Intel 320 SSD aka G3

**Brahmzy** · 03-29-2011, 02:06 PM

Originally Posted by Anvil

I noticed that comparison, question is, is that using the shipping firmware.

Also, the most noticeable "speed" penalty is the random read on the 120GB, half the iops of the 240GB, for most users it won't change a thing though.

2R0 V3 at any capacity should be great fun.

Yeah, the random read IOPS do take a major hit on the 120's - wish I knew EXACTLY how that translates to real-world for me though...

I suppose buying 2 of the 240's would be the way to go, but goodness that's a lot of snaps for one of the fastest depreciating items in a PC that'll be obsoleted in 2-3 months, if it hasn't been already on release day.

**Anvil** · 03-31-2011, 01:49 AM

I'm sitting here looking at 2 160GB G3s

One is the brown box (10) and the other one is a reseller (K5).
The brown box is just like the G2s, it includes a sticker with "SPEED DEMON" printed on it,
the reseller includes 3.5" adapter bracket and screws + SATA and power cables.

I won't be able to do any tests for another 4-5 hours but I'll post a photo in a little while.

320_photo1_wo (1) - Copy.jpg

I just had to perform 1 AS SSD test, result is as expected, lower than the G2 on 4KB stuff.

as-ssd-bench INTEL SSDSA2CW16 3.31.2011 12-39-43 PM.png

**F@32** · 03-31-2011, 07:13 AM

The fact that new gen 25nm drives neglected 4K performance improvement is pretty sad. Years later and we are still stuck at 20MB/s on MLC SATA drives. Acard and io-Xteme still rule.

**ghmz** · 03-31-2011, 08:12 AM

Originally Posted by Anvil

I'm sitting here looking at 2 160GB G3s

One is the brown box (10) and the other one is a reseller (K5).
The brown box is just like the G2s, it includes a sticker with "SPEED DEMON" printed on it,
the reseller includes 3.5" adapter bracket and screws + SATA and power cables.

I won't be able to do any tests for another 4-5 hours but I'll post a photo in a little while.

320_photo1_wo (1) - Copy.jpg

I just had to perform 1 AS SSD test, result is as expected, lower than the G2 on 4KB stuff.

as-ssd-bench INTEL SSDSA2CW16 3.31.2011 12-39-43 PM.png

My 80GB G2 in comparison..

**Anvil** · 03-31-2011, 09:35 AM

A few more standard tests before I do some iometer and finally raid-0

atto_g3_P67_3gbs.JPG CDM_1GB_G3_P67_3Gbs.JPG

**Anvil** · 03-31-2011, 12:43 PM

Random read 4K through 128KB on single G3 160GB on P67 3Gb/s port.

rr_4K-128K_G3_160GB_PCH_3GBs.jpg

**Ao1** · 03-31-2011, 12:47 PM

Nice graph Anvil

Any chance of doing a comparison with the G2?

**Anvil** · 03-31-2011, 12:58 PM

Working on it

I have collected the data, just need to compile a new chart.

A mixed bag so far, it looks better in mixed tests (read+write) at low QD than the G2 and then when the throughput is exhausted it's more or less a flat line.
The G2 scaled better at high QDs but after all we are looking for improvements at QDs < 32 for the most

About to test some raid-0 before I dive into more charts...

**Ao1** · 03-31-2011, 01:11 PM

Any chance of adding 512 byes and 1MB xfers as well?

It would be good to also see how other SSD's perform with that test.

Edit: Interesting to see how xfers sizes peak at different QD's

**Tiltevros** · 03-31-2011, 01:35 PM

from 1-4QD i think it scales better than a G2

**Computurd** · 03-31-2011, 07:23 PM

interesting. cant wait to see your comparison graphs anvil!

**Ao1** · 03-31-2011, 11:17 PM

Hi Anvil,

If I look at your graph it seems to provide a way to determine tangible “real life” benefits.

For discussion, but as a 1st stab:

Xfer weighting is based on stats I’ve obtained from hIOmon, which broadly line up with what JDEC define as SSD Enterprise Endurance Workload.
QD weighting is based on stats I’ve obtained from hIOmon. As QD 1 dominates for desktop users I’ve rated it at 85%. QD 2 = 10%. QD 3 = 5%.

In the example below, based on approximates for your IOmeter results it gives a score of 33. At this stage I can only guess how much that score would have to vary before a tangible real life improvement could be seen. As a guess however the score would have to be +/- 5 before an observable difference could be seen.

I’ve attached excel sheet, which has formulas that allow MB’s speeds to be entered and automatically calculate a score based on the inputs.

What do you think?

(Anvil I've made some cosmetic changes to the excel sheet. (The formula's remain the same, I've just made it easier to see how the score is calculated).

**Anvil** · 04-01-2011, 12:04 AM

Looks interesting

I've downloaded the Excel file and will play with it, got lots of data but 512B and 1MB I'd have to regenerate.

QD is a tricky one as it depends on usage but I'd like think that QD is more like 1.x rather than 1.0, 1.0 basically means that there is no queue or multitasking going on.
Lets play with that spreadsheet and see what different setups scores, I have collected QD1..16 Linear in a separate iometer test so I've got all details except for 512B and 1MB.

I ran a few 512B tests on the 320 and it didn't look bad at all, will post both the raid-0 charts and benchmarks today.
I'll load the results into that spreadsheet asap

**Anvil** · 04-01-2011, 03:48 AM

OK, this is 2R0 320 strip size 128KB

I've changed the QD3 to QD4

and the figures for 1MiB is just a guesstimate.

2r0_320.PNG

for comparison, a single Vertex 2 60GB (used)
v2_60.PNG

**Ao1** · 04-01-2011, 04:39 AM

Interesting. I’d guess that +/- 5 on a score of 33 would be hard to notice. I’d also guess that it would take and Acard or io-Xtemeto go above 5.

Did you estimate 512B as well?

**Anvil** · 04-01-2011, 04:44 AM

The ones on the 320 is correct (except for all 1MB figures)

On the Vertex I just kept the figures for 512B from the Intels, so they are not correct.
(I'll rerun the 512B on the Vertex 2 later, at 4% it doesnt tip the scale)

**felix_w** · 04-01-2011, 10:03 AM

Originally Posted by Anvil

OK, this is 2R0 320 strip size 128KB

I've changed the QD3 to QD4

and the figures for 1MiB is just a guesstimate.

2R0 Intel 320 Random 4KB read QD1 is 13.6 MB/s (on Io-Meter if i understand correctly) ? Or did i get something wrong ?

**Anvil** · 04-01-2011, 11:06 AM

That is the result, using that strip size and the P67 3Gb/s ports.

4K QD1 ~14-19MB/s depending on strip size and other settings like wbc, wbc flushing.

Not alarmingly low imho, 25nm took a dive at 4K, 8K looks great though

**Anvil** · 04-14-2011, 02:46 AM

Received the 320 Series 600GB today

Nothing remarkable about it except for the capacity and write speed of course.

CDM_G3_600GB.JPG as-ssd-bench INTEL SSDSA2CW60 4.14.2011 12-21-56 PM.png

Should be perfect for computers where capacity is an issue, laptops in particular.