Slashdot Mirror
Sun's new UltraSPARC workstation: the Blade 1500
Roman Hauptmann writes "Here's a review of Sun's newest single-CPU workstation based on the UltraSPARC IIIi processor. According to the review, the system barely performs on the level of a P4 1.8ghz machine yet it sells for several times the price. Despite that, the Blade series still brings value to those who do visualization and imaging."
I'd hope that, for $3-4k, they could do a bit better than an 80GB (2MB cache) Seagate drive. Do "those who do visualization and imaging" really not care about the performance of their storage?
I've never yet seen a machine which skimps on its essential components justify its price tag. No surprise here.
Maybe on sheer performance it will be beaten by x86 however for crunching big data sets the UltraSparc is just more effecient.
----------
If by "efficient" you mean "more instructions per clock" than yes, UltraSPARC is more efficient. But workstation people really don't care about efficiency. They care about total instructions executed per second. And x86 machines have the upper hand here.
There are lots of advantages to Sun hardware generally, but this machine doesn't seem to have those:
- Sun machines usually have high-quality SCSI disk drives. This machine has a standard PC IDE drive.
- Sun machines usually have support for many CPUs. This machine supports one.
- Sun machines usually have insane memory bandwidth. This machine has less bandwidth than a P4.
- Sun machines usually have extensive I/O capabilities. This machine has your standard 64/66 PCI slots.
A deep unwavering belief is a sure sign you're missing something...
Hopefully anyone who made the mistake of a Blade 1000 will stay far away. Performance from Sun workstations has been sub-par for years now.
I had a good laugh when one of my Intel workstations and a colleague's Blade 1000 were both hooked up to a compute grid. The benchmarks for BLAST, the bioinformatics tool we were running on the grid, showed my PIII running circles around the bioinformatics geek's favorite machine. What's better is that the Intel machine (an IBM), was bought new for less than $1000, and the Blade had been purchased for over $5000!
I think the SPARC IV is due sometime in the next few months (probably just for the big iron for 6 months or so), if I recall correctly it's largely a dual core SPARC III with more incremental improvements. There is at least speculation that SUN will offer an Opteron based workstation in addition to the already announced entry-level server. I think there is development on a SPARC V, Fujitsu seems to be having better luck with their SPAEC implementations currently. There are also rumors that a bigger partnership will develop between the two firm's development.
Degaussing scares the bad magnetism out of the monitor and fills it with good karma.
It only got 1MB Cache and a rather slow harddisk. And it can't take more then 4GB ram so I really can't imagine what kind of task it would be good at.
It does have a nice 3D card but 3d is one of the things that really DO require number chrunching, so putting the Wildcat in PC with the fastest Pentium IV/Athlon would give a faster and cheeper system.
The only use for this system as far as I can see is for people who need to run Solaris and for some reason can't run it on intel.
Commodore did an "Bridge Card" for the Amiga 2000 way back in 1986-1987. Both 8088 and 80286 and it "bridged" the Amiga's Zorro slots and the included ISA slots, allowing the use of both Amiga and PC hardware.
This concept has been around for a while, this is just a refinement.
Learning HOW to think is more important than learning WHAT to think.
RTFA. It has one CPU socket, and I've heard a maximum of 2.0GB RAM. Also, I'm not having any problems with less gigahertz - keep in mind, I'm pushing the Pentium M, which has a very high IPC as compared to the P4. I'm saying that a rig that performs like a P4 1.8 and costs $5K is a total ripoff. Sure, it has a great video card, but I'd like to take a Blade 1500 Light, and take an Athlon 64 3000+ (which is used for two reasons: "I'm cheap, but my dick is still longer than yours", and it's a cheap way of having a CPU that can handle 64-bit apps when they become available) with a Radeon 9800 Pro, 512MB of whatever the best RAM for that system is, a nice fast HDD (maybe SATA, just to make it unfair), a Plextor DVD +/- RW, etc., etc., and find out how much it costs, and if the US3i is blown out of the water (if a 3000+ can kill a P4EE, and a P4EE, by nature, can kill a P41.8, it's kinda obvious), and do the same on video card (3d rendering tests, maybe?).
The problem with Sun is that it's three times more expensive and three times slower. We would spend $60k and get a whopping two new Sun servers. Then all the engineers would start throwing jobs at it and it would be dog slow again. Do you know how many Linux machines we could have bought for that much?
Primarily we need computers for raw number-crunching (big simulations) and large memory (big circuits). Linux can handle these just fine, and it's frustrating when other groups blow a load of cash on more Sun equipment.
Komi
The ultimate goal of science is to unify all forces of nature to a single law that can be silk-screened onto a T-shirt.
Okay, I know this is feeding the trolls and such, but I knew this issue would come up.
I did ask Sun, not only for benchmarks that they used for testing, but at very least for results that they'd gotten from their SPEC benchmarks that everybody runs. I waited, re-requested and did not receive them.
The reason why SPEC ViewPerf wouldn't install was because of a problem with GCC that I couldn't figure out and couldn't get from Google. Since it wasn't an issue with Solaris 8 (well, sort of) and wasn't an issue with the hardware, I didn't publish anything that I couldn't verify personally. If you feel that's poor journalism then, quite frankly, you don't belong on the Internet.
The Blade 1500 has been for sale since November. It's completely unreasonable to assume that only I had access to it...
-JemBriefly stated, you're wrong. I'm a sysadmin for both Solaris and Linux, and trust me, both have their place in the Enterprise. When you're looking for a good overall performer, and speed is more important than overall efficiency, Linux is GREAT, especially for webservers and other similar tasks. When you're talking about applications such as Oracle, you need the big iron that Sun can deliver pushing the envelope of performance on more robust systems.
Personally, I'm going to be getting a 1500 or 2500 in the next few weeks at work (still haven't decided which to buy). I have a SGI Indigo2, an Ultra 1, a few x86 based machines, an AIX server, HP-UX server, and a microVAX. Each has one or two things that they're good for (like, only the SGI or x86 systems make good desktops), but together you start to see why each flavor of *NIX has its own quirks, and value. Each job has a tool best suited for it, and x86 / Linux / BSD isn't always the right answer.
Clinton made me a Republican. Bush made me a Libertarian. Trump is making me question reality.
Sounds like Apple's undercutting Sun Also,
Dual 2.0 Ghz G5 $2,999
64bit, up to 8gigs of ram, 160GB SATA, 1Ghz frontside bus.
Spend the extra 2grand on ram, or hopefully a mini Xserve RAID some day.
Warm regards indeed; pleased to meet you and sorry about the troll comment.
I didn't just fall off of the silicon truck -- I've written reviews of Sun products before and I'm working on one more right now. I gave Sun several days to read the article before it was posted. This gives them a chance to correct any major mistakes that I might have made, and it also gives them a chance to respond if they feel I've been unfair. Then I wrote one last warning saying I was going to publish it if I hadn't heard back within another day.
I have a pretty good relationship with Sun, and I don't feel that the article was at all negative or unfair... and if they did, they had every opportunity to work with me to change anything biased or factually incorrect. And if they hated the review, why did they post it in their Press section? I don't think I've been unfair with them at all; it seems that they don't feel that way either.
In regards to the benchmarking tests, it was my guess that they only wanted to show that it was faster than the Blade 150 and didn't care about much else, or perhaps they didn't have anything to send me. Their primary target with the Blade 1500 is customers who already have a Blade 150. Benchmarking is just gravy anyway; I value a good review with a few pictures over a poor review with lots of graphs any day. That's what makes my site unique among review sites. Anyway, all that potential customers (readers, in other words) really want to know is that the Blade 1500 is twice as fast CPU-wise as the Blade 150 and there is no need to change software when upgrading. In the workstation market that's a tremendous value, even if it seems trivial to us desktop users.
Sun's DBX debugger is an excellent tool (GDB doesn't give you the fork-following options that DBX supports). AFAIK, it only runs on Solaris. Very useful for development of software that requires to fork.
Follow me
There are lots of advantages to Sun hardware generally, but this machine doesn't seem to have those:
- Sun machines usually have high-quality SCSI disk drives. This machine has a standard PC IDE drive.
- Sun machines usually have support for many CPUs. This machine supports one.
- Sun machines usually have insane memory bandwidth. This machine has less bandwidth than a P4.
- Sun machines usually have extensive I/O capabilities. This machine has your standard 64/66 PCI slots.
You forgot to mention that Sun USED to manufacture their own machines. Now they have Acer Computers do it for them (literally!).
If "disco" means "I learn" in Latin, does "discothèque" mean "I learn technology"?
I just installed 3 Blade1500 workstations. We run a legacy medical PACS system that is based on Sun boxes. We are running anywhere from Sparc 4s to the Sunblade range. We are currently using the Blades to drive 4 three megapixel x 10 bit Dome monitors. They work great in that application, and that is what our software runs on. The vendor that we have our PACS system with is moving to a PC/Linux platform, but for the legacy software we run now, the Blades offer a lot of bang for the buck.
BTW, the build quality of the machines is to the usual high Sun standard. I like the looks of them as well.
"He's lost in a 'floyd hole"
I've been in the same situation. I've had Sun replace everything but the power supply on a machine and still have it experience inexplicable lockups. In that time I did not get the impression that Sun support knew what the hell they were doing.
Could you provide hard evidence of UltraSPARC systems beating comparably priced Athlon64 or Opteron systems for large data set problems? There are a lot of people in this discussion regurgitating that old chestnut. While it might have been true 5 years ago comparing an UltraSPARC workstation to a 32-bit Pentium III system with a constipated little 133MHz bus, times have most definitely changed. Show me solid benchmarks of a Blade 1500 beating out an Athlon64/Opteron system of the same price, and I'll happily run out and tell everyone that they should be buying an UltraSPARC for their next workstation.
Also some software only runs on Solaris so for that this box is good
Ahh NOW we're getting somewhere. Agreed. People who can only run their apps on SPARC/Solaris are locked into the platform, and have no choice. Kind of like all the graphic designers and desktop publishers who were locked into the Mac until serious ports started showing up on the PC. So what's going to happen when the inevitable happens and your specialty app appears for x86/Linux?
Sure, the hardware cost might only be 1% of the software cost for some specialised applications that (for now) only run on SPARC. But why would anyone choose to run their app on a slower, single-vendor proprietary plaform when a faster, open one is available?
They aren't going to, and we can all look forward to seeing the end of Sun's anachronistic SPARC workstation line in the near future.
Uhhhh...technical workstations have traditionally been used for stuff like large scale CAD and industrial design work, complex graphic visualizations and mathematical modelling. The traditional realm of the 'workstation' (before the term was highjacked by every x86 vendor with a minitower case and a 3 button mouse) was CPU, memory and graphics intensive work that would normally make a 'kiddie's game box' break down and cry. Having the ability to have a lot of RAM in a workstation is a key feature.
The technical workstation market was how Sun got started. The Blade 1500 is a very poor excuse for a technical workstation, yet it is priced like one. It's lack of RAM capacity is yet another illustration of this point.
Do you expect this sort of reliability from Dell? Your applications may be more suited for linux now, but there are still tasks which we run in VLSI design. Which require a lot of CPU cache. The MHz is not really important. All that matters is the consistant 100% CPU utilization for 6 months on a 4 CPU 750MHz machine and not a single powerdown, and no faliures. Try running a Pentium or Athalon 6 months flat out.
Comparing Sun to an Intel is like apples to oranges. Both are for a different purposeMy Aurora : http://www.youtube.com/watch?v=o91ZsGwJYyg
FB : https://www.facebook.com/TanveersPhotography
Even more so, Sparc is an actual IEEE standard...
"Also, I don't have a lot of respect for someone who claims that WD hard drives are of better quality than Seagate drives. As someone who has killed every WD drive I've ever seen in a real workstation environment, I'll take a Seagate any day, even if it is IDE."
Ever since hard drives were sold for PC's, people have claimed one brand is more reliable than another.
That's not true. For years, people have derided Seagate as "Seacrates", but really, different lines of HD's have different levels of reliability (i.e. the first Barracudes had a 100% failure rate within 2 years), but that is meaningless across an entire product line, and people who avoid entire manufacturers of HD's are simply using anecdotal information as if it were a statistical fact.
Stop it. Be a computer *scientist*.
... than the stability and scaling already mentioned. The NFS implementation is still superior to that of Linux, particularly in areas of caching (works out of the box) and throughput on busy networks - although it's certainly true that the Linux implementation is improving.
Oh for fuck's sake not the "RISC vs. CISC" crapola yet again.
There have been thousands and thousands of posts here to debunk this obsolete myth. But apparently not enough for the diehards.
Mainstream x86 processors have been hybrids (sorta CISC outside, kinda RISC inside) for years. And "pure RISC" is more an academic ideal than reality.
And what is this cycle that is not a clock cycle? Mmm? [Perhaps "more instructions per given task" was intended, to compare to "more clock cycles per given instruction"; this is how the usual zealot comparison of "RISC" vs. some x86 implementation of younder goes, anyway.]
Explaining the fallacy in the "bandwidth for more fetch operations" argument is left as an exercise for other readers. (Clue: your CPU arch hardly dictates how much input your op takes. How come you fetch more data in a RISC -- or any other -- environment?)
And the "C" stands for "Computing", not "Computer".
What is interesting is who was the dingbat that modded this junk "Insightful"...