Gigabyte i-Ram cards

Gigabyte i-Ram cards

[Jeff Garzik]

While responding to email related to Bugzilla# 6163 [1], I took a look 
at the Gigabyte i-Ram[2]:  PCI card, where PCI only provides power to 
some DIMMs.  Data xfer to/from DIMMs via SATA cable.  So, at current 
prices[3] one can get a 1GB Solid State Disk for US$206:  US$153 for the 
card, and US$53 for 1GB pc2700 ddr.

I'm impressed.  I ordered one to help debug BZ#6163, but I think these 
cards will help expose bugs in libata due to their high speed.  With 
these cards, we can push more data through libata than would normally be 
possible with a standard disk drive.

I've seen these for PATA before, but this is the first inexpensive SATA 
SSD that I've seen.  Let's hope its not riddled with hardware bugs 
<crosses fingers>.

	Jeff


P.S. Full disclosure:  I've had no contact with Gigabyte at all, nor has 
Red Hat (my employer) to the best of my knowledge.  I just like the gadget.


[1] http://bugzilla.kernel.org/show_bug.cgi?id=6163
[2] http://www.gigabyte.com.tw/Products/Storage/Default.aspx
[3] http://www.cdw.com/shop/products/default.aspx?EDC=929424

Re: Gigabyte i-Ram cards

[Eric D. Mudama]

On 3/20/06, Jeff Garzik <jeff@garzik.org> wrote:
>
> While responding to email related to Bugzilla# 6163 [1], I took a look
> at the Gigabyte i-Ram[2]:  PCI card, where PCI only provides power to
> some DIMMs.  Data xfer to/from DIMMs via SATA cable.  So, at current
> prices[3] one can get a 1GB Solid State Disk for US$206:  US$153 for the
> card, and US$53 for 1GB pc2700 ddr.
>
> I'm impressed.  I ordered one to help debug BZ#6163, but I think these
> cards will help expose bugs in libata due to their high speed.  With
> these cards, we can push more data through libata than would normally be
> possible with a standard disk drive.
>
> I've seen these for PATA before, but this is the first inexpensive SATA
> SSD that I've seen.  Let's hope its not riddled with hardware bugs
> <crosses fingers>.

I'll see if I can get Maxtor to buy one for me, then I'll be able to
hook it up to a bus analyzer.

Re: Gigabyte i-Ram cards

[Jure Pečar]

On Mon, 20 Mar 2006 21:24:38 -0500
Jeff Garzik <jeff@garzik.org> wrote:

> I'm impressed.  I ordered one to help debug BZ#6163, but I think these 
> cards will help expose bugs in libata due to their high speed.  With 
> these cards, we can push more data through libata than would normally be 
> possible with a standard disk drive.

I hope I'll manage to post my benchmarks by the end of the week. So far it looks very useful. 

-- 

Jure Pečar
http://jure.pecar.org/

Re: Gigabyte i-Ram cards

[Mogens Valentin]

Jeff Garzik wrote:

> I'm impressed.  I ordered one to help debug BZ#6163, but I think these 
> cards will help expose bugs in libata due to their high speed.  With 
> these cards, we can push more data through libata than would normally be 
> possible with a standard disk drive.
 >
> ...I just like the gadget.

I was thinking.. Someone on the Xen list suggested using suspend/resume 
for dom0 systems. Now, doing that to an i-Ram card would be nice.
I do some systems integration, and when AMD releases their Pacifica 
cpu's /and/ decent graphics (GLX/openGL based, I think it'll be) gets 
working in Xen, I could imagine having XP and/or W2K3 sleeping on such a 
card, for almost instant wakeup for a quick test.
Mind you, I don't live my life with those OSs, fortunately

If only they'd make them for more than 4GB ram.. maybe that'll change 
when AMD starts pumping DDR2-based cpu's into the market, so both [AMD 
and Intel] will be using DDR2, and prices may go down.

-- 
Kind regards,
Mogens Valentin

Re: Gigabyte i-Ram cards

[Jens Axboe]

On Mon, Mar 20 2006, Jeff Garzik wrote:
> 
> While responding to email related to Bugzilla# 6163 [1], I took a look 
> at the Gigabyte i-Ram[2]:  PCI card, where PCI only provides power to 
> some DIMMs.  Data xfer to/from DIMMs via SATA cable.  So, at current 
> prices[3] one can get a 1GB Solid State Disk for US$206:  US$153 for the 
> card, and US$53 for 1GB pc2700 ddr.
> 
> I'm impressed.  I ordered one to help debug BZ#6163, but I think these 
> cards will help expose bugs in libata due to their high speed.  With 
> these cards, we can push more data through libata than would normally be 
> possible with a standard disk drive.

But they are still only 1.5Gbps, right? Which is really a shame, since
that makes it only about twice as fast as the good drives out there in
terms of sequential transfer. Seeks are in a different league of course
:-)

Still strange why they did not go for the full 3.0Gbps link speed.

-- 
Jens Axboe

Re: Gigabyte i-Ram cards

[Bryan Fink]

Jeff Garzik wrote:

> I'm impressed.  I ordered one to help debug BZ#6163, but I think these 
> cards will help expose bugs in libata due to their high speed.  With 
> these cards, we can push more data through libata than would normally 
> be possible with a standard disk drive.


Hi.  I tried to respond on this topic about a week ago, but I haven't 
seen my response show up on any of the mirrors, so I think it must have 
not made it through.

Anyway, I just wanted to add that I am also fiddling with the Gigabyte 
i-Ram.  I haven't tried modifying and recompiling the kernel yet, but I 
did have some success another way:

My desktop is a Dell Dimension 5150.  In the BIOS, I can set the "SATA 
Operation" to either "SATA" or "RAID".  If I set it to RAID, linux will 
see the i-RAM just fine (under Ubuntu 5.10 Live CD).  It loads ahci, and 
just takes off.  If I have the setting on SATA, then linux does not load 
ahci, and does not talk to the i-RAM.

Of note is also the fact that Windows has no issue with talking to the 
i-RAM when my system is in SATA-mode.

So, I guess the question is, are Windows and my BIOS ignoring this 
"invalid" feature query reply, or are the prodding the card in some way 
other than how linux does, which makes the card respond properly?

Good luck to those who are working on this problem, and please let me 
know if there is any way I can help.

-Bryan

Re: Gigabyte i-Ram cards

[Jeff Garzik]

Bryan Fink wrote:
> Jeff Garzik wrote:
> 
>> I'm impressed.  I ordered one to help debug BZ#6163, but I think these 
>> cards will help expose bugs in libata due to their high speed.  With 
>> these cards, we can push more data through libata than would normally 
>> be possible with a standard disk drive.
> 
> 
> 
> Hi.  I tried to respond on this topic about a week ago, but I haven't 
> seen my response show up on any of the mirrors, so I think it must have 
> not made it through.
> 
> Anyway, I just wanted to add that I am also fiddling with the Gigabyte 
> i-Ram.  I haven't tried modifying and recompiling the kernel yet, but I 
> did have some success another way:
> 
> My desktop is a Dell Dimension 5150.  In the BIOS, I can set the "SATA 
> Operation" to either "SATA" or "RAID".  If I set it to RAID, linux will 
> see the i-RAM just fine (under Ubuntu 5.10 Live CD).  It loads ahci, and 
> just takes off.  If I have the setting on SATA, then linux does not load 
> ahci, and does not talk to the i-RAM.
> 
> Of note is also the fact that Windows has no issue with talking to the 
> i-RAM when my system is in SATA-mode.
> 
> So, I guess the question is, are Windows and my BIOS ignoring this 
> "invalid" feature query reply, or are the prodding the card in some way 
> other than how linux does, which makes the card respond properly?

Your description seems to imply this is a BIOS+driver issue, not 
anything related to the gigabyte card.

Intel ICH boards can be driven using either the ata_piix or the ahci 
driver, depending on BIOS mode.  Most likely, when you switched to SATA 
mode in BIOS, it started programming the motherboard to boot in 
IDE-compatible mode (ata_piix) rather than SATA FIS mode (ahci).

	Jeff

Re: Gigabyte i-Ram cards

[Tejun Heo]

Jeff Garzik wrote:
> Bryan Fink wrote:
>> Jeff Garzik wrote:
>>
>>> I'm impressed.  I ordered one to help debug BZ#6163, but I think 
>>> these cards will help expose bugs in libata due to their high speed.  
>>> With these cards, we can push more data through libata than would 
>>> normally be possible with a standard disk drive.
>>
>>
>>
>> Hi.  I tried to respond on this topic about a week ago, but I haven't 
>> seen my response show up on any of the mirrors, so I think it must 
>> have not made it through.
>>
>> Anyway, I just wanted to add that I am also fiddling with the Gigabyte 
>> i-Ram.  I haven't tried modifying and recompiling the kernel yet, but 
>> I did have some success another way:
>>
>> My desktop is a Dell Dimension 5150.  In the BIOS, I can set the "SATA 
>> Operation" to either "SATA" or "RAID".  If I set it to RAID, linux 
>> will see the i-RAM just fine (under Ubuntu 5.10 Live CD).  It loads 
>> ahci, and just takes off.  If I have the setting on SATA, then linux 
>> does not load ahci, and does not talk to the i-RAM.
>>
>> Of note is also the fact that Windows has no issue with talking to the 
>> i-RAM when my system is in SATA-mode.
>>
>> So, I guess the question is, are Windows and my BIOS ignoring this 
>> "invalid" feature query reply, or are the prodding the card in some 
>> way other than how linux does, which makes the card respond properly?
> 
> Your description seems to imply this is a BIOS+driver issue, not 
> anything related to the gigabyte card.
> 
> Intel ICH boards can be driven using either the ata_piix or the ahci 
> driver, depending on BIOS mode.  Most likely, when you switched to SATA 
> mode in BIOS, it started programming the motherboard to boot in 
> IDE-compatible mode (ata_piix) rather than SATA FIS mode (ahci).
> 

Hello, Bryan, Jeff.

iRam is not detected if the controller is configured as legacy mode 
because then ata_piix got attached and it checks diagnostic code which 
iRam fails to report properly. On the other hand, if you configure it as 
AHCI (or RAID), the ahci driver gets attached which accidentally doesn't 
check the diagnostic code, so the iRam is detected normally and can be 
used. sata_sil24 also doesn't check the diagnostic code, so iRam would 
work with sil24 too.

Well, apparently, BIOS and Windows driver don't check the diagnostic 
code and that's probably why gigabyte missed it. I'm voting for joining 
the crowd and ignoring it (with a warning). We have plenty of other 
mechanisms to verify attached devices and giving users an opportunity to 
try seemingly failing device isn't a bad idea.

-- 
tejun

Re: Gigabyte i-Ram cards

[Mogens Valentin]

Jeff Garzik wrote:
> Intel ICH boards can be driven using either the ata_piix or the ahci 
> driver, depending on BIOS mode.  Most likely, when you switched to SATA 
> mode in BIOS, it started programming the motherboard to boot in 
> IDE-compatible mode (ata_piix) rather than SATA FIS mode (ahci).

So, this could apply to Dell Dimension 9150 as well?
I only have limited experience with Gentoo (used at work), so I'm not 
sure what's actually what as yet on those boxes.
However, using stock BIOS setup, i.e. SATA mode, I couldn't install 
Slackware test26 (2.6 kernel), which use libata.
Didn't have time to investigate further.

-- 
Kind regards,
Mogens Valentin

Re: Gigabyte i-Ram cards

[Jure Pečar]

On Tue, 21 Mar 2006 23:53:13 +0100
Jure Pečar <pegasus@nerv.eu.org> wrote:

> I hope I'll manage to post my benchmarks by the end of the week. So far
> it looks very useful. 

Finally, I have something: http://nerv.eu.org/iram/

I plan to do some more tests, when time permits.

-- 

Jure Pečar
http://jure.pecar.org/