Sunday, September 18, 2011

AMD SB850 SATA controller vs eSATA port multiplier

Last year, I bought three 4-drive eSATA enclosures from SANS. They each come with their own PCI-E eSATA controller card with a Sil3132 Silicon Image controller.

So far, this is the only SATA controller I have ever found that can recognize multiple drives in this enclosure.

The list of the controllers I tested that failed is long :
- Intel ICH9R built-in to my GA-P35-DS3R motherboard
- Promise TX4302
- VIA VT6422 PCI card (technically SATA, not eSATA)
- Gigabyte / J-Micron

And now I can add the AMD SB850 built-in to my GA-890GPA-UD3H motherboard.
What's interesting about this latest controller is that AMD claims port multiplier support with this chip.

And indeed, the BIOS can see multiple drives if I boot with the enclosure connected.
But after the OS boots up (Win7), it only recognizes the first drive in the enclosure.
Very frustrating. I hope someone at AMD is listening. I am using the AMD AHCI compatible driver version 1.2.296 dated 4/15/2011, under Win7 x64.

I tried designating the port as "ESP" in the BIOS, but the only difference that makes is showing the "Eject" icon in Windows. It doesn't help recognize the extra drives in the enclosure.

Gigabyte GA-890GPA-UD3H rev 2.0 motherboard vs SATA 3.0 devices

Earlier this year, I purchased the above motherboard. It was advertised as supporting SATA 3.0 on its AMD SB850 controller, with speeds up to 6 Gigabits/s. I didn't have any SATA 3.0 devices at the time I purchased the motherboard to test with . I do now - two hard drives and two solid state drives.

Much to my dismay, I was initially unable to achieve these speeds. The burst speed in popular benchmarks such as HDTach topped around 200MB/s, which was clearly SATA 2.x 3 Gigabits/s territory.

After much googling, I found that I was not alone with the problem. Fortunately, I found that a beta BIOS update from Gigabyte, version FGf, solves this problem. But it took many hours and grief to figure it out.

The beta BIOS did introduce another problem, though. When using AHCI mode on the AMD controller, the machine would no longer go to sleep under Win7. More precisely, it would go to sleep and immediately wake back up. The fix was to disable "USB wake from S3" in the BIOS. I didn't have to do this before. It isn't required if I use RAID mode on the SB850 contrller, either. Just another BIOS bug. Hope this will save somebody else time.

Wednesday, September 14, 2011

If you buy new memory, make sure to run memtest right away.

ALWAYS. Especially if there is a rebate on it.
Unfortunately, cheap RAM appears to mean that quality control is low, ie. it's now done by the customer.
Astute readers of this blog may have noticed a picture of a failing memtest yesterday.
I had a bad experience with a 16GB kit. The manufacturer, Patriot Memory, is local to Silicon Valley, and was helpful. They let me get a replacement in person the same day. Unfortunately the replacement also failed memtest :-(

This Patriot DDR3-1333 16GB kit was only $89.99 at Fry's, with a $20 rebate, a price unheard of . The store took it back. I bought a DDR3-1600 kit from Corsair for $94.99, with a $10 rebate. I hope this one will work better and will pass memtest tonight.

Is Level 3 cache on the AMD Phenom II x6 CPU slower than RAM ? No.

Update: I was running with an old version of memtest that didn't support my CPU.
Version 4.20 supports it and shows proper numbers. The cache speed, is in the 8GB/s range and memory access is 5GB/s. Sorry about the useless post.

That's what memtest 4.0a would like me to believe.

If so, AMD could have saved a few transistors by not including that cache at all. And perhaps I could gain some speed by disabling it ?

All joking side, hopefully the explanation is that this is a glitch in memtest.

Or perhaps the slower L3 cache is a requirement of the AMD architecture to avoid inconsistencies in memory state between cores.