAND they do not feel it is necessary. I was surprised by that as well...but its not like Intel is going to slap TLC on it!

I would look for the price to drop after a time...of course.

[Trox]

On Newegg:
Intel 520 240GB - 550$
http://www.newegg.com/Product/Produc...82E16820167086

Samsung 830 256GB - 360$
http://www.newegg.com/Product/Produc...82E16820147135

Crucial Performance Pro 256GB - 400$
http://www.newegg.com/Product/Produc...%20pro%20256gb

Plextor PX-256M3 - 330$ (5 years warranty)
http://www.newegg.com/Product/Produc...82E16820249015

Corsair Force GT 240GB - 375$
http://www.newegg.com/Product/Produc...e%20gt%20240gb

Intel 520's price is crazy high for an "hacked firmware" and 5 years warranty

Totally agree with you, that price is overkill

[Ao1]

LOL the 60GB version is £114.72 or £1.91/GB in the UK. A Samsung 830 64GB drive is £89.98 or £1.40GB. I wonder what makes Intel think they can charge a premium for a product using a soon to be EOL crapforce controller.

[Computurd]

they know that the enterprise boys are gonna eat this thing up. Finally a SF controller that you can put in a server and not worry about reliability.

Highly compressible workloads are a huge portion of server loads, so these will perform very well.

even though intel isnt marketing these for the enterprise, they arent stupid. they know that a good many of them will land over there. even though its pricey for us, its pennies by enterprise standards.

[Leeghoofd]

Hardware heaven has a good review where you can compare the Corsair Performance Pro, the Samsung 830, and the Intel 520. Similar performance, except the Corsair does very well in the BlackMagic video test for writing video data, and the Samsung 830 falls down in Microsoft Outlook PST repair test.

http://www.hardwareheaven.com/review...ead-write.html
http://www.hardwareheaven.com/review...intenance.html

Looks good though pretty weird they give a 9 out of 10 for value, yet don't know the price... I'm shocked :p Hardwareheaven at it's best...

[johnw]

Looks good though pretty weird they give a 9 out of 10 for value, yet don't know the price... I'm shocked :p Hardwareheaven at it's best...

I usually ignore most of the commentary in SSD reviews. Usually the only worthwhile thing in any SSD review is the benchmarks, and the variety of SSDs that they run the benchmarks on. I happen to like many of the benchmarks that hardwareheaven runs, since they tend to be realistic tasks rather than some poorly documented "trace" that they play back. And they also are one of the few review sites to have the same benchmarks run on the CPP, Samsung 830, and Intel 520.

[Ao1]

@ John. I’m the opposite. I ignore the benchmarks, pile on the salt and then read the intro and conclusion. When someone starts benchmarking using the SNIA benchmark standard I will take notice. Talking of which did you ever get round to making some scripts to mimic the SNIA benchmark?

@ Comp, I agree Intel are indirectly targeting enterprise users and most will end up in enterprise applications. Do you know which OEM is buying the drives? I’d guess they are sold at a considerable discount. It might be a way to get one at a reasonable price (assuming anyone would want one).

@ anyone: I take it the 520 does not have a temp sensor? Intel just disabled the SMART attribute?

[Gilgamesh]

When someone starts benchmarking using the SNIA benchmark standard I will take notice. Talking of which did you ever get round to making some scripts to mimic the SNIA benchmark?

Here some bench with SNIA:
http://www.xbitlabs.com/articles/sto...1_4.html#sect0

Inviato dal mio GT-I9000 usando Tapatalk

[iboomalot]

Intel is very fast but cost is way too high

Iam hoping the 830 follows the M4 pricing soon and drops down into the 650-700 range. Kinda mad at myself for not pulling the trigger when egg had the 830-512 on sale for $729

If Intel treats the 520 like it does its top CPUs prices will remain high

[johnw]

@ John. I’m the opposite. I ignore the benchmarks, pile on the salt and then read the intro and conclusion. When someone starts benchmarking using the SNIA benchmark standard I will take notice. Talking of which did you ever get round to making some scripts to mimic the SNIA benchmark?

Why? Do you think it is irrelevant how long it takes an SSD to boot and load applications? Or load a game level? Or perform some disk intensive MS Office task?

I'm still working on the SNIA benchmarks. A few issues came up that I am still looking at. But I will be sure to post my results when I have them.

[johnw]

Here some bench with SNIA:
http://www.xbitlabs.com/articles/sto...1_4.html#sect0

Their claim that the tests are based on SNIA is misleading. Those tests are really only very loosely based on SNIA. And what is worse, they used the "pseudo-random" data setting on IOMeter, which is actually highly compressible. They should have used the full random setting for the data.