Five months ago we published a review of the SanDisk Extreme 240GB SSD. This drive, the first of its kind in more ways than we knew at the time, started a long and frustrating experience. The Extreme was the first product to cross our desk with new SanDisk 24nm Toggle Mode Flash. At that time, it was easy to point to the flash and say that any problem discovered had to do with the new exotic flash. What we didn't know at the time was the Extreme also brought with it our first look at Series 5 firmware from LSI SandForce.
The firmware system in place works roughly like this. The base code is produced by SandForce and that is where the revision number comes from. Before February 28th, drives coming in used what we call Series 3 (3.3.2, 3.3.4 and so on). The base code goes to manufacturers where it is key coded and customizations are added. In the customizations, features that produce Max IOPS, CapX 'Host Power Loss Protection' capability, throttle rate and other differentiating points are added. At that point the manufacturers release the code to the public. If there is a problem with the underlying base code, it affects everyone who builds on top of it. We see the same thing with motherboards, video cards and other products as well - if Intel's foundation code is off, then the problem is found on all motherboards using that underlying code.
In our testing the SanDisk Extreme 240GB was very fast, faster than any other SandForce based drive we've ever seen before" at first. The problem started to show up as we worked our way through the benchmarks. After a standard run we went back and re-tested the drive without a secure erase cycle and the performance dropped like a brick. These were the very same tests we'd just run a couple of hours before and achieved record setting scores. For the next month we worked our way through all of the typical issues that could pop up. System stability, drivers and so forth were all tested out of the equation. By that time I had six test systems running in the lab 24/7 and a whole fleet of SanDisk Extreme drives.
During all of this, other new drives hit the office and they all tested normal, all with 3.3.2 and 3.3.4 based firmware. It wasn't until the Biwin Elite arrived that we started making sense of the problem. Biwin uses standard firmware numbering and this drive arrived labeled as 5.0.1. At that point we were able to match the performance profile of the SanDisk Extreme to the Biwin Elite, both drives showed the same high performance that quickly dropped off in our testing. What was different about the Biwin Elite was the flash, 25nm synchronous IMFT flash. With that knowledge we determined the issue wasn't with the exotic 24nm flash, but the base code. The firmware was screwed and TRIM was no longer working. How could I of missed it went through my mind at least once a day for the next week.
The Biwin Elite arrived roughly three and a half months ago. Since then we've received 25 other SandForce based SSDs, all with the same issue, all with broken TRIM. A few of these products we've went ahead and reviewed, like the ADATA SP900, the first SandForce based drive with 0-provisioning and the SuperSSpeed Hyper Gold SLC, the first SLC flash drive we've tested in quite a long time. Basically we determined that any SSD coming to our lab would need to be re-tested and re-written once a firmware fix was available. Since the option wasn't very attractive we only published reviews of 'special' drives like the 0-provision and SLC.
At the time we were spoon fed the same line over and over, "two weeks". A new firmware "is close", "coming soon" and "in testing now." We really didn't expect two weeks to turn into five months, but our hands were tied to some degree because writing one review is tough enough, but writing two over the same product is exhausting.
TRIM arrives on 5-Series Code - But you have to wait for it
As you can imagine, with so many drives in for review we've been all over the next firmware release. Last week we learned from a drive manufacturer that new firmware was given to manufacturers, but it had to be tested first in validation labs. As you can imagine I had quite the four letter filled reply for validation testing this time around. It didn't help much though, but here we are a week later and the TRIM is fixed, the firmware code is now in my hands, from four different manufactures so far, all of it arriving in the last four hours.
I want it and I want it NOW
While I'm sure everyone with a SandForce based SSD wants to get their hands on working TRIM, you will have to wait for it "just two more weeks." At first, we were told there was a compatibility issue with 5.0.3 and an unnamed notebook chipset, but then that was redacted from the unnamed manufacturer. We were told and later confirmed that a new release, 5.0.4 was already in the oven. We're not sure what 5.0.4 brings with it, but some of the manufacturers stated they are going to wait for 5.0.4 to come out before releasing TRIM fixing code to their support download pages.
That said, we've heard from a few other manufactures and they plan on releasing 5.0.3 as soon as they finish their validation.
When Doing Wrong is the Only Right
In our reviews we talk specifically about how one company might have a solid track record with firmware releases. Some companies do a really good job and release code very quickly, but others might take months to release new code. In this case, one company stood out for NOT releasing firmware with broken TRIM, which we know of. OCZ Technology never did anything with 5.0.1 or 5.0.2. If another company should be on this list we apologize, but we tried to cover all of the companies we work with in the review cycle.
So, what about everyone else? That is a complicated issue, but I'll take the bait. Besides breaking TRIM, the 5 Series code also included the use of new flash. If a company wanted to use 24nm Toggle Mode flash, it had to go with the updated code to make it happen.
Series 5 firmware also brought with it a massive performance increase, at least until the drive started to slow down from normal use. With Series 5, SandForce based drives were able to outperform those based on Marvell controllers in PCMark Vantage when both drives were void of data, the way most reviewers test SSDs. The temptation was just too great for manufacturers to overcome and Series 5 was quickly posted to the support pages of most Team SandForce manufacturers. It should be said that not all of them were aware of the TRIM issue either... but some of them were aware of it.
But what about all of those reviews?
I'll be honest, this is a topic that just pisses me off. If a product was reviewed and an issue this big wasn't caught, then one of two things happened. The first is ignorance; the issue wasn't caught even though it's painfully obvious. Not every review is written by a full time, this is the only job they have, member of staff. If that was the case then no harm no foul, but I would avoid reading those reviews or be sure to add other sources for your information.
The second is absolutely in-freaking-conceivable; you were lied to in a cover up of the facts. We've talked about the TRIM issue for at least the last three months and just about everyone who reads TweakTown knows TRIM is not working on newer SandForce drives. The issue is discussed in various forums, on LinkedIn and other places where SSD reviewers openly discuss SSDs. As we said, this issue isn't difficult to find, anyone who spends more than 20 minutes testing a drive with broken TRIM and compares the data to SandForce based drives with working TRIM knows that something is not right! You really can't miss it; it's a giant red blinking sign two feet away from you.
While waiting for the new firmware to arrive we sat back and watched review after review hit the web, nearly every new drive shipped was without working TRIM. Time and time again the issue was passed over without mention. We knew the drives didn't work right; we had them all in our office for testing. That's when we ran across the image you see above that answered why the issue wasn't being talked about, it was being ignored.
In the image above you can see at the bottom the words VANTAGE AFTER SE. SE stands for Secure Erase and secure erase is the process used to bring a SSD back to fresh out of box (FOB) state. FOB is the point in which an SSD has fresh flash, ready to be written to without a full copy, erase write cycle. If someone runs SE in between every benchmark then they are not just ignoring the issue, they are covering it up. As one manufacturer put it, "it should be criminal." I'll stop just short of calling anyone out and let you make up your own mind, but it looks like bullshit to me.
There were a few reviewers that stepped up and openly discussed the TRIM issue. I managed to find the issue discussed by Bob from Think Computers, Jon from RWLabs and Hugh from HardOCP. Most likely there were others that I didn't find, but these are the confirmed TRIM talking guys who weren't worried about their sample supply train derailing.
Benchmarks - Test System Setup and A Modified Look at Testing
You can read more about TweakTown's Storage Product Testing Workstation and the procedures followed to test products in this article.
As I mentioned earlier, we now have firmware in hand that has working TRIM. This is the first time Series 5 firmware with working TRIM has been out of the validation lab and we've worked really hard to get it before anyone else.
As I write this, 5.0.3 is running through its paces on an SSD that was tested just hours before with 5.0.2. We want to thank Corsair for getting this over to us first. This was literally a race; the first to get us 5.0.3 was going to be the one to make this article happen. While running through the testing cycle we received code from three other companies and those full reviews will come out over the next few days. Given the large amount of drives we have on hand, we're hoping to publish a new SSD review every day leading up to Flash Memory Summit.
The presentation format for this review will be different than what you normally see at TweakTown. The benchmarks will be presented side-by-side and in the same order we run the tests. This is important because we'll be able to show how rapidly performance degradation occurs without TRIM.
The new 5.0.3 FW isn't just a TRIM fix, new features were added, and the drives gained some performance in some areas, but went back to 3.3.4 levels in others. We would love to just go side-by-side with code that was a little closer together, but LSI SandForce isn't going to make a TweakTown Edition firmware for testing" I already asked months ago.
Each benchmark will be presented side-by-side with the older 5.0.2 on the left and the new, TRIM fixed, 5.0.3 on the right. Let's go!
Unformatted Tests, Sequential and Access Times
HD Tune Pro
Just like with the reviews, these early tests are more for conditioning the drive, getting the flash read and written to. At the same time we get to see fresh out of box performance.
I'm not going to comment on all of the tests, but a few will get the analyst treatment. After the HD Tune Pro tests we can see in HD Tach that the working TRIM drive is in much better shape than the broken TRIM drive. The average write speed is already 100MB/s faster or in this case, 100MB/s more performance is retained.
Notice that each graph is on a different scale.
AIDA64 Random Access Time
The write access times (latency) are much smoother on the working TRIM model. As you can see, just from the light use we've already put these drives through the non-TRIM firmware is already going through the full read, erase, write cycle whereas the other drive gets the shortcut we call TRIM.
ATTO and PCMark Vantage Light Use
ATTO measures peak performance so we aren't going to see much deviation there.
PCMark Vantage Light Use
Here we see a complete smack down. My kids watch this every Friday night on TV. One little guy gets into the ring with a big massive dude who just beats the snot out of the pretender. So far I wouldn't say we've hit the drives very hard, about what your Windows and Office installs would do with your swap file on the OS drive after a week of use.
Even though we haven't hit it very hard, you can see a massive performance difference between these two sets of numbers. The 5.03 is very close to the FOB performance of roughly 79-80K, but the 5.0.2 is already showing signs of a stubbed toe.
Incompressible Data Tests
One of the area's hit the hardest that we've seen is the 4K write performance. When we first found the issue it was the 4K QD1 write speed that we noticed at first.
This benchmark is about to go away and be replaced with a new real-world test.
The scale is different on these charts as well.
Crystal Disk Mark
The harder we work the drives the lower the 5.0.2 goes, but with TRIM, the 5.0.3 is able to quickly recover on its own.
The first two tests are 100% read so there isn't much difference between the two other than the older FW is a little faster when straight reading data.
As soon as we mix a few writes in the non-TRIM FW panics and the TRIM FW keeps chugging along.
Compressible vs. Incompressible
Anvil Storage Utilities
46% Mix Application
Compressible Data Read IOPS, 32QD
Compressible Data Write IOPS, 32QD
92K write IOPS after a brutal run through our benchmarks. The Corsair Force GS is a manic with the SanDisk 24nm Toggle Mode flash.
Data on Drive - PCMark Vantage
Data Purge 0% Fill - The TRIM Test
With all of the data purged off the drive we see the 5.0.3 regain full performance and running within 2-3K points of FOB state. The 5.0.2 non-TRIM working FOB state starts out much higher, nearly 90K, but after a good work out, loses around 30% of its performance, until you clean the drive with a secure erase cycle.
When looking through the numbers you might think that the new 5.0.3 is just a faster firmware when compared to the older 5.0.2. That is NOT the case; 5.0.2 is a much faster firmware when fresh out of the box. 5.0.3 is built using more of the 3 Series code than the 5 Series. The difference on the drive we looked at today is 5.0.2 at roughly 90K Marks in Vantage FOB compared to roughly 80K Marks at FOB. Even though 5.0.2 starts out strong the performance quickly falls behind the TRIM enabled 5.0.3 and even though it's not shown today, it falls behind the amazing drives Plextor is putting out as well.
5.0.3 starts out a little slower, but is able to keep its composure when cycling through the benchmarks. This means it'll also be able to cycle through your daily use workload and stay fast too. If you had an SSD back before TRIM, with Windows XP then you know the hassles of secure erasing your SSD every three months and re-imaging your OS and programs. Nobody liked to do it then and no one should have to do it now. 5.0.3 will let you keep your OS install longer before the drive feels slow. It's really that simple, you get the performance right out of the box and you keep it.
Since I started writing this article two days ago I've heard from more manufacturers and more of them are getting ready to announce the new 5.0.3 FW on their websites. It sounds like within a week many of you with SandForce based SSDs will have access to this firmware. A few others seem to think 5.0.4 is right around the corner and they are holding off and seeing what happens.
In my testing 5.0.3 felt solid and we didn't have any issue. We've only run this drive on two Z77 motherboards so far and don't have the massive validation test lab that a Corsair or SanDisk has, but I'm going to keep using 5.0.3 and work on moving all of my drives over to it.
If you would like to learn more about TRIM, garbage collection and other advanced SSD technology we found a good white paper that covers all of the fine details. It's worth your time to look over it.
PRICING: You can find products similar to this one for sale below.
United States: Find other tech and computer products like this over at Amazon.com
United Kingdom: Find other tech and computer products like this over at Amazon.co.uk
Australia: Find other tech and computer products like this over at Amazon.com.au
Canada: Find other tech and computer products like this over at Amazon.ca
Deutschland: Finde andere Technik- und Computerprodukte wie dieses auf Amazon.de