So I have a SunFire T2000 - what now?

Mr. Baz

2[H]4U
Joined
Aug 17, 2001
Messages
2,815
Inherited SunFire T2000 server for free. I hate Solaris with a passion, so I'm in the process of netboot installing Debian for SPARC processors. I figure an 8-core processor with 32 threads should be useful for something. It also has 4 x 146GB 10k RPM SAS drives. Unfortunately the RAID controller only works with striping or mirroring. I figured I'll stripe all 4 drives just for the speed. I don't need redundancy.

Question: What am I going to use this for?

I was thinking about offloading rendering tasks on this server? (I use SolidWorks mainly. Might start using Blender and some others).

I use windows machines for workstations and servers already. This server I thought would make a great R&D server since every other core tasks are already spoken for.
 
Last edited:
Not much. It's quite old and quite slow. It's as old as the Pentium 4 and can only do a single floating point operation at once (not 8), so it would be terrible for rendering. A modern Celeron would best it by a large margin. Seeing as they go for all of $100 on eBay, I wouldn't bother with it really. Recycle it somewhere.
 
Not much. It's quite old and quite slow. It's as old as the Pentium 4 and can only do a single floating point operation at once (not 8), so it would be terrible for rendering. A modern Celeron would best it by a large margin. Seeing as they go for all of $100 on eBay, I wouldn't bother with it really. Recycle it somewhere.

Dang. Forgot about that. Maybe I'll just through apache, mysql, and php on there for some tinkering.
 
those sparc chips are great for databases (with LOTS of users).
 
those sparc chips are great for databases (with LOTS of users).

Maybe for 2000s applications and databases, but they wouldn't hold up so well today.
When an Ivy Bridge Celeron from 2013 can decimate these processors, that just kind of shows, unfortunately, how obsolete they really are.

While systems like this are super fun to get working, especially to have them do something modern, it isn't a realistic goal to have them work in full production, either in home or enterprise, as their power consumption, heat output, and noise is massive, especially when compared to a modern x86 system which has a tiny fraction of all of the above while doing more and having more features.

To put this in perspective, I had an Apple G5 Quad with dual 970MP PowerPC CPUs @ 2.5GHz with 16GB DDR2 ECC memory, running either OS X 10.5.8 or Debian 7.1 PPC.
With the ODROID-U3 I have in my sig running a wimpy ARM Cortex-A9, I was able to transcode audio and video, as well as other processes and services, at about 1.1x the speed of that G5 Quad.

The G5 Quad, under load, used about 600-700 watts from the wall (the heat output was ungodly, even in the winter), where as the ODROID-U3 uses no more than 15 watts from the wall.
Again, these old systems are truly fun to work with, and especially push to their limits with modern applications, but beyond that, they are just hobbyist systems now. ;)

@ OP: If you do get it up and running, share some pics, they would be a lot of fun to see for everyone!
 
How can A9 that score 2750 in Geekbench beat G5 with 3700 points?
How can 2.5 DMIPS/MHz times 1700MHz beat 2.9 DMIPS/MHz times 2500MHz?
Better optimizations are one thing but difference in raw power is too great for Odroid U3 to come faster. In some benchmarks made in web browser like Chrome it might be true (due to lack of JIT on PowerPC) but not in normal programs.
I would suspect Odroid XU3/XU4 with Cortex A15 at 2GHz to be be slightly faster than this G5 but not A9 @ 1.7GHz. A9 being faster would not make any sense.

That being said I agree with your point of new computers being much faster using fraction of power. Espeically if we take single core performance which is very important in many situations and which is rather bad on those kind of server processors.

And I definitely agree it would be awesome to see some photos and benchmarks of such PC. I would love to see results of full Phoronix suite benchmarks.
 
How can A9 that score 2750 in Geekbench beat G5 with 3700 points?
How can 2.5 DMIPS/MHz times 1700MHz beat 2.9 DMIPS/MHz times 2500MHz?
Better optimizations are one thing but difference in raw power is too great for Odroid U3 to come faster. In some benchmarks made in web browser like Chrome it might be true (due to lack of JIT on PowerPC) but not in normal programs.
I would suspect Odroid XU3/XU4 with Cortex A15 at 2GHz to be be slightly faster than this G5 but not A9 @ 1.7GHz. A9 being faster would not make any sense.

That being said I agree with your point of new computers being much faster using fraction of power. Espeically if we take single core performance which is very important in many situations and which is rather bad on those kind of server processors.

And I definitely agree it would be awesome to see some photos and benchmarks of such PC. I would love to see results of full Phoronix suite benchmarks.

I don't see how synthetic benchmarks can really compare two completely different CPU architectures.
What is it comparing? FPU? Instructions? ALU/integer?

I'm using real world programs in order to make the comparisons.
A 720p video file, being transcoded on the G5, would not transcode in real-time, and would transcode about 7-8 frames for every 10 frames viewed in real-time.

The same video file, being transcoded on the ODROID-U3, would transcode around 11-12 frames for every 10 frames viewed in real-time, meaning that the video could be watched without stopping to buffer.

The G5 had a gigabit connection and two SATA HDDs in RAID1 to work off of.
The ODROID-U3 has a 100Mbps connection and a USB 2.0 connection to two HDDs to work off of, yet it was still capable of more.

Transcoding FLAC to MP3, using the same files on both, was also faster by a bit on the ODROID-U3.
This is why I only trust synthetic benchmark tools when comparing two CPUs from the same architecture.

MIPS means almost nothing, and is a superficial way of benchmarking, especially when comparing two different architectures with different instructions which function very differently, regardless of how many operate per second.

Another thing, the G5 has TWO 970MP CPUs (each is a dual-core), making the A9 quad-core much faster than either 970MP on their own, again, using a tiny fraction of the electricity, let alone heat output and noise from the cooling solutions.

I would suspect Odroid XU3/XU4 with Cortex A15 at 2GHz to be be slightly faster than this G5 but not A9 @ 1.7GHz. A9 being faster would not make any sense.
I don't see why.
Clock speed has little to do with it, unless you are comparing apples-to-apples, which this is not.
An Intel Core i5 2500k @ 1GHz will run circles around those two 970MPs @ 2.5GHz, yet it isn't even clocked half as fast.

The A15 @ 2GHz would be closer to the A9 @ 2.7-3Ghz, as the A9 @ 2GHz wouldn't be as fast.
I'm not trying to brag about how amazing the ODROID-U3 is (it really isn't), but just to use it as an example of how even a very low-power system today is much faster and more efficient than a very high-end, high-power system was just a decade ago, let alone longer ago. ;)
 
The answer to that question is built in hardware encoding/decoding on the SoC (assumption) to save power. That's something that G5 system doesn't have.
 
The answer to that question is built in hardware encoding/decoding on the SoC (assumption) to save power. That's something that G5 system doesn't have.

The video is being served to, and is played, on another system, not the ODROID-U3 or the G5 themselves.
They are just transcoding and sending the data to be viewed on another system.

What hardware encoding/decoding are you talking about?
These systems aren't using their respective GPUs for anything as they both run headless, and the transcoding software is strictly CPU-based.
 
unlike Sandy Bridge this Cortex A9 is not faster than G5 but slower so it should be clocked higher than G5 to beat it. Cortex A9 is rather comparable to Atoms (these before Silvermont) and NetBurst and G5 is comparable with AMD K7/K8 in performance per clock. With your posts its almost like you were suggesting that 1.7GHz Pentium 4 can beat 2.5GHz Athlon, just ridiculous in itself :eek:

If you get such results if all programs I would rather try to find what is wrong with yout G5 than post Cortex A9 is faster than PowerPC 970MP. There can be number of issues that hamper performance of this G5 machine from not detecting second CPU to thermal throttling.

If software you use is very badly optimized for PowerPC AltiVec and have good NEON support then it should be noted and you should not make A9 as example of low power chip beating high power old chip for which you should take some A15, A57 or some Intel NUC board or something and use A9 thing as exmple of software optimalization, that even slower CPU can perform much better in normal applications than old less popular processors due to better optimalizations for popular ISA.
 
With your posts its almost like you were suggesting that 1.7GHz Pentium 4 can beat 2.5GHz Athlon, just ridiculous in itself :eek:
I never said anything remotely similar to this.
Not sure where you are getting this info from.

If you get such results if all programs I would rather try to find what is wrong with yout G5 than post Cortex A9 is faster than PowerPC 970MP. There can be number of issues that hamper performance of this G5 machine from not detecting second CPU to thermal throttling.
There was no thermal throttling on those CPUs under either OS, I monitored everything the system was doing when transcoding.
Both CPUs were detected and were fully utilized; running top in either OS X or Debian showed between 370-398 (400 is all four cores @ 100%) for ffmpeg, among others when used separately.

If software you use is very badly optimized for PowerPC AltiVec and have good NEON support then it should be noted and you should not make A9 as example of low power chip beating high power old chip for which you should take some A15, A57 or some Intel NUC board or something and use A9 thing as exmple of software optimalization, that even slower CPU can perform much better in normal applications than old less popular processors due to better optimalizations for popular ISA.
You might be right about the software being more optimized for NEON than for Altivec, I won't deny that as I am not one of the designers of the software used.
However, that further shows why I trust synthetic benchmarks so little, unless they explicitly state which areas they are testing/scoring, not just presenting a "generalized" score.

The FPUs wre definitely at work (not software libraries) on the ODROID-U3 as Debian is armhf compiled and not armel.
I will look into this further, but I do agree with the A15 (quad-core) certainly beating dual 970MPs, hands down and optimized or not, as it is certainly a good jump up from the A9.
 
The video is being served to, and is played, on another system, not the ODROID-U3 or the G5 themselves.
They are just transcoding and sending the data to be viewed on another system

I have two words for you:

Quick Sync.

Accelerates both decode and re-encode stages SIMULTANEOUSLY , allowing you to trans-code files without using the CPU.

Is also available on Bay Trail. The trans-code bandwidth is 3MB/s for Bay Trail. Since a typically 1080p stream is only 10Mbit/s or less, it has more than enough bandwidth. Even the Jaguar part can handle real-time trans-code.

tC8mmJB.png


http://www.sisoftware.co.uk/?d=qa&f=gpu_byt

If you think a modern A9 SoC doesn't have such a feature, you're in denial. Intel added a feature like that to compete with both ARM and GP-GPU accelerated encoders.

In fact, I did some research, and the ODROID supports Samsung's MFC unit for video encode/decode:

http://linux-exynos.org/wiki/MFC

If both AMD's Jaguar and Intel's Bay Trail support hardware trans-code, you can be sure it''s on ARM platforms (that are much less powerful).;
 
Last edited:
I'm having to rework the entire network infrastructure right now, but once I get an opportunity to play with the Sparc server I will. I'll post some pics/numbers when I get a chance.
 
While this was a beast at it's time, that's not so much the case now. A good way to compare them to other CPU's is transistor count (yes it's not perfect but what is?). The T2000 has 279 Million. That isn't much compared to a modern Xeon E5 (5,560 Million) or a Spark M7 (10 Billion).

If you can get hold of few M7's (and a few terabytes of memory), THEN you can have some fun (if you're that way inclined).
 
Last edited:
I'd advise redundancy on the drives. I've replaced a ton of failed drives in those old T1000/T2000's.
 
While systems like this are super fun to get working, especially to have them do something modern, it isn't a realistic goal to have them work in full production, either in home or enterprise, as their power consumption, heat output, and noise is massive, especially when compared to a modern x86 system which has a tiny fraction of all of the above while doing more and having more features.

This is really the bottom line. While an interesting novelty item, a machine like the T2000 is useless. It makes no sense to run one 24/7, or even for an hour or two on a regular basis, when far more powerful and energy efficient alternatives can be had for cheap.

Even if I were given one and I had a use for it, I'd be better off spending $200 on a NUC-like device which can do more and would pay for itself in energy savings before too long. That T2000 consumes about 325W while operational. That makes it horribly inefficient.
 
You can't run LDOMS 3.x on it, you can't run Solaris 11 as the control domain. The fan units in the ass end of the thing like to die grizzly deaths. It's about time for the EEPROM to die on those as well.

It's hardly even useful as a learning device: Due to its age, there's no SPARC-specific stuff to be learned on that gear that is particularly applicable in a modern SPARC environment, or that you can't quickly learn in a modern environment.

When it comes to the software entitlements you get, all of them and then some are available to under the "development" terms of the Oracle Solaris licensing, and you can run those items under an x86 VM considerably faster and with less power consumption.

Recycle it.
 
Last edited:
As much as I love old tech and retrocomputing, SPARC systems with their proprietary nature and battery-backed PROM chips (which completely die upon power loss), are just not very good systems for retro computing.
Not to mention, the older units from the late 1980s and early 1990s won't be lasting much longer due to said ancient batteries failing.

This is what made x86, m68k, and POWER/PowerPC devices so appealing back then, and even now for our retrocomputing needs.
SPARC was (and still is) just a bit too finicky, flaky, and fuggin' expensive! :D
 
As much as I love old tech and retrocomputing, SPARC systems with their proprietary nature and battery-backed PROM chips (which completely die upon power loss), are just not very good systems for retro computing.
Not to mention, the older units from the late 1980s and early 1990s won't be lasting much longer due to said ancient batteries failing.

For the older EEPROMs you can hack in a replaceable battery
https://www.youtube.com/watch?v=ql9xo_zhNy0

something as new as the T2000 has a replaceable battery
http://g03.a.alicdn.com/kf/HTB1V7dRIXXXXXckXpXXq6xXFXXXm/150-1204-SUN-T2000-Battery.jpg
 
This almost makes me want to dig out my old Sparc 10 with dual CPU's and get it up and running again... almost.

Perhaps make a Youtube video for it... just have to find a 13w3 to VGA adapter...
 
This almost makes me want to dig out my old Sparc 10 with dual CPU's and get it up and running again... almost.

Perhaps make a Youtube video for it... just have to find a 13w3 to VGA adapter...

Here you go!
Not too shabby for under $10. ;)
 
As an eBay Associate, HardForum may earn from qualifying purchases.
Back
Top