ACARD RAM Disk 9010 series

[wmaciv]

I should have captured this in the previous post, but here goes...
When I remove the ECC function, and effectively enable "all" the memory, no matter what combination of RAM I use in what DIMM slot, Port1 ALWAYS reports itself as 15GB even though Port0 shows the proper 16GB. And remeber, this is prior to any formatting, or even building a RAID set in BIOS. Is the ACARD bootstrapping some sort of real-time kernal or OS and using a portion of RAM at boot to run its protocols? Even one stick of 4GB ram in any given slot (Port0 or Port1), has 1GB take from it. Now, I did NOT try to see if it steals 1GB from the top when the 9010 is set to single port mode, i.e. all the RAM in one contiguous block accesable through Port0. Maybe it would hand over all the RAM then. Jason, any thoughts?

Wade

[josh1980]

My guess is this. Keep in mind that the 'disable ECC' function is not really supported by Acard for some reason. It's 'reserved', but that was it's original intended function. I have a PDF of the original version that shows that jumper's function.

It sounds like the RAID0 function always accounts for the ECC function being enabled, and when it's not you still can't really use that space for RAID0. I'm guessing that when you use the RAID0 function it uses the equation:

Port1 RAM allocated = Total RAM/2 - Total RAM * 1/9
Port0 RAM allocated = Total RAM - Port1 RAM allocated

This would give the exact results you are seeing. It seems that jumper disables the ECC function(maybe it really doesnt???, but I'll talk about that in a sec) but the Port1 mappings equation still subtracts the ECC memory size.

Ok, now to elaborate on the ECC function idea. What happens if you partition that extra 1/9th memory and actually put data there. I'm wondering if the "ECC disable feature" isn't really completely disabling the function. Maybe the 9010 gets confused and says the RAM is free, but is actually still using the ECC function(and using that space). If this were actually the case and you format that space and fill the space with data, then your drive would suddenly be corrupt because you overwrote the ECC data that was actually there with your data. Weird idea huh? I'll have to try this out and see what happens sometime.

I would strongly recommend you not disable ECC. The ECC feature, at least IMO, doesn't seem possible to be 'incompatible' with your RAID controller. The ECC feature is purely handled by the ANS-9010. All hard drives made in the last 25+ years(if not from day one) have built in ECC. It's handled entirely at the hard drive level. If it is not working correctly with the ECC feature enabled, I would start pulling the RAM out and testing it in a computer as system memory to determine if there is 1 or more bad sticks. Maybe your RAM isn't compatible with the 9010 or the 9010 is bad. This sounds more like a problem with the RAM/9010 than an incompatibility between the motherboard and 9010. All hard drives end up correcting data by the ECC pretty regularly. I was surprised at how often the ECC is actually used even though the sector isn't considered 'bad'. ECC can only improve the reliability of the data, and disabling it could result in corrupt data(not to mention possible undesired results due to the function being 'reserved').

I got my ANS-9010b's set up. One is going to be RMAd for a broken RAM connector. The RAM will stay in the slot on it's own, but since I haven't used the box yet, why not fix it now while I can RMA it? The 9010b internals look exactly like the 9010, minus the added RAM slots, SATA port, etc. They use the same physical board for both. I'm wondering what would happen if someone soldered on the missing components. Would a 9010b become a 9010?

Can you elaborate on the 1GB missing when using a 4GB RAM stick? You said "Even one stick of 4GB ram in any given slot (Port0 or Port1), has 1GB take from it. " There's more than 2 ports for 4GB ram sticks to go :P. Just wanting to understand your comment on the "missing 1GB". What exactly happens with ECC enabled and disabled with 1x4Gb stick of RAM? If possible what about if you are in RAID0 or 1 port mode.

[wmaciv]

I still am convinced that the error on volume one (port 1) shown at boot time was COMPLETELY eliminated by jumpering the ECC function/RESERVED jumper. I'm not saying WHY it worked, but it fixed my problem, and I have been running continuously for about 5 days with no problem. I could barely make it a dozen boot cycle before throwing an error before.

The comment about the single stick of RAM was just that... Any of my 4gb sticks (out of 8) show only 3gb when installed in any slot (as the only stick of ram). Whatever slice it is taking from the pool of RAM, it takes it from 4gb or 32 gb or any amount in between.... I'm not sure, but it would seem if I put in a 2gb stick, it would probably show up as 1gb. Anyway... still learning the box here... still like it, but it has its quirks.

I still have no joy with my 32gb CF card that worked so well early on with 24gb, but now, no matter what erase or format tool I use, notheing but a red LED. One of these days I will get another card, but not right now...

Wade

[Chosen.]

I just finished a video for my review and I can't publish it on Youtube, as it's 12:30 long. It's a video of a fresh Windows Vista Ultimate 64-bit installation on the ANS-9010.
The installation took under 8 minutes, but with the 2.5 minutes of the system performance benchmark and some minor fiddling around in the desktop and a couple of restarts, the video turned out to be too long for Youtube...
I'm uploading it to GoogleVideo as we speak =]

[F.E.A.R.]

Don´t use reg ECC!!!
Only ECC or Non-ECC.

[josh1980]

What happened? The fact that you used 3 exclamation points makes me wonder if you had a beautiful fireball occur when you applied power or something. If you look at posts 119 and 121(both from me), you'd see that the box is not compatible with registered memory.

After reading about how registered memory works, I figured out that it is impossible for registered memory to operate in this box. The signals for registered are different. Normally, any signal on the input pins is directly placed on the output pins. With registered memory this is different. Any signal on the input pins is not on the output pins. The inputs go to the registers on the RAM stick. During the following clock cycle, the signal will be duplicated on the output pins. Basically, if a given device works with 'unregistered' memory(the quotes are there because there really is no such thing as unregistered RAM, the correct term is asynchronous RAM), then the fact that the device requires the output signal to be present on pins on the same clock cycle means that the 1 clock cycle added to registered memory would throw off the hardware, and would not work. .

Unfortunately Acard's manual for this is effectively crap. If you read my comment in post 119 of this thread, you'll see how the manual at their website conflicts with the manual that was provided with the hardware. If I had bought registered memory and had to RMA it, I would have called Acard and demanded a refund for my restocking fee for the RAM because they would clearly be at fault for providing incorrect information in the manual. I totally would have gotten it too, even if it meant getting a lawyer. I see no reason why the manual could be so horribly wrong with identifying the correct RAM to use. That's like selling a vehicle, and the manual tells you that the vehicle uses Unleaded gasoline, but it really requires Diesel.

[Particle]

No, it doesn't catch on fire with registered memory. I tried it. It wasn't until I found the fine print in the manual that I understood why eight 2GB ECC Reg DIMMs weren't working in the thing. It's not very common to have ECC memory that isn't registered, so it hadn't..."registered" (heh?) with me that reg wouldn't work after reading it had ECC memory support.

[Chosen.]

My review is ready

[Serra]

My review posted here.

[josh1980]

Strangely, I must have 'rev1' of the box, because my 'lid' on my ANS-9010 is a solid piece of metal. There's no holes in it at all. What would be great is a fan for these things. They get kind of warm, so I left my cover off.

Serra,

Nice review. I just have one comment. I'd research more before you take my word for this...I tried to look this up quickly before work, but i couldn't get all the info to verify this for myself. Based on what you can research about ECC, your review might be inaccurate as to the ECC calculations being performed on the RAM stick itself .

If i'm not mistaken, ECC is 72 bit RAM instead of 64 bit.

- For non-ECC RAM, 64 bits of data are sent via the RAM to the motherboard.
- For ECC RAM, 72-bits of data are sent via the RAM to the motherboard. The 8 additional bits are handed off to the memory controller, where the ECC checking is performed.

If you think about it, most ram are even number of chips(2,4,8,etc). You'll see an extra 'missing' chip slot on alot of RAM. The missing slot is because the PCB is made to support anything. The ram manufacturer solders on the number of chips and size for the amount of ram they want, and if the stick is intended to be ECC, the extra chip is added. In my example above, the chips would have been 8 bit, with the 9th chip being identical to the rest providing the 72 bits of data.

Several people have noted using ECC and non-ECC RAM(with the ECC feature enabled) and could not find a performance difference. The answer is simple. The board has to perform the ECC calculation regardless of the type of RAM used. The real difference is that ECC ram has 1/9th more RAM on the stick, but is not user available space because those memory locations are designated for ECC checking. So really, if you buy a 2GB stick of ECC RAM, it's technically 2.25GB, but your available space is 2GB.

I believe(of course, i'm not the designer so I have no information to prove this) that the ANS-9010 detects the total quantity of RAM available(this includes the ECC space, so 2x2GB ECC sticks would really be detected as 4.5GB). But since the box uses ECC we'd still see the 4GB.

This could have changed. I looked at pictures on newegg and stuff, and they were showing pictures of ECC ram with only 8 chips, but it could be they were just using a canned picture for all memory made by the same manufacturer. I'm not sure if its changed, that's why I think you should research it further.

Also, if what I said above is true, then the implications for using the pseudo ECC feature are more pronounced. This means that any RAM request would actually be 2 requests. 1 for the data, and a separate request for the additional ECC data. This would effectively mean that the RAM's theoretical max is 1/2 of the speeds. I'm sure they aren't using RAM at the DDR2-400 speed, probably more like DDR-200 or something along those lines. I believe the Gigabyte I-RAM was DDR-83 or DDR-100... It was underclocked whatever it was.

[m^2]

"What I find most encouraging is just how close we get to the 200MB/s maximum per-port throughput"
Sounds like you mean SATA2 maximum, which is 300 MB/s.

But overall, a very nice review. Especially the "Under the hood" part.

[F.E.A.R.]

I believe the Gigabyte I-RAM was DDR-83 or DDR-100... It was underclocked whatever it was.

I think, iRam uses PCI 2.0 - 33MHz...
My bandwidth with iRam is max. 133MB/s.

[Serra]

"What I find most encouraging is just how close we get to the 200MB/s maximum per-port throughput"
Sounds like you mean SATA2 maximum, which is 300 MB/s.

But overall, a very nice review. Especially the "Under the hood" part.

Nah, I was talking about the processor max... but I can see how that would easily be missed. Maybe I'll have to take a swing at re-wording it.

[lowfat]

I think, iRam uses PCI 2.0 - 33MHz...
My bandwidth with iRam is max. 133MB/s.

The iRam doesn't use PCI for bandwidth, just to power the card. All the data is over SATA1.

[F.E.A.R.]

@ Chosen

In cooperation with a french forum (Nokytech.net) we start a collective order for Acard 9010 (365$ of piece - normaly 532$)
But it´s hard to get cheap 4GB-Modules in germany (less than 64$ of piece). Qimonda (Aeneon) are broken!?!?
Now, i bought 16x 2GB Kingston-Modules (320$) and wait for 2x Acard 9010.
I hope to break the 700MB/s.-Wall @ 0,02ms with the ARC-1261D-ML


@ lowfat

You´r right
PCI is only for electricity supply

[Chosen.]

I think, iRam uses PCI 2.0 - 33MHz...
My bandwidth with iRam is max. 133MB/s.

Your 133Mb/sec are basically SATA150's theoritical bandwidth (150Mb/sec) minus a 10% overhead.

@ Chosen
In cooperation with a french forum (Nokytech.net) we start a collective order for Acard 9010 (365$ of piece - normaly 532$)
But it´s hard to get cheap 4GB-Modules in germany (less than 64$ of piece). Qimonda (Aeneon) are broken!?!?
Now, i bought 16x 2GB Kingston-Modules (320$) and wait for 2x Acard 9010.
I hope to break the 700MB/s.-Wall @ 0,02ms with the ARC-1261D-ML

Very nice, do you use Live Messenger by any chance? If so, PM me your MSN address

The cheapest 4Gb modules I found, were some Kingston N2E6/4G for £79 so I didn't fill up my ANS-9010...

[F.E.A.R.]

The cheapest 4Gb modules I found, were some Kingston N2E6/4G for £79 so I didn't fill up my ANS-9010...

Yes... Infineon/Qimonda/Aeneon have big problems


So i will buy 2x Acard 9010 to get this --> (it´s only onboard - ICH9R) <-- 2 Acard´s


With a separate controller you get better 4k-benches (1x Acard 9010 with LaCie-Controler) <-- 1 Acard


And no, i haven´t a messenger. But you can PN me

[Boogerlad]

why is a raid controller worse than the onboard one for seq?

[F.E.A.R.]

It´s only one Acard on a raid controller. Onboard 2
I red-marked it only for you

A good raid-controler + 2 Acard´s must be the hell on earth

[wmaciv]

First e-mail....

Hi,

1. For ANS-9010 8*4GB total 32GB,

Memory total capacity: unit:1024Byte
32GB=32*1024B*1024B*1024B=34,359,738,368Byte

ANS-9010 has ECC simulation function and it needed total capacity of 1/9,

so it will be 34,359,738,368 *1/9=3,817,748,708Byte

According to calculate, we can use data of RAM Disk is

RAM DISK - ECC = DATA
34,359,738,368-3,817,748,708=30,541,989,660Byte

CF Card capacity: unit:1000Byte
32GB=32*1000*1000*1000=32,000,000,000Byte

Therefore, we know RAM Disk capacity is less than CF card capacity, so you can feel free and use 32GB CF card to backup / restore data with ANS-9010.

If you are use DIMM with ECC for ANS-9010, we will suggestion move up to next size of CF card. For example, if total RAM Disk capacity is 32GB, please use 48GB CF card for backup / restore.

2. For that button, it is for engineer debug use only, please do not press that button in any time, because it may take risk or cause data lose.


ACARD Technical Support Dept.
FAE-Jack
Mail: support@acard.com
Tel: 886(2)85122290
----- Original Message -----
From: supportrc
Sent: Saturday, December 27, 2008 8:43 AM
Subject: <<9700080531>>Help!!--English--ANS-9010


Second e-mail...

Hi,

1. No need do format on CF card before you insert it into ANS-9010 and backup data.

2. The new CF card will copy all ANS-9010 data (included disk partition) if you pressed backup to cf button or auto backup data to cf when power off ANS-9010.

3. When restore data in progressing, CF be removed suddenly, you may see "Backup to CF" LED solid red. 2 solutions for this situation:
A. You can try to remove all power source from ANS-9010 (include battery) and install power source again.
B. Insert CF card to ANS-9010 again.

If you read previous mail, you will find out uses 32GB CF card to backup / restore data from ANS-9010 with 32GB DDR2 DIMM (NON-ECC) should be no problem.

ACARD Technical Support Dept.
FAE-Jack
Mail: support@acard.com
Tel: 886(2)85122290
----- Original Message -----
From: supportrc
Sent: Monday, December 29, 2008 8:41 AM
Subject: <<9700080662>>Re: Help!!--English--ANS-9010


Third e-mail...

Hi,

If your DDR2 is non-ECC and disable ANS-9010 ECC function (jumper on ECC), too, your ANS-9010 total size will be 32,768MB, but CF card is 32,000MB only, that's why you can not backup and see Backup to CF LED are steady red.

ANS-9010 are provided ECC function (no jumper on ECC) already, it makes date more safety and security, this function take total size of memory * 1/9, meaning 32768-(32768*1/9)=29127, and CF card can fully backup ANS-9010 32GB without any error because CF card size big than ANS-9010 now.

If your DDR2 DIMM is ECC type, ANS-9010 total size will be 32,768MB, but your CF card is 32,000MB only, that's why you can not backup and see red light on Backup to CF LED, so please use 48GB CF with ANS-9010 32GB or downgrade ANS-9010 to 28GB with 32GB CF card for complete backup.

Please beware that power source (include battery) must remove or press button battery connector nearby when you are install DDR2 DIMM.

Please provide your DDR2 DIMM model and ANS-9010 firmware version to us for check more information.
Here is how to check ANS-9010 firmware version:
1. Right click on "My Computer"
2. Click "Properities"
3. Check ANS-9010 on "Device Manager"
4. Refer to attach file "9010-FW 31 32.jpg"


ACARD Technical Support Dept.
FAE-Jack
Mail: support@acard.com
Tel: 886(2)85122290
----- Original Message -----
From: supportrc
Sent: Monday, February 09, 2009 8:51 AM
Subject: <<9800004593>>Help!!--English--ANS-9010 R 3.1

[F.E.A.R.]

I´m ready for the fight!!!
16x Kingston ValueRAM DIMM 2GB PC2-6400U CL6 (DDR2-800) (KVR800D2N6/2G)

[josh1980]

So.. today I hooked up my ANS-9010(using both ports) and my 2 ANS-9010B in RAID0. The speeds were....wait for it... 702MB/sec on HD Tach. I couldn't tell the difference in speed between this setup and my 1 ANS-9010 using both ports. It booted in about the same time, copying was a lot faster, that was about it. I'd post the benchmark results, but I deleted them on accident . Grr.

[F.E.A.R.]

Which controller?

Look this test with ICH9R, Areca ARC-1210 and ARC-1231ML (german)

http://www.hyperossystems.de/Erfahru...perdrive_5.pdf

[josh1980]

*Update*

Well, I now own a 9010 and 2x9010B models. I decided to buy my first CF card for one of my 9010B drives since it was going to be the boot drive for my server. Got a Kingston 16GB card (only have 12GB of RAM in it). Plugged it in, and the backup light turned green, hinting that all is well. That was a mistake assuming all was well. I decided to move my server into another room. After getting it there, I looked and the light was blinking and all seemed right. However things were about to go wrong.

Due to a busy day, I decided not attempt to plug the computer back in and boot it up. I woke up this morning and noticed that all 4 battery lights were still lit. I was shocked as the battery isn't supposed to last 12 hours. When I got home from work, again all 4 lights were still lit. I just plugged in the computer and turned it on. Surprise! There's no data on the drive. The backup light is red. Not happy right now as I didn't make an Acronis image because it all seemed okay until the next morning when I woke up to find 4 lights lit.

There is a firmware update for the ANS-9010B (of course, the website says it's 'Beta' and to not install it until it has been thoroughly tested). There is also a software tool for download. Again, with a catch. The drive cannot have any partitions on it per the instructions with the software tool. Note the underline. It tells you to backup the data (obviously as a warning), but it also tells you that you cannot have a partition on the drive. The software tool allows you to test the RAM to verify compatibility. I see no ability to test the functionality of the CF card, so my situation is still not solvable.

Currently, I'm stuck with having to reload Windows Server 2003, and now I have a CF card that I am unable to prove will work in a 9010B. Guess for me I'm lucky enough to have a spare 9010B, so I'll be duplicating data from 1 to the other to determine what the heck is wrong.

My advice: Be sure to back up your data, and not trust the CF feature. Someone else in this thread had a similar issue where their CF just wouldn't work. I'd expect that if the backup failed then I'd see the red backup light BEFORE I turn the computer back on, and I'd see the battery level drain slowly to zero while the battery powered the RAM. It appears that what happens when power is lost is the CF backup starts.

If the CF backup fails for any reason then the battery just gives up completely, dumps the contents of your RAM and sits there with 4 power lights on. Not the kind of design I'd call 'high reliability'. I'd expect a big fat red backup light telling me something is wrong and the battery power the RAM until the battery died. This obviously isn't the case because my server sat with all 4 lights on and the computer was unplugged for almost 24 hours before I finally got it reconnected.

Edit:

After giving up for about 45 mins I turned the computer off. 1 light is currently lit for the battery power. Remember, I had 4 lights lit for the last 24 hours.

[wmaciv]

My ANS-9010 is rev 3.1 hardware, not the newer 3.3 I wonder if the CF functionality is improved/fixed on the 3.3 hardware? I am pretty sour about the level of technical advice provided by ACARD about the CF problem. I think I just straight up have a lemon of a drive in the CF department. Other than that, it's a great 32GB ram drive.

Wade