OCZ Vertex Drives

[Ao1]

^^^^

Can you even get three consistent benchmarks in quick concession on the same set up? I haven't tried it but I'd bet not. Each benchmark is fragmenting the drive and slowing down read/ write and IOPS and in addition the results are likely to be variable depending on prior usage patterns.

Benchmarking on an OS drive is also going to throw out results. (I seem to recall OCZ saying you need to benchmark on a non OS drive for accurate results).

Benchmarking seems to be the best way to cripple your SSD drive performance and get useless results in the process.

[Hoss331]

This is my single 120gb with the original firmware and the new 1199 FW. This was with the offset at 512, 256 and 512 seem to be about the same and 128 has about the same writes but lower reads.

original


1199 FW

[Halk]

^^^^

Can you even get three consistent benchmarks in quick concession on the same set up? I haven't tried it but I'd bet not. Each benchmark is fragmenting the drive and slowing down read/ write and IOPS and in addition the results are likely to be variable depending on prior usage patterns.

Benchmarking on an OS drive is also going to throw out results. (I seem to recall OCZ saying you need to benchmark on a non OS drive for accurate results).

Benchmarking seems to be the best way to cripple your SSD drive performance and get useless results in the process.

I can get similar results. Have a look at this thread where I'm mucking about with different BIOS SATA modes. The results are not identical, but you can see a theme.

[Griff805]

I can get similar results. Have a look at this thread where I'm mucking about with different BIOS SATA modes. The results are not identical, but you can see a theme.

Try 3-5 benchmarks on the same setup just to see consistency- I think that's what he was after-

[F@32]

I´m using a single 60GB vertex on my system in specs. So far very pleased. What kind of real world performance would I get if I bought another and put them in raid0? How many boot seconds would I be able to cut of approximately?

Haha, well, you'd actually add time to your boot sequence due to RAID initiation

Try 3-5 benchmarks on the same setup just to see consistency- I think that's what he was after-

You can get consistent results by following this order: set-up the drive -> bench ->write zeros -> set-up the drive -> bench. It's not possible otherwise.

[Ao1]

I can get similar results. Have a look at this thread where I'm mucking about with different BIOS SATA modes. The results are not identical, but you can see a theme.

Halk, was it the AMD drivers/ AHCI mode or was it fragmentation that changed your results?

I'd suggest that unless you format after each test you will not get proper results because benchmarking adversely fragments all SDD drives. Maybe you get a ball park idea, maybe not.

For a valid benchmark you need a formatted non OS drive, otherwise you are just testing how badly you have fragmented your drive.

I can't understand why anyone would want to cripple the performance of an OS based SDD by benchmarking (assuming they are not formatting afterwards)......it's a lot of money to spend to then loose performance by pointlessly fragmenting it.

It would be worth asking the guys over at the OCZ forum for their opinion on how long it takes their drives to recalibrate after a test if you don't format. That would enable people to decide if they should be benchmarking or not on drives they are using for their OS.

[Halk]

Halk, was it the AMD drivers/ AHCI mode or was it fragmentation that changed your results?

I'd suggest that unless you format after each test you will not get proper results because benchmarking adversely fragments all SDD drives. Maybe you get a ball park idea, maybe not.

For a valid benchmark you need a formatted non OS drive, otherwise you are just testing how badly you have fragmented your drive.

I can't understand why anyone would want to cripple the performance of an OS based SDD by benchmarking (assuming they are not formatting afterwards)......it's a lot of money to spend to then loose performance by pointlessly fragmenting it.

It would be worth asking the guys over at the OCZ forum for their opinion on how long it takes their drives to recalibrate after a test if you don't format. That would enable people to decide if they should be benchmarking or not on drives they are using for their OS.

It was AHCI + AMD drivers together that caused the performance drops. If I run in either IDE mode, or AHCI without AMD drivers then the ATTO scores are pretty much the same. Running ATTO doesn't seem to degrade performance. I think you're maybe overreacting to the effect benchmarks have on the drive. Someone had a figure of 50GB writes per day for 16 years before the drives "run out". I don't know if that's true or not, but I suspect a couple of dozen ATTO runs will be neither here nor there.

[Ao1]

It was AHCI + AMD drivers together that caused the performance drops. If I run in either IDE mode, or AHCI without AMD drivers then the ATTO scores are pretty much the same. Running ATTO doesn't seem to degrade performance. I think you're maybe overreacting to the effect benchmarks have on the drive. Someone had a figure of 50GB writes per day for 16 years before the drives "run out". I don't know if that's true or not, but I suspect a couple of dozen ATTO runs will be neither here nor there.

I don't have an OCZ drive so I can't test it to find out, but logic says that benchmarking will fragment the drive. Why not give it a try or ask OCZ what it is doing to the drive?

EDIT: It's not about the drive "running out" it's about the way that SDD's relocate data after a benchmark, which is causing the slow down. The PC Perspective review has already shown how badly benchmark performances are affected on the X25-M drive. The post I made with the extract from the Intel statement is explaining why that it happening. Maybe OCZ drives are not affected as badly in terms of how long they take to adapt back, but I think it is safe to say that benchmarking is not doing an SSD drive any short/ mid term favours if you continue to use it without formatting.

[Ao1]

I´m using a single 60GB vertex on my system in specs. So far very pleased. What kind of real world performance would I get if I bought another and put them in raid0? How many boot seconds would I be able to cut of approximately?

IMO.....you don't need raid O to get more (tangible) performance out of Intel or the Vertex drives.


Hard raid with cache was only a benefit to earlier OCZ drives models that stuttered, but that does not seem to be the case with the Vertex.


The problem with hard raid is they are geared to HDD, they have not had chance to adapt to target high-speed random data in small chunks. Consequently performance of an SDD may even be reduced.


The only benefit of hard raid 0 is additional capacity. The down sides are very long boot times, heat, power consumption & added latency.


I use hard raid and raid 0 but I do that for capacity and imo hard raid gives more stability over soft raid, but I came to that conclusion before I switched to SSD. If I did not need capacity I would not be using raid 0.

[HDCHOPPER]

audienceofone thats exaly what eye have been seeing with my set up hard or soft raid , think will keep the hard card till eye get 2 more drives and re test

probbly end up selling it lol

[Halk]

I have a hardware RAID card arriving in the next few hours which is for the 4 WD RE3 1TB drives I bought last week. They'll be going into RAID5. I'll probably buy another Vertex drive and whack that on there too. I noticed better performance on a hybrid (cheap) RAID card, the 1430SA than I did on the motherboard. I've also noticed that people running 2 or 3 Vertex drives on RAID seem to get better performance than just double the single drive performance on the low numbers on ATTO. I suspect this is somewhat down to cache, but maybe also down to the drives supporting RAID but not AHCI. I don't expect to see spectacular gains under RAID0, but I think there will be gains. 60GB is just a little too small for an OS drive though. Currently I've got less than 30GB, but as I add apps and games etc it'll get very close to 60GB, and I'd rather have plenty of space.

As for performance degradation, I've noticed my write speeds had gone down from about 150MB/sec to around 110MB/sec after I had installed stuff to take it up to 40GB - I was really just filling the drive up. Repeating the same benchmark didn't decrease the write speeds, but that may have been because I hadn't written to the whole drive by that point. I probably have now. I've cleared stuff back down and I'll see if the performance stays the same or goes back up - and as I type ATTO is running, I'm capping out at 120MB/sec on writes, reads are unaffected.

A guy from PerfectDisk has been on the OCZ forums talking about defragging, and Tony has mentioned (but avoided any detail) some kind of OCZ software to enhance performance. It would seem to me that what we want is some kind of defragmentation/maintenance routine that erases any areas of the disk that the OS has marked as deleted. I think that'll come in the near future, either through third party defrag tools or through OCZ. Even with performance degraded though the results are better with the new firmware on an OS drive than they were with the old firmware on an empty non-OS drive.

[Ao1]

^^^^
Is the issue not how much valid data is showing but how full the drive is with "deleted" data? A 30/ 60GB drive is soon going to be "full" after an OS install along with a few basic programs due to the temporary files that are written in the process.

That is one issue. How the drives fragment is another. From what you are saying ATTO benchmarks are consistent on your Vertex, so it would seem that the Vertex is not affected as is the case with Intel drives, but I find that a little surprising.

EDIT: What happens if you run multiple Iometer or Crystal benchmarks thattake cache out of the equation? Do you get consistent results?

EDIT: Let's not forget that ATTO did not give a true reflection of the user experience performance of the Core's. It didn't give any indication that the drive could potentially stutter. That is the prime reason for me to believe that ATTO is useless for giving any form of validation on how an SSD is going to perform in real world usage.

EDIT: OCz's web site states: **Rated speeds are based on NVIDIA 680i chipset and may vary slightly dependent on the benchmark used, drivers, windows version, bios version and file size. We recommend using ATTO and PC Mark Vantage for benchmarking SSDs.

[One_Hertz]

I've also noticed that people running 2 or 3 Vertex drives on RAID seem to get better performance than just double the single drive performance on the low numbers on ATTO. I suspect this is somewhat down to cache

It is entirely due to cache. As soon as they turn cache off in ICHXR their small numbers go back to normal.

[s3rv3r]

Hi there, so, I was thinking, which one would be better, OCZ Vertex or OCZ Apex? According to the specs the Apex is better (Well, looks like), but I am seeing a lot of good results with the Vertex.

Thanks!

[One_Hertz]

vertex by a large margin.

[F@32]

vertex by a large margin.

True, stay away from JMicrons even in internal RAID config. Titan/Apex - same thing. It's either vertex or Intel/Samsung SLC/MLC at the moment.

[Griff805]

Still no pcmark vantage scores? Anyone? Single drive or RAID-

[Ao1]

You can get consistent results by following this order: set-up the drive -> bench ->write zeros -> set-up the drive -> bench. It's not possible otherwise.

QFT. Benchmarks kill the short/ midterm performance of SSD and if you don't follow the above you are wasting your time. This has now been well documented on the OCZ forum.

It makes a joke of the scaremongering PC Perspective review on the X25-M. I have since found out that the quote from Intel on my earlier post was not in response to the PC Perspective review it was information available at the release of their drives, which kind of makes me wonder where they were coming from and what they were trying to prove.

PC Perspective failed to demonstrate anything more than a drop in benchmark performance, which if they had done their homework is hardly a revelation. So is there a perceivable drop is user performance after fragmenting an SSD with benchmarks? Not on my X25-E, but considering that it is designed for server applications that is not really surprising. I doubt that the X-25-M or Vertex are affected either, however reading the PC Perspective review it seems that the benchmarks they undertook were instigated by this statement "Once the newness wore off and I got some hours logged with it, I discovered that my speedy SSD had lost some of its luster." That statement would seem to imply that they had noticed a drop in perceivable speed in normal use (within hours).....but who knows? If they had been talking about the X25-E in server applications they might have been on to something, but they were talking about the X25-M on a workstation. Considering the damage they must have done to Intel I'm surprised they haven't been more forthcoming with a more robust response.

Do benchmarks pointlessly wear out the SSD as well as give useless results if you don't follow the advice above? Yes, which is obviously more of an issue on MLC.

[Ao1]

Duplicated by error.....Is it just me or has anyone else noticed a severe slow down on this site since they went SSD last week?

[Halk]

Still no pcmark vantage scores? Anyone? Single drive or RAID-

Single 60GB



HDD : 20107

[Rhys]

QFT. Benchmarks kill the short/ midterm performance of SSD and if you don't follow the above you are wasting your time. This has now been well documented on the OCZ forum.

It makes a joke of the scaremongering PC Perspective review on the X25-M. I have since found out that the quote from Intel on my earlier post was not in response to the PC Perspective review it was information available at the release of their drives, which kind of makes me wonder where they were coming from and what they were trying to prove.

PC Perspective failed to demonstrate anything more than a drop in benchmark performance, which if they had done their homework is hardly a revelation. So is there a perceivable drop is user performance after fragmenting an SSD with benchmarks? Not on my X25-E, but considering that it is designed for server applications that is not really surprising. I doubt that the X-25-M or Vertex are affected either, however reading the PC Perspective review it seems that the benchmarks they undertook were instigated by this statement "Once the newness wore off and I got some hours logged with it, I discovered that my speedy SSD had lost some of its luster." That statement would seem to imply that they had noticed a drop in perceivable speed in normal use (within hours).....but who knows? If they had been talking about the X25-E in server applications they might have been on to something, but they were talking about the X25-M on a workstation. Considering the damage they must have done to Intel I'm surprised they haven't been more forthcoming with a more robust response.

Do benchmarks pointlessly wear out the SSD as well as give useless results if you don't follow the advice above? Yes, which is obviously more of an issue on MLC.

After doing multiple Lowish Level re-formats and swiss cheesing my drives until my benchmarks at one point were less than half of what they were, I noticed XP startup took 0.5 bars extra to load. There was no perceivable difference though once in OS.

[Ao1]

After doing multiple Lowish Level re-formats and swiss cheesing my drives until my benchmarks at one point were less than half of what they were, I noticed XP startup took 0.5 bars extra to load. There was no perceivable difference though once in OS.

So no need to lend you my Intel crutches?

[Ao1]

Let me clarify the point about the X25-E in server applications. The X25-E is sold for server applications. It's selling point is IOPS. In a server application an Iometer benchmark is reflecting server use, so it is relevant to server applications.

If an Iometer benchmark is causing a drop in IOPS that is a problem for Intel, but it is still unclear how the X25-E is performing long term in server applications.

Rhys, my choice to get an X25-E for a workstation was 1, because it is SLC, 2, because I didn't want to risk anything less than that top level performance having previously purchased a Core. I don't regret that decision one bit and would do it again given the chance. (My logic, flawed other otherwise.... 10 Ford's might cost the same as a Rolls Royce but you don't end up with a Rolls Royce with the money spent on 10 Ford's.)

Not to unpset anyone....I'm not implying the Vertex is a Ford, it wasn't available when I got the X25-E. That said I would still get the X25-E if I had the choice, but that is just my opinion.

[One_Hertz]

Single 60GB

HDD : 20107

A bit slow? Should be ~24k?

Edit: lol vantage is so screwy, just ran it and it gave me 1200mb/s for the media test. Seems to be influenced by cache.

[Halk]

I've got a live OS with games, and I've been benchmarking a fair bit
Currently I'm seeing if PerfectDisk and Diskeeper can sort things out. I'm noticing performance comes back after a reboot. I'm not sure if it's just the case that the reboot does it, or if "defrag" benefits come after a reboot.