The collected works of mapesdhs - Page 4

Tezro Quad-R16K/1GHz (16MB L2), 4GB RAM, 36GB 10K system, 48 x 146GB 15K JBOD RAID, 4 x QLA12160/66 PCI64,
DM3 card (not in hinv as drivers not installed). Normally connected to a DM5/VBOB, but again not shown here. The system
originally had two QLA2342s, RAD Audio, dual-port Gbit NIC and dual-port serial card, but I've removed all these.




I have a second similar system (same CPU/gfx/RAM) that still has the original other cards.

Ian.

indyman007 writes:
> Do you have any particular use for it?

There are things I can use it for (movie conversion tasks), but neither system is for keeps. I bought them stricly as an
investment, because the ISA interest rates are now so useless. They were expensive (thousands); I couldn't justify
spending so much on something like this for personal use, not at present. My bank savings balance would have to
have at least one extra digit on the end for that to be viable.

A US movie company is interested in one of them, perhaps later in the summer, not sure yet.

Oh, I have a quad-700 system aswell.


bri3d writes:
> This is what I hope to turn my O350/Onyx system into - ...

Is it still a dual-1GHz V10 atm?


> replace my V10), and a DM3 (I probably won't end up getting one, as I don't have much use, but...) and I've got it :p

I got one from Harry, for the quad-700 system I have. He probably has more.


> @indyman007 - would probably be used as a Discreet system in "the real world" - there's not much else that uses
> a DM3 and that much power.

Correct, they were all Smoke systems. The unit pictured at the start of this thread still has a sticker on the front panel,
behind the front flap, that says, "SMOKE2". Naturally, the system disks were wiped though. Each system came with
either an original IR36 or IR73 Stone FC array, ie. a 15-bay FC chassis filled with either 36GB or 73GB FC drives
containing proper Discreet firmware.


sgtprobe writes:
> Wow, Nice system Ian.

Thanks!

I've added a few benchmark results for the system, but have a few more to do.


hamei writes:
> Ian, since you pulled the serial card, could you look and see what
> UART's it has ? Maybe 16550's ?

The board (PN 030-1657-003, dated 2000) has just four main ASICs:



There are two more much smaller 14-pin ASICs, but the writing on them is too small to discern as I type this.

I have a second serial card, marked SGI Imagesync, PN 030-1723-001. It has a completely different layout, the main
ASIC being a XILINX SPARTAN XCS40XL (PQ208AKP0029, D1106126A).

While I think of it...

The dual-port Gbit NIC in the systems is SGI PN 9210292, Tigon3 I think (Broadcom ASIC), marked Silicom Ltd
Connectivity Solutions, model number PXG2-SGI.

The RAD Audio card is PN 030-1649-001, LSI ASIC L2A0470 RAD1 REV B.

Hope this helps!


Btw, does anyone know if Blender 2.48 can do renders using networked systems? If so, is it easy to setup? I was
wondering how hard it would be to run my benchmark using all three Tezros, see what happens.

Ian.

Richtom1 writes:
> I'll be looking to get a set up like that next..........unless I die first bummer

In that case, take all possible measures to live longer.


> btw the DM3 I have is PN: 030-1927-003 Rev B and it's PCI not XIO

Ah, no, it's not a PCI card. Check a DM3 installed in a Tezro, it does not plug into a PCI slot.

The only PCI SGI video board I know of is the DM6, a card Discreet never officially supported.

Ian.

Just a thought about the DM3...

It fits into the same type of double-socket that the V12 is plugged into (definitely XIO, or rather XIO2 I suppose) Thus, with a
bit of metal hacking of the rear chassis plate, could one fit two V12s into a Tezro tower unit? This would mean no DM3 though.

Wild thought: if one knew how, maybe the spare XIO port could be linked to the same port in a different Tezro, make an 8-CPU
system. Hmm, probably not.

Funny thing - before Tezro launched, some of the comments I heard from SGI people led me to infer that the system would
indeed be linkable in this way, but alas that didn't happen. Pity, would've been cool.

Ian.

Quick update...

I've tested a 1TB SATA drive on a Fuel with an LSI SAS3442X-R SAS/SATA card and it works fine!

Check the hinv , and here's the diskperf/df output (the SATA drive is the one mounted on /0):



So, the 1TB SATA gives 953GB of usable space (via 1MB = 1024KB, etc.). I'm just about to run a speed
test to compare access times to the system disk and Maxtor 15K data disk. More shortly...

Oh, I also tested a SUN SAS drive which worked ok, a Hitachi 146GB 10K, model H101414SCSUN146G. Have
a few more to test this evening...

---

More results from the SAS drives! First, a SUN-badged Hitachi 146GB 10K (H101414SCSUN146GSA23):




Here's a SUN-badged Seagate 146GB 10K SAS (ST9146802SS):



Nice! Now here's two of the Seagates in an XLV optimised for HD:




And now top of the pack, the SAS version of the Seagate 15K mentioned in my earlier post (ST3146855SS):



The SCSI version of this drive is significantly slower for sequential read/writes of smaller request sizes, but
much the same for larger sizes, though the SAS is more consistent.


Lastly, though not an optimal arrangement since it's using three different models, here's an XLV of four
146GB 10K SAS drives, optimised for HD:



Either way, the LSI card works very well!

Ian.

ramq wrote: Darn. You beat me to the SAS drive test.
All because my Fuel is still at your warehouse...

Yeah, sorry about that; couldn't post anything last week, spent all my time tidying up the place ready for
a Landlord/property inspection (rented house). Most inconvenient...

Ian.

ramq wrote: I've got a couple of Seagate ST973402SS drives I want to evaluate.

They should perform the same as the ST9146802SS mentioned earlier, as it's the same model
range (Savio 10K II), ie. 90MB/sec for one drive, just under 180MB/sec for two in XLV.

Ian.

ShadeOfBlue wrote: Note the "decrease SGI revenue", the export laws are probably just an excuse

Given how fast commodity clusters had already become by then, probably yes.

Irony is, Russia had a better chip anyway (Erebus2K), though it floundered in the end, bought out by SUN.

Ian.

hamei wrote: We need to institute IQ testing as a voter qualification.

Hmm... in the UK, I reckon simply being able to speak properly would be a better starting point. 8\


Anyway! Back to more cheery things. Any more upgrades planned for Iceberg?

Ian.

A quick bump to help spread some useful info:-

I've replaced the LSI 22320 card with an LSI SAS3041X-R SAS/SATA card, and installed a 1TB SATA, so now my Fuel has
a 1TB SATA to act as a backup clone/copy of my movie archive disk in my PC. Here's a df (oh yeah, I also replaced the
146GB 15K Maxtor system disk with a 300GB 15K Fujitsu MBA3300NC):



I've not yet finished copying the data over to the SATA in the Fuel from my PC, but it's pretty quick via ftp, about 75MB/sec.
An intriguing test given that the both the source and destination for this transfer is exactly the same model of drive (Samsung
Spinpoint F1 1TB 7200rpm SATA, model HD103UJ). Would be interesting to see how two of these operate as a RAID0 with
a Fuel, but that's for the future.

The SAS3041X-R just has 4 internal ports, but it's quite cheap and thus ideal for installing SATAs in a Fuel. Also note that
using the fastest possible SATA (WD VR 10K series) does work fine with Fuel/Tezro and the performance is good, beating
all SCSI 10Ks and most SCSI 15Ks (see my disks performance results page ), though 2nd-hand prices are not quite as good
as 15K SCSI. OTOH, the WD VR makes less noise, uses less power and generates less heat.

I'll update the hinv, etc. in my o.p. shortly.

Ian.

PS. Am hoping to be able to test an SSD in October.

hamei wrote: There's no room for discussion, Ford GT.

Nah, I'd rather have a Veyron any day. GT is waaaay too slow. ;D

Ian.

ajerimez wrote: ... and a system that can only do NTSC video in realtime ...

Or PAL of course. ;D

Note that Effect on O2 can only stream video to/from RAM in 30 second bursts, not to/from disk continuously,
as opposed to Octane/Flame which has no such limitation. I think Flint on Indigo2 works the same way.

To the original poster: MovieMaker works ok for basic things, but try and do complex stuff and it will barf.
There are GUI tools for recording video and playback aswell (MediaRecorder/MediaPlayer) but in both cases
it is much better to use the command line equivalents which are more reliable. Premiere is ok aswell, and is
able to use the O2's acceleration hardware, but again it's not bug-free.

There are various freeware tools (Nekoware) you can use of course, mencoder/mplayer being the main items,
but even a max-spec O2 cannot use mplayer to play back full-size/rate DivX/MPEG2/MPEG4 video in
real-time (alas, O2's ICE hardware is not used to accelerate mplayer). O2 is very good at MJPEG video though,
and indeed can record direct to AVI which makes it easy to use a PC to convert movies into other formats. I use
an O2 for capture/editing/playback (CEP for short), but I use a PC for final format conversion. VirtualDub, DivX
Converter, PICVIdeo JPEG codec, etc. Plenty of tools available. I'll be building an i7 system soon to aid with
conversion (current system is only a 6000+).

Note that I do have some professional analogue video units for O2, but I've not yet looked into exactly what they
can do.

Lastly, my site has some hints/tips on capturing video using IMPACT Compression, but many of the concepts
apply to O2 aswell. Read the O2 man pages for full details of the individal commands such as dmrecord and
dmconvert, also note the presence of online books such as the Digital Media Tools guide. And read the release
notes too - all useful info.

If you want it *simple*, don't use an O2. Get a Mac/PC/Linux box. If you want it *interesting*, get an O2.

However, don't bother with an O2 if you're not able to afford or obtain a decent spec system. Using a lesser spec
O2 for this would not be a pleasant experience. I would recommend a minimum of an R10K/250 with 512MB RAM,
and whatever storage you need. For the R5K line, don't use anything less than an R5K/300, though even then an
R10K is preferable. Or get an R7K/600, that would be good. See my site for benchmark results , eg. video conversion.

Ian.

Found this in the Jan/Feb 1998 issue of Developer News, part of the launch PR for the R10K/250 CPU and E-series MGRAS gfx:



Just goes to show what companies were willing to pay in order to have, at the time, superior performance, reliability, etc.

Anyone else got some old original list price examples? I mean new pricing, not 2nd-hand units from a reseller.

I remember seeing a quote for a 256MB RAM kit for Challenge/Onyx in around 1995, for 21000 UKP.

Ian.

I found this handy site:

http://www.xe.com/ict/

ramq writes:
> I have, for whatever reason, a copy of an old invoice from SGI Stockholm dated 12th nov 1999:

At that time, 1 UKP = 13.5 SEK, while $1 US = 8.38 SEK.


> [*] HU-2P300 - Upgrade to 2x300MHz R12000 for Octane: 105.500 SEK

(7814 UKP, $12589 US)

Ouch!


> [*] CR-2CPU-175/195 - Credit for return of dual R10k-175/195 CPU: -19.200 SEK

(1422 UKP, $2291 US)

Heh, and to think I gave some away for nothing recently...


> [*] WT5-1P300SE9 - Octane/SE, R12000 300MHz, 256MB, 9GB, monitor: 191.700 SEK

(14200 UKP, $22876 US)

And that was just an SE? Yikes.


> [*] SC4-AWE-6.5 - IRIX 6.5 Advanced Workstation Enviroment (O2, Octane, Onyx2): 3.400 SEK[/list]

(252 UKP, $406 US)

That's strangely cheap. In the UK 6.5 cost 2X more (500 UKP + tax).


> For whatever reason there were no purchase that time, apart from the Diff-SCSI HBA. I think I recall that the
> department who needed the bits thought it was too expensive...

Hard to argue with that. 1999 was about the time x86 CPUs really started to edge ahead. The PIII/700 offered
only slighty lower SPECfp95, but almost 2X better SPECint95 than the R12K/300. SGI did release faster CPUs
of course, but by then the gap was significant even for fp, eg.:

R12K/400: http://www.spec.org/cpu95/results/res20 ... 04235.html
PIII/1GHz: http://www.spec.org/cpu95/results/res20 ... 04136.html

On the other hand, the gfx in the test PC sucks (heh), and the bandwidth in the Octane still offered an edge. Around 1999
I talked to a guy at Chevron/Nigeria, he said their Octane SI/texture systems blew away PCs completely for processing
with their volumetric data (750MB+ minimum typical dataset size, crunched in less than 3 seconds), so presumably the
higher costs of SGIs was still worthwhile. They had an 18-CPU POWER Challenge for processing the larger datasets.

Ian.

Never ceases to amaze me how much extra RAM costs in original pricing.

Low RAM was a problem I ran into with the SGIs I adminned, both at the university in Preston (18 x R4600PC/133 Indys with 32MB)
and Salford University some years later (12 x R5K/180SC O2s with 128MB). I think resellers did such deals because it enabled them
to sell more seats a typical example of what I meant in other threads about bad PR/sales messing things up. Tiny disks were also a
problem (549MB in the Indys, 2GB in the O2s). In both cases I was eventually able to get *some* systems upgraded, but by then it
was too late. And of course, the Indys were only 8bit (ugh). Students were amazed at how much better a 24bit Indy with R4K/200
and 4GB disk was, ie. able to install all the sw instead of NFS-mounting everything, images looked much nicer, way quicker, etc.

Hmm, some of the prices in that PDF look kinda wierd. How come the R10K/195 I2 Max is only $1K more than the R4K/250 I2 Solid?
The product codes suggest the I2 is a Max aswell.

Oh, and I think the extra R10K in the Octane might be for a dual-175, judging by the product code.

Ian.

bri3d wrote: or SGI's sales reps very bad at demonstrating the 24-bit vs 8-bit - 8-bit is not only ugly but slower in most apps due to dithering.

I think it was just a case of selling a larger number of lower spec systems, thus satisfying the academic requirement for the
number of seats, ie. reps didn't want to say, sorry, the cost of *decent* systems would be too high if you wanted 20 of them.

The dithering is a hardware function via the CLUTs, AFAIK it has no impact on speed (check my Inventor results). For 3D
speed on Indy XL systems, the main CPU is much more important.

Ian.

Megatron-UK wrote: Why can't I find SGI kit as cheap as that, here in the UK?

'Cos I ain't had time to list anything on eBid yet. ;D

Btw, there's no point using the hinv command in the PROM to identify a VPro card, it doesn't
give enough detail. Use the 'system' command instead, and take a note with of the various
possible gfx board SGI part numbers.

Ian.