Slashdot Mirror
Sun's New Workstations and Graphics Cards
An anonymous reader "Sun Microsystems has released the Sun Blade 2000 workstation, along with a new graphics accelerator, the XVR-1000. This could very well give SGI's lineup a run for its money in the CAD and Visualization fields, although its fillrate and 38-bit colour may make it less desirable for animation. Make sure to check out Ace's article. " (page down
a couple times to read it)
http://news.com.com/2100-1001-860701.html
http://www.sun.com/2002-0314/feature/
The system ships with a 73gb fibrechannel harddisc, 900 or 1.05 UltraSparcIII (dual capable), and a gig of ram. nice box. It sets a world record in workstation performance (halfway down the press-release).
Direct link to the post as a stand-alone page.
I have Sunblade 100, from which I write this comment. From my experience, this machine is by a factor 4 (yes, four) time slower than a new Athlon XP 1.9... And it costs much much more.
If it wasn't for endianness compatibility with existing binary data, I wouldn't be using it.
Ford Motor owns about 7000 Suns, and still buys them. PCs just don't have the applications that CAD/CAM desisgners need to get real work done. There are some big software packages ported to Windows, like I-DEAS, Unigraphics, and Catia, but the whole workflow and ancillary apps are non-existant.
What do you mean 38 bit color makes it less desirable for animation?
That is just wrong. This has 10 bits per component RGB. Typically that's more than enough. In addition animation apps like Maya tend to be geometry and state limited not fill limited.
Ofcourse the tag 'animation' is a bit to vague to mean anything in the first place.
Well done Sun, this should cause SGI some pain, but I'd say more because it gives the impression that Sun is doing something interesting where SGI hasn't done anything genuinely interesting in a LONG time.
As for the color, does anyone know if you can actually see any difference there ? I mean - 24 bit color is 16M colors ?
It's not about the number of colors, or whether you can see the difference. You want more bits of color precision for handling multiple lighting/shading/blending/etc. ops that happen throughout the rendering pipe, before the end result's precision is scaled down and displayed.
For example, when adding more and more lights to a scene, you will eventually start clipping against those 24 bits of precision.
I'd like to see 128 bpp internal rendering pipes and 128 bit Z buffers. It would take a lot to exhaust that kind of precision.
Well we have a number of older Sunblades (1000's) and Ultra 60's (also workstation class) that we use for chip design. When trying to route some of our bigger asic's we use all 8GB's of ram in the 1000's. Show me an Intel workstsation that can handle 8GB of ram. Since these runs typically take days having even a single crash is unacceptable, and yes I know about checkpointing but afaik the tools from the chip factories don't do it, and even if they did that's a lame answer. For the most part it's about stability, and memory addressing not about raw cpu power (though since the jobs take days more cpu power is always apreciated =)
There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
>> although its fillrate and 38-bit colour may
.plan once.
.plan about this...
>> make it less desirable for animation
I'm not sure what that's supposed to mean. The fillrate is just fine for a workstation, games generally are the only programs that need a high fillrate, memory bandwidth and size, and of course T&L are *much* more important. The 38 bit internal color is excellent, nicely comparing to SGI ( http://www.sgi.com/workstations/comparison.html ), and unmatched by 3dlabs. The bit-depth of the graphics card has nothing to do with the color rendering accuracy, which is usually 48 or 64 bits for high end stuff. Games really need high bit depth precision for multitexturing, which multiplies color errors. I think Carmack mentioned this in a
Nvidia will probably have 64 bit color in NV30, and 3dfx's rampage was supposed to have 52 bit color ( http://www.digit-life.com/articles/3dfxtribute/ ) Games start needing high bit depths when you have massive multi-texturing, which tends to multiply errors. I think Carmack had a
As for raw compute performance, if you believe Sun's SPEC ratings from their product site, a 1.05GHz SPARC CPU is only just lagging behind an Intel 2.2GHz PIV on integer performance and beating it on FP. As FP is what drive 90% of scientific applications, Intel hasn't got the SPARC beaten yet by a long shot (especially since you can get a 106-way SPARC box, but Intel is limited to 32-way last I heard).
It's probably also worth noting that list price is rarely what a company will end up paying.
Very interresting If they Pull it out.
--- Metamoderating abusive downgraders since my 300th post.
Doh. That would be everythingunix.ORG. Or go straight to the article here.
Yep, you have to have the right tool for the job.
I also do chip design and we have been using Xeon with Linux for many applications. For cell level hspice and block level synthesis you can't beat the speed of the PCs.
For top level jobs like extraction we need the ram.
One should not theorize before one has data. -Sherlock Holmes-
Does 10 or 12 bits really make a difference over 8 bits? Of course it does. Most film work these days is rendered in either 12 bits, 10 bits logarithmic, or 16 bits. Think about it: in a dark movie theatre room, 256 levels of grey (for instance) is not a lot. And if that doesn't convince you, think about image manipulation: after a few multiplications and compositions, you'll end up with very little color resolution with 8 bits. And yes, these things are often done in hardware in the color buffer (eg flame).
As for PCs, NOBODY's doing large model work on them. Small shops might use them because they're economical, but no one would use a PC to work with multi-thousand surface/100k+ element geometry/FEM. Perhaps this is a Windows limitation, not hardware architecture; it's hard to tell because most of the big 4 (Catia/ProE/Unigraphics/Ideas) don't have a Linux port yet, AFAIK.
You really need to work on your understanding of "high end"; Sun's high-end is boxes like the E10K and E15K -- and it's an area where Intel has no leverage. An E15K can support multiple hardware domains, up to 106 US3 900MHz CPUs, and over a half *terabyte* of RAM.
You find me an Intel machine with those specs. Oh, and it must be fully managable from a remote site down to the hardware level; you have to be able to turn CPUs on and off, power the machine up and down, re-assign drive IDs, and such -- remotely.
The eight-way xSeries competes more with Sun's low-end server hardware, which is comprable in price; I can't really give an exact figure without knowing what this server is for.
--
I Hit the Karma Cap, and All I Got Was This Lousy
it matters for print.
it is difficult to get screen colors to match printed colors. they simply use different color space. (although you can simulate cmyk with rgb somewhat). and differences are easier to see in print. in 24 bit color there are 16M colors, but only 8 bits (256) of variation for each of the 3 primaries. this also one of the reasons that many scanners and printers are capable of more than 24 bit color.
Then there's the alpha (transparency) that isn't considered at all in 24 bit RGB. so that matters, too.
you probably shouldn't have read this.
sgi going bankrupt? check the papers dude, who made profits recently? sgi did while sun's diving in the red... don't say things w/o checking your sources dude..! with all the .bombs, I guess sun's stuck with so much inventory...
No innovative products? check their products webpage, they have stuff no one else can even close!
Ace is wrong about one thing:
Currently, the XVR-1000 targets primarily the engineering and CAD markets, as opposed to 3D animation, given the rather limited fillrate. However, Sun intends to use the MAJC-5200 to scale the performance of its graphics solutions to higher levels in the future (as seen in this older roadmap), so we may yet see a solution attacking the 3D animation market at some point in the future.
The MAJC-5200 will improve geometry performance (number of triangles, floating point math required), not fillrate (number of pixels/texels shaded, integer math).
Animation requires better fillrate, and more MAJC-5200s won't provide that. MAJC-5200 *will* provide Sun with stronger geometry performance (FLOPS, remember?), which is just what Sun's core engineering and CAD markets most want. Lots of small triangles to accurately show the precise shape of things of digitally-created parts. Nothing about MAJC-5200 will strengthen Sun's penetration into new SGI markets per se. That'd be dependent on some other, presumably fill-rate enhancing, technology.
--LP
When you've got some serious number crunching to do, a PC is lame.
That was true up to the late 1990's. Today, it is not. A P4 blows away a SunBlade 1000 for both integer and floating point number crunching. In fact, go check the SPEC results. P4 is faster than any Sun workstation. If you want to beat a P4, you need to be talking to IBM, not Sun.
The DEC Alpha has blown away SUN, IBM, SGI, HP etc in SPEC benchmarks for years, and look where it got them. Their market share was always lagging behind, even with the much faster hardware.
http://spamdecoy.net - free throwaway anonymous email - avoid spam!