ACARD RAM Disk 9010 series

[Speederlander]

Sorry, it does sound confusing. The normal operation for non-ECC memory is to enable the quasi-ECC feature. This feature uses 1/9th of the total size of the memory, so whatever size you have installed you will lose 1/9th. If you put a jumper over the "RESERVED" jumper, which is the second jumper from the right when looking at it, this disables the quasi-ECC feature giving you the total capacity of the RAM sticks. I have a copy of their manual from 2 months ago, and it did have a feature to disable the quasi-ECC feature, but they strongly recommended you use it. I disabled it and ran read/write tests all night with no problems.

I would still recommend you not use a jumper there since they seem to have intentionally wanted to remove the 'feature' by not explaining the jumper in the manual.

I just had to know what was the purpose of the jumper since they deliberately covered up the sticker on the case for the jumper configuration.

Anything 8 gig and up acting as a permanent RAM drive you want to run with ECC. Doing it on the cheap with non-ECC is the wrong approach. Large ram drives simply have to use ECC RAM. So not using ECC and then disabling their "quasi-ECC" (however well that works) is a bad plan IMO.

[wmaciv]

From what you said earlier, I'm thinking of splurging for 2 4gb sticks at an unsavory cost, then finishing out the slots with 2gb sticks for a rough total of 20 gig, since 16 is cutting it a little close for what I want to install (OS and a few key apps). I really wonder how this thing will behave with the Intel ICH9R chipset on my MB (Supermicro C2SBX). It had some early issues with the I-RAM's, before a BIOS update.

[josh1980]

Speederlander, why 8GB? What happens at 8GB that makes ECC important? Or are you just trying to say that if you use it as a permanent RAM drive you should use it? I wouldn't recommend ever disabling the feature. 1/9th of your capacity isn't much to lose to prove your data is correct via ECC.

Me personally, I intend to buy 4GB sticks for all 8 slots and make it my boot drive, with 2x1TB hard drives for storage space. I have a RAID5 array consisting of 8x1TB drives, so I don't really need to keep alot of storage space on my workstation.

I'm sorry I haven't had a change to test the RAID0 using the onboard RAID chipset. Work has kept me busy this weekend, but I intend to get results as soon as I can. I'm just as curious as others to see what the RAID chipset can do. My motherboard is an Asus P5E if anyone wants to look up the RAID chipset. I think it's the ICH9R.

[Speederlander]

Speederlander, why 8GB? What happens at 8GB that makes ECC important? Or are you just trying to say that if you use it as a permanent RAM drive you should use it? I wouldn't recommend ever disabling the feature. 1/9th of your capacity isn't much to lose to prove your data is correct via ECC.

Me personally, I intend to buy 4GB sticks for all 8 slots and make it my boot drive, with 2x1TB hard drives for storage space. I have a RAID5 array consisting of 8x1TB drives, so I don't really need to keep alot of storage space on my workstation.

I'm sorry I haven't had a change to test the RAID0 using the onboard RAID chipset. Work has kept me busy this weekend, but I intend to get results as soon as I can. I'm just as curious as others to see what the RAID chipset can do. My motherboard is an Asus P5E if anyone wants to look up the RAID chipset. I think it's the ICH9R.

I say 8GB only because people tend to add in 4GB increments. At 4GB, RAM suffers from the occasional transient bit error but probably not enough to worry, especially when used as simple volitile memory (though I would, since I really value my data).

The original Corsair rule I recall is: 1 bit error occurs in 256MB of ram every month.
4GB = 15 bit errors/month
8GB = 31 bit errors/month
16GB = 62 bit errors/month
32GB = 125 bit errors/month

However, other more recent sources maintain 1 bit error per gigabyte per month, so that cuts those numbers by a factor of 4, i.e.:
4GB = 4 bit errors/month
8GB = 8 bit errors/month
16GB = 15 bit errors/month
32GB = 31 bit errors/month

Which is more correct? My guess is that it's somewhere in between, governed by the quality of the RAM, the luck of the draw, and a host of other issues.

There is a reason that server boards only use ECC RAM.

At 8GB, is the incidence still acceptable? You have to decide. But if that RAM is now operating as an always on, permanent drive, to not use ECC when you will probably be using 16GB or more would be nuts.

[IanB]

But if that RAM is now operating as an always on, permanent drive, to not use ECC ... would be nuts.

+1 QFT. Which is why the software quasi-ECC feature is such a brilliant idea, at the cost of a small amount of capacity, if the cost of ECC memory is too high.

[wmaciv]

My ANS-9010 just arrived this evening. I have to go out of town tomorrow morning for 10 days... argggghhhhh! Anyway, I don't have any memory for it yet, think I may go with 16 gigs of 4 x 4gb dimms initially... And will spring for the 32 meg CF card also... wonder if it's worth spending the premium on high transfer rate CF memory? Bet not...

Hopefully I can post some Platypus to ACARD and IRAM to ACARD transfer stats in the near future.. I can't WAIT to see this as my boot drive...

Wade

[RADCOM]

Who has it in stock? Website still saying pre-order!! Here's an interesting thread re the Acard on the Anandtech forum.
http://forums.anandtech.com/messagev...VIEWTMP=Linear

[tiro_uspsss]

Here's the Benchmark results:

Single Port tests
----------------
[INDENT]HD Tach -
[INDENT]Random Access: 0.1ms
CPU Utilization:3%
Average Read: 167.8 MB/sec
Average Write: 140.3 MB/sec
Burst Speed: 171.9MB/sec
The graph was perfectly horizontal, as expected.

I thought single port would have maxed out 1x SATAII port = ~300MB/s ... why is it only 168?

[B.E.E.F.]

So only ECC memory can be used with it. That sounds pricey.

[Speederlander]

So only ECC memory can be used with it. That sounds pricey.

No one should be surprised. ECC is the only serious option for larger RAM drives.

[Serra]

I thought single port would have maxed out 1x SATAII port = ~300MB/s ... why is it only 168?

I hope I can eventually get my hands on a unit to give it a thorough investigation, but I currently theorize that the reason is the controller on the ram disk. Technical specs show that it provides ~200MB/s throughput maximum (which is one reason it's sure not going to hit 300MB/s), but the question is whether that includes SATA overhead or not.

If it does, then something else is to blame and that bears further investigation.

If it does not, then 168 is pretty close to right if the controller can provide a little over 200MB/s performance.

[tiro_uspsss]

I hope I can eventually get my hands on a unit to give it a thorough investigation, but I currently theorize that the reason is the controller on the ram disk. Technical specs show that it provides ~200MB/s throughput maximum (which is one reason it's sure not going to hit 300MB/s), but the question is whether that includes SATA overhead or not.

If it does, then something else is to blame and that bears further investigation.

If it does not, then 168 is pretty close to right if the controller can provide a little over 200MB/s performance.

mm.. well.. personally I had hoped for more considering the price

*sigh*

[josh1980]

The unit does not require ECC memory, but can use it. If ECC memory is installed, the unit will use the ECC feature to fix errors. If ECC memory is not installed, then 1/9th of the total capacity of the RAM is used for ECC functions, and is not used. The big difference between ECC memory and non-ECC memory is a 9th chip on a RAM stick. This 9th chip is identical to the other 8, but the data stored in the 9th chip is similar to a checksum of the RAM bank. If you chose not to buy ECC memory for this unit, then it will create it's own ECC function utilizing 1/9th the space.

I agree with tiro_uspsss that 168MB/sec is somewhat limiting, but it could be a limit from something else besides the box. I do own a Gigabyte I-RAM card, and I benchmark it at 120MB/sec(the limit for SATA 150).

I also did order 4x4GB of RAM today for my box. I'll buy 4 more 4GB sticks next week when I get paid again. This box is going to make my core i7 computer boot at light speed.

Also, the motherboard RAID benchmark's I've been promising to do will be performed tonight. I'll post the results when I get them done.

[Buckeye]

Just catching up on this thread and have a question. I see reading back a bit that Access time is 0.1ms ?

I thought that these units are faster than that ?

[B.E.E.F.]

No one should be surprised. ECC is the only serious option for larger RAM drives.

I'm surprised. They can increase sales by making it non-ECC compatible.

[RADCOM]

I'm surprised. They can increase sales by making it non-ECC compatible.

Features:-
Supports up to 64GB amount of memory
Supports ECC/Non-ECC DDR2 400/533/667/800**
Automatic data backup/restore between DDR2 memory and CF card
Built-in Lithium Battery
LED indicators for battery capacity, power status, SATA ports activity, backup status
Driver less: need no driver on host side
**Note: Please refer to Compatibility List for DDR2 RAM Module
source:
http://www.acard.com.tw/english/fb01...ve&idno_no=270

[Speederlander]

I'm surprised. They can increase sales by making it non-ECC compatible.

Running a big RAM drive without ECC means the person has no concern whatsoever for data integrity.

[wmaciv]

Josh, what kind of ram and where are you getting 4gb density sticks? Are you paying a premium, or have you found an inexpensive source? I would sure like to find some for my 9010 without spending what I think I am going to have to.....

Wade

[josh1980]

Just catching up on this thread and have a question. I see reading back a bit that Access time is 0.1ms ?

I thought that these units are faster than that ?

You are correct. But most benchmark programs are written to give the results for the hard drive access time in ms, and they can't calculate below 0.1ms. They all round, and in this case, they round up to the smallest possible division of 0.1ms.

The h2benchw tests showed the following:

Random access read: average 0.05, min 0.03, max 0.07 [ms]
Random access write: average 0.05, min 0.04, max 0.08 [ms]

Even at 0.1ms, you wouldn't hear a peep out of me. I don't know of any other hardware out there with these kinds of seek times. These are 100ths of a millisecond. All I can say is "Awesome".

I was playing around some today, and I figured out something else that I figured I would share. My BIOS has SATA options of Disabled/Compatible/Enhanced. If I chose "Enhanced" I lose about 10% performance vice "Compatible". Also worth mentioning is disabling ECC feature does not result in any change in performance.

I spent some time attempting to use the onboard RAID controller for my motherboard, and I was unable to get it to work. Unfortunately I do not have a spare hard drive to install another copy of Windows on, so we'll have to wait until either I get my new core i7 computer next week, or someone else can perform some benchmarks.

[Speederlander]

You are correct. But most benchmark programs are written to give the results for the hard drive access time in ms, and they can't calculate below 0.1ms.

Absolutely not. These are my RAM drive.

[B.E.E.F.]

Running a big RAM drive without ECC means the person has no concern whatsoever for data integrity.

That's right.

Features:-
Supports up to 64GB amount of memory
Supports ECC/Non-ECC DDR2 400/533/667/800**

My bad. Thanks.

[Serra]

Absolutely not. These are my RAM drive.

Yes, but I assume that is a drive carved out of your system RAM, not a peripheral. To speak with your physical RAM there is effectively no wait time... but the question is, does the program perhaps perceive that there is a "non-zero" wait-time to a hard-drive (or SSD) and therefore just round up to 0.1? I personally think it is likely.

[Speederlander]

Yes, but I assume that is a drive carved out of your system RAM, not a peripheral. To speak with your physical RAM there is effectively no wait time... but the question is, does the program perhaps perceive that there is a "non-zero" wait-time to a hard-drive (or SSD) and therefore just round up to 0.1? I personally think it is likely.

I was responding to:

they can't calculate below 0.1ms

Also, I-ram shows as 0.0ms, and it's a peripheral, not directly created from system RAM. And IIRC, doesn't it also use the SATA interface, just using PCI for power? Not that .1ms is a lot.

[Buckeye]

Great ! Thanks for that info on access time.

I am looking at 2 of the 9010 installed onto my ARC-1231ML and set in Raid 0. From what I gather the 9010 has 2 SATA ports on each unit, so I would need 4 SATA ports total. This matches up just fine with the 4 SATA ports I have left on my controller card.

I wounder now how well these will scale in a setup like this. That would be 128gig total and a OS installed on that. I am thinking that would be in the range of 800mb/sec bandwidth with 0.05ms access time. Pretty fast.

However with my current SSD Raid 0 I am in the range of 950mb/sec bandwidth and 0.1 access time.

Side by side these 2 setups should perform pretty close but with the ACRAD mopping the floor with writes and small file access.

Should be an interesting test thats for sure.

I-RAMS are looking rather interesting to me also, I don't like the fact that they need a slot for power. I can just see 12 I-RAMs connected to my ACR-1231ML as a Raid 0, but as far as being usefull for anything other than a bench is not good.

[tiro_uspsss]

I-RAMS are looking rather interesting to me also, I don't like the fact that they need a slot for power. I can just see 12 I-RAMs connected to my ACR-1231ML as a Raid 0, but as far as being usefull for anything other than a bench is not good.

i-rams dont work with areca cards