ACARD RAM Disk 9010 series

[Serra]

Who the heck is talking about the one port version? It is pure garbage more or less. The fact remains, the proper version is 2x faster than the old IRAM.

I personally like the 1 port version, and for one simple reason: non-RAID drive size. The 1 port version offers 6 DDR2 slots (in the pictures anyway), and @ 2GB/stick density (best bang for your buck right now) 12GB vs 8GB is - to me - a huge difference in usability. Plus being cheaper is a bonus.

Would I buy it? With a faster controller - yes, undoubtedly. As it stands though the controller speeds just aren't enough for my excessive e-peen requirements which is why I'm eyeing a regular 2-port 9010 and 2x 4GB ECC sticks of DDR2-5300 (and hoping to all get out that the prices on those sticks drop dramatically as time goes on). We'll see whether I end up pulling the trigger or not, Nehalem-EP is right around the corner and I may need to see prices on motherboards for that first.

[Levish]

I personally like the 1 port version, and for one simple reason: non-RAID drive size. The 1 port version offers 6 DDR2 slots (in the pictures anyway), and @ 2GB/stick density (best bang for your buck right now) 12GB vs 8GB is - to me - a huge difference in usability. Plus being cheaper is a bonus.

Would I buy it? With a faster controller - yes, undoubtedly. As it stands though the controller speeds just aren't enough for my excessive e-peen requirements which is why I'm eyeing a regular 2-port 9010 and 2x 4GB ECC sticks of DDR2-5300 (and hoping to all get out that the prices on those sticks drop dramatically as time goes on). We'll see whether I end up pulling the trigger or not, Nehalem-EP is right around the corner and I may need to see prices on motherboards for that first.

probably because people would like to setup the two port model with the onboard raid0 for nearly double the throughput

[wmaciv]

Serra, you can set a jumper on the 9010 to allow all memory on one port.... of course at non-raid speed.

Wade

[josh1980]

I still love my 9010. I find it a very worthwhile investment. Although, I'm starting to get the feeling Acard released this product now, only because it wanted to recoup the development costs for this device. They showed the ANS-9010 for 2 years at various conventions, always giving us a new release date. I think they intended to tweak the hardware for more performance, but decided to release it to recoup costs before SSDs take over the market. I was intending to purchase a second box for another computer, but i have decided to hold off on account of SSD. SSD is dropping so fast in price, I expect I'll be able to find 32GB SSD drives with comparable performance benchmarks next year for <$200. The ANS-9010 with no ram is twice that price. I expect that next year I'll be buying a 128MB SSD drive with similar benchmarks to the ANS-9010 and I'll be using the ANS-9010 for temp files and swap file.

The only advantage I see in this hardware in 12 months is it has infinite durability(ie. it can be written to indefinitely without degrading).

If this box had been released 2 years ago, it might have set the stage for a performance revolution that will now be seen with SSD becoming mainstream. The seek times of SSD/RAM makes all the difference. I installed Windows XP on my box using the 1 port setup vice 2 port, and I could not measure a difference in boot times.

Those of you looking at the 1 port and 2 port versions, the big difference(IMO) is only the number of RAM slots. Benchmarks don't tell a story that reflected real world application. Don't let the 'lower' performance numbers fool you. If I had to buy all over again, I'd look at the number of ram slots of each, not the lack of dual port support. You will ultimately be limited in size by the number of RAM slots. I think that reason for there being little real world performance, is that there are very few sources of data at these speeds to begin with. If you copy files over gigabit LAN, you are limited by LAN, if RAID0 the hard drive seek/throughput, if CD-ROM/DVD-ROM then seek/throughput...etc(I think I made the point).

I'm thinking that when 32GB CF disks are cheaper, I'll shut down my computer and let it do a backup to CF, then make a duplicate of the CF to another CF. If I kill the OS, I'll always have the "backup CF" to just plug and play.

[Levish]

Josh, did you get a chance to check latency on it? (preferrably something that measures to the hundreths or thousands)

[josh1980]

I didn't. The latency with h2bench was hundredths of a ms(0.05ms or so)... I haven't found a program that gives quality numbers for hundredths of a ms(or better). If you know of one, I'd be more than happy to run it. I tried searching for programs that give latency in microseconds, but I couldn't find any.

[Serra]

Serra, you can set a jumper on the 9010 to allow all memory on one port.... of course at non-raid speed.

Wade

Awesome!

The 9010B is still a better deal on a slot-per-dollar basis tho (which may/may not be offset by the lack of RAID speeds, depending on setup of course).

[Spoiler]

I'm thinking back to several discussions we all had
when the Gigabyte i-RAM Box was released with a SATA-I
interface:

Many of us were saying that "RAM should SATURATE the SATA interface".

In other words, using any RAM regardless of age (DDR, DDR2, DDR3)
should automatically increase the transfer speeds in a manner
proportional to the interface speed e.g. from 150 to 300 MB/second.

Then, along those same lines, we expected that
running 2 such devices in RAID 0 should nearly
double throughput.

What was missing from those discussions, imho,
was a realization that each end of a SATA cable
is controlled by firmware, software and harware logic
which varies considerably in efficiency, but we haven't
developed very good tools to isolate measurements
of that particular effiency.

Just to illustrate with a very simple example,
we scaled an inexpensive 16GB Super Talent SSD
from 1 to 2 drives, the latter in RAID 0, and
the "raw reads" increased from 130 to 150MB/second --
not even close to a linear scaling result.

The controller is the Highpoint RocketRAID 2340
using x8 PCI-Express lanes in an ASUS P5W64 WS Professional.

And, using these SSDs eliminated seek times and rotational
latencies inherent in conventional rotating disk drives.

So, what is the cause of the "penalty" that prevented linear
scaling from 130 to 260 MB/second with that simple configuration?

If RAM is truly "random access" -- as it should be --
the quantity of RAM present should have no significant
effect on the transfer rates in either direction (read or write).

What needs to happen, in my professional opinion,
is a 4-port SATA/6G ramdrive that is designed -- in advance --
with logic at both ends of the SATA cables that scales efficiently
i.e. in near-linear fashion.

I'm not asking for 100% perfection here, OK?

However, there are plenty of measurements already
reported on the Internet of 4 x i-RAMs in RAID 0
(e.g. see youtube.com for video illustrations).

I think it is reasonable to accept 10-20% raw overhead
from 4 such ramdisks wired to Intel's ICH10R, provided
that the logic overhead inside the ramdisks is not any larger
than that.

Thus, using Intel's latest single SATA/3G SSD,
250/300 = 83.3% or ~17% overhead.

Allowing that overhead to reach 20%, we should be able to
design and manufacture controllers at each end of the SATA cables
that achieve the following scaling with SATA/3G interfaces:

300 MB/second x 4 devices x 0.80 efficiency = 960 MB/second

And, we should also expect linear scaling with the arrival of SATA/6G:

960 x 2 = 1,920 MB/second

Anything less than these target rates would indicate
controller logic with efficiencies that are unnecessarily
inferior.


Sincerely yours,
/s/ Paul Andrew Mitchell, Inventor and
Systems Development Consultant

All Rights Reserved without Prejudice

That was a lot of writing just to say that certain controllers are inefficient. I do agree with you though. Why produce a device that can can take advantage of such high bandwidth such as DDR2, then limit it through crappy interface design?

[NapalmV5]

id be all over this ram disk if acard would come up with a pcie controller of their own

similar to pcie ssd.. but better/faster/flexible by sporting multi ram disk ports

can you guys see 8x ram disks connected to a full x16 pcie2 fledged controller ?

i certainly can.. up to acard to bring it

[nFo]

i-rams dont work with areca cards

Works fine

[wmaciv]

Here is an odd one.... I formerly had the 24GB of DDR2 backed up to a Transcend 32GB 133x CF card with no problem. Now, after installing the last 8GB of DDR2 for a total of 32GB, the "Backup to CF" LED illuminates solid red, which, according to the manual indicates that the CF card is damaged, or that the capacity is not equal to or greater than necessary to back up the on board RAM. After talking to ACARD tech support, they seem to think that since the overall RAM size shows as 28.4 GB out of 32 physical (ECC functionality stealing 1/9th of total available RAM), that a 32 GB flash is enough to backup a 32GB bank of onboard RAM. I have a nasty suspicion that I will have to jump to a 48gb card to b/u a 32 gb drive. I have my questions in with ACARD right now... will see how this goes.

Wade

[Speederlander]

Here is an odd one.... I formerly had the 24GB of DDR2 backed up to a Transcend 32GB 133x CF card with no problem. Now, after installing the last 8GB of DDR2 for a total of 32GB, the "Backup to CF" LED illuminates solid red, which, according to the manual indicates that the CF card is damaged, or that the capacity is not equal to or greater than necessary to back up the on board RAM. After talking to ACARD tech support, they seem to think that since the overall RAM size shows as 28.4 GB out of 32 physical (ECC functionality stealing 1/9th of total available RAM), that a 32 GB flash is enough to backup a 32GB bank of onboard RAM. I have a nasty suspicion that I will have to jump to a 48gb card to b/u a 32 gb drive. I have my questions in with ACARD right now... will see how this goes.

Wade

Just get acronis and back it up to a hard drive.

[wmaciv]

Yeah, I love that program... so much better than Ghost... I bought a copy of Acronis TrueImage 9 a while back, and use it quite often, especially now with the 9010. I think everyone is thinking along the same lines with this product - really fast and nice, but NAND flash drives sre so quickly overcomming their limitations, in less than a year, this device will seem quaint.

Wade

[josh1980]

About the long comment above about the inefficiencies in the controllers. There's alot of variables involved here:

1. CPU - faster transfers do use more CPU. If your CPU is loaded, I'm sure that affects performance.
2. Drivers - Are the drivers optimized well enough to give you actuall access to the full bandwidth allocated to SATA?
3. OS - I know that all Windows OSes to date do not write data in 4k blocks. Writing data in 4k blocks for system writes is very important for SSD performance, which is why Windows 7 has it built in. MacOS has been using 4k blocks for system writes for quite some time. (this is off the top of my head, but I could be mistaken, got this from somewhere where they compared OSes to SSD to see if the OS affected performance)
4. SATA has 8 bit to 10 bit encoding, resulting in an immediate loss of 20% of your possible bandwidth. If someone understands the encoding and that it does not cause an immediate 20% decrease in performance, then I stand corrected. I believe this fact alone is why the i-ram performance wasn't 150MB/sec, but was limited by just a little over 20%.
5. Some motherboards gave the SATA controller(read: all SATA devices, ie 4, 6, 8 etc) a 1xpcie connection. That's an instant bottleneck of 250MB/sec for all SATA devices combined. Who knows if the northbridge/southbridge bottlenecks any further based on the design. This is why I have a corei7 . WHen I was shopping for a motherboard for my new computer, this was a factor in my choices. Sure, using a 1x pcie wasn't a limitation 2+ years ago, but it will be in the future.
6. I'd bet money that the RAM is operated at DDR2-400 speeds with a CAS6 latency. Since that's the slowest RAM according to the JEDEC standard, everything should be able to perform at those speeds.
7. Possibly some other limitation none of us fully understand.

Remember data from the ramdrive to the CPU has to go through many parts to get to the end result. Yes. I think the ans-9010/b should be able to perform alot better, even in single port operations, but who really knows where the limitation really is? Or is it simply an exponentially larger problem as data goes from beginning to end to eventually get to the CPU, like someone throwing a small rock over a mountain that causes an avalanche?

Just as an analogy, how many of you have benchmarked your RAM transfer rates and gotten anything close to the theoretical maximum for your computer? I don't know that I've ever hit 80% of the theoretical maximum. I was always so disappointed by how far off the theoretical maximum was from actual that I never looked at the numbers after the first or second time.

I got quite a few private messages as to which RAM I have in my box. The following links are all of the RAM that I have tested personally in the box. I mixed and matches sizes as well as brands in all sorts of random orders, and I never had a problem with compatibility. Choose name brand and do not buy high density DIMMs and you should be fine. Having alot of RAM from old computers helped me build this list.

http://www.newegg.com/Product/Produc...82E16820231181

http://www.newegg.com/Product/Produc...82E16820134863

http://www.newegg.com/Product/Produc...82E16820227334

http://www.newegg.com/Product/Produc...82E16820220269

http://www.newegg.com/Product/Produc...82E16820146565

[josh1980]

5. Some motherboards gave the SATA controller(read: all SATA devices, ie 4, 6, 8 etc) a 1xpcie connection. That's an instant bottleneck of 250MB/sec for all SATA devices combined.

Found a good example of this...

I went and looked at my motherboard(Gigabyte GA-EX58-UD5) block diagram more closely. The ICH10R chipset uses the QPI Interface. However the Gigabyte SATA2 chipset has 4 SATA 3Gb/s plugs and has 1x PCI Express lane for data transfers. For me, that means I shouldn't ever use more than 1 SSD drive on all 4 of those SATA plugs. Even then, 1 SSD could saturate the 1x pci-express lane. If I had set up the ANS-9010 I own on the Gigabyte SATA2 instead of the Intel ICH10R, I wouldn't have gotten the benchmarks I have shown here.

Remember, this is a motherboard that was released less than 60 days ago. This is the stuff I look at when I build a computer. The gigabyte SATA2 would be perfect for using CD-ROMs and other lower performing devices. Add to this the fact that some 3rd party chipsets aren't bootable. Fortunately mine is though.

Just goes to show how deeply involved effective planning for a powerful computer needs to go.

Some people think i'm some kind of maniac for looking into this much detail. I call it being thorough.

[Buckeye]

Thanks Josh for all the info

4gig RAM's is not cheap. The ones you posted come to $767.92 for a full load

[Speederlander]

Thanks Josh for all the info

4gig RAM's is not cheap. The ones you posted come to $767.92 for a full load

Yeah, pretty expensive. Add in $400 for a unit and that's a ton of cash for 32GB (actually 29.7GB in windows).

[wmaciv]

Ok, this is getting frustrating. Originally, for the first two weeks, I had 6 x 4GB sticks in a RAID0 config, with a (recommended) Transcend 133x 32GB CF card, and everything working as advertised. When my last 8GB arrived, I backed up via Acronis to a spare HD, powered down, and removed the CF card. I opened the ANS-9010, disconnected the battery backup pigtail, installed the new memory (all memory from same manufacturer; all mached sets). After reversing the whole process, rebuilding the RAID0 in Bios, and cloning the OS partitin back to the ACARD, I tried reinstalling the CF card. Now I get a steady RED LED, which according to the manual menas either the CF card is corrupt, or it is too small to backup the amount of DRAM on board. Here is where all of this begins to break down. ACARD tech support swears that a 32GB CF is enough to back up a 32GB mem install, especially with the ECC function enabled. OK, i'll buy that. Second, the card formats flawlessly to FAT32 and NTFS, and reads and writes effortlessly through my USB2.0 card reader and formats to full capacity (no problem here). Finally, I went back, removed the "newer" 8GB mem set, reimaged the drive once again, and I still get the RED LED. WTH? I have run out of ideas. The crazy thing is, it worked like a champ for the first run install. I noticed the rev. 3.3 MB has a wider DIMM compatibility list; could PCB revision play a part in this drama? Any ideas from you smart folks? 32GB CF is just too spendy to be playing these games with. Early adopter penalty?

Wade McIntyre

[wmaciv]

Josh1980, any experimentation with cluster size on your end? I used to fool with that on the Platypus Qikdrive8, and still not sure of the right answer. I have seen and a review where the smaller the cluster size, the greater the CPU utillization, but the deduction of the reviewer was that the increased CPU utilizaiton came from keeping the CPU busy with near instantaneous data feed. Not sure about that logic, but the graph seemed to give some merit to the idea. Overall trnasfer rates dropped precipitously, though. Any thoughts?

Wade

[Speederlander]

Honestly, why did you even bother with the 32GB card? Skip it. Acronis the base install and forget about it. It's faster anyway.

[josh1980]

It's funny how many of you went WOW at the price for 8x4GB sticks. That's what I have. Expensive, but I have the cash to experiment. I've always lived on the bleeding edge. I think they call it bleeding edge because it bleeds your wallet fast. I had a dual processor Xeon workstation years ago, before anyone could buy a dual core computer. I had the latest and greatest at 800Mhz FSB. The system cost me almost $5k for everything.

Yesterday I bought 2xANS-9010B and 12x2GB sticks to populate those with. I bought the RAM drives from 2san, and they have a christmas special where you buy 2 ANS-9010Bs and get a free ANS-9012... So now I'm looking at 16GB SD cards... hehe. I bought 12x2GB sticks from newegg for $14 each. These ANS-9010Bs are perfect for running a virtual OS using VMWare Workstation. Fast seek and transfer rates to help offset the performance penalty for running 2 OSes.

@wmaciv-

I have questioned the usefulness of CF for my purpose since day 1. I still don't have a CF card, so I can't really provide much troubleshooting help. What I would recommend you try to do is zero out all of the sectors on the drive. This would wipe out all data on it, and possible some location that the ANS-9010 uses to keep data stored for it's purpose.

My thoughts are that the CF is actually a backup of the total userspace on the RAM. If you have 32GB of non-ecc ram, then that would be 8/9th(ECC enabled). So you are really backing up about 29GB of data. I'd bet there's something on the CF card that the ANS-9010 uses to identify how much data is backed up. You have more than you had, so the box doesn't want to use your card because it's not the right size. Formatting likely won't erase those bits since formatting doesn't really write data. If writing zeros to all sectors doesn't fix it, I don't have any other ideas to try, except RMA your CF stick.

I use acronis and do an autobackup every night. In the event I lose power and my UPS doesn't last.. acronis is cheaper and probably faster to recover my data.

I did experiment with stripe size for RAID0... look back at my post for benchmarks in RAID0 mode. 16kb was optimum for my setup using an ICH10R chipset. I left the sector size at 4kb though. I am formatted for NTFS running Windows XP Pro. I didn't see much of a reason to optimize my drive much further because when i run programs and such, the hard drive is not my limitation. My corei7 goes to max CPU usage for the associated core when programs are loading. Using HDDLED, I rarely exceed 50MB/sec when doing anything on my computer(aside from copying files). The only exception is playing World of Warcraft. WoW loads up all those graphics and such, and that goes ALOT faster, and the transfer rates go to about 80MB/sec for 2 seconds.

[Serra]

It's funny how many of you went WOW at the price for 8x4GB sticks. That's what I have.

Yeah, it's funny when we have problems with the rent too. We say WOW because the percentage of people that can afford this level of hardware are not just slightly off the tip of the bell curve (I know you didn't mean it like that, but I think this is a fair retort).

Seriously though, let us know how that 9012 works out. It's certainly an interesting little device, but just doing some numbers off the top of my head based on the last time I looked up SD speeds it seems like an investment for particularly specialized areas. My hope is that you've looked into it a bit more than I have (which isn't hard, I haven't looked up max speeds on SD cards in quite awhile) and can wow us with some interesting speeds at a decent GB size.

[nFo]

I got the Acard ANS-9010 yesterday.
With 1mb to 8mb blocksize you get uber write results.

[josh1980]

I tested my ANS-9012 tonight with 1x8GB Sandisk Ultra II SD card. I used the SD card from my HD camcorder for these tests. The SD card says Class 4. For any of you familiar with the SD classes, Class 4 requires 4MB/sec. I have no idea why Sandisk would market a product as class 4 when it's performance is 15MB/sec... That's almost 3x faster than Class 6(6MB/sec). Anyway, the Sandisk performed a solid 17.5MB/sec across the board. This was using a SATA->USB dongle!

Now I just have to figure out what size/speed/brand to buy. I was thinking if I buy a brand of SD with lifetime warranty, that's a guaranteed size forever. Just have to deal with RMAs and such. This does prove that 1 SD card will operate this device, for anyone that was wondering. That allows someone to add SD cards as time goes on. That's actually a pretty good deal for us folks that want to buy 32GB cards, but want to buy 1 or 2 to start, and the rest when they are cheaper.

One of my ANS-9010B arrived broken. One of the RAM clips was broken, so I'll be testing just one in the next day or 2, and then 2 ANS-9010B in RAID0 as soon as I get the other ANS-9010B in the mail. I just might try doing a RAID0x4 using both ANS-9010B and both ports on my ANS-9010. Just depends on how silly I want to be next weekend.

[wmaciv]

Was having the devil's own time with the 9010 getting a corrupt volume on Port 1 (secondary port). Tried all sorts of RAM configs, still it would soon error on me either while running or between boots. Decided to get a little crazy, and put a jumper on the "RESERVED" block, a.k.a the ECC funtion disable, and I have had no problems since. I plan to leave the ECC functionality disabled, not for usable RAM, but it truly seems that it causes problems, at least in my RAID0 setup. I have a feeling there is something happening in the ECC that my motherboard's implementation of the ICH9R does not like, or vice-versa. Still working on getting another CF card; a full "erase" vs. just a format did not fix that problem either. Wonder if HW rev. 3.3 has these problems?

Wade