Slashdot Mirror
SGI launches R16000
nkrgovic writes " SGI has just launched a new CPU - the long expected R16000. The new CPU works on 700MHz, has 4MB secondary cache and more goodies.
For now the new CPU is only used in SGI's Fuel workstations, but we should expect to see it pretty soon in SGI's Origin servers as well. With new high density compute nodes this should make the Origin's the fastest supercomputing server per square foot."
So fast, it helped me get first post.
I'm confused. I thought SGI was dropping support on IRIX. Why are they releasing new Irix boxen?
Information doesn't want to be anthropomorphized anymore.
I thought enough material had finally invaded the net for people to realize Mhz means nothing... I guess I was wrong.
Let's play what if... cause I don't have any facts on this processor: What if the mov operation of said processor is 1 cycle, whereas mov of pentium is 7?...
Where does that put you?
Books are written on CPUs. pick one up, and you'll understand Mhz means nothing.
processor performance has never been SGI's strong point, except for breifly after the R3000 and R10000 were introduced.
SGI's workstation line is largely unimpressive, especially for the 99% case of computer users, hell, even engineers.
The problem is, for a small set of jobs, for a small set of people, nothing else is suficient - at any price. You're either using an SGI, or the work isn't taking place.
That market is continuing to erode, but i dont think it will ever dissolve completely. I think eventually SGi will effectively become a US govt subsidized entity. SGI continues to build the systems that only governments need and only government agencies can afford.
Clustering has nothing to do with the markets SGI sells in. Please don't mention it, it makes me think you don't know what you're talking about.
Do you ?
My opinions are my own, and do not necessarily represent those of my employer.
I'll be running my P4 at 3 Gigahertz, thankyouverymuch.
;)
True, but your architecture still sux
"Backups are for wimps. Real men upload their data to an FTP site and have everyone else mirror it." -- Linus Torvalds
I guess that makes it faster than my car. :)
Linux is only free if your time has no value. Windows is only free if you threaten to use Linux.
Sigh...
Come on people. You all root for the Athlon when it is clocked well under the P4, yet you believe that SGI's MIPS line is crap when it tops out at 700Mhz???
Sun's UltraSPARC III Cu tops out at 1.05Ghz last I checked. Does that mean that the P4 at 3Ghz stomps the hell out of it? If you said yes, you are a fucking idiot.
People, the Unix world is far far different from what you are used to in PC land. High speed backplanes, dedicated busses, huge amount of L1 cache, insane L2 cache, incredibly efficient cpu designs (where 1 clock per instruction is pretty much the norm and cache misses don't occur every 3 operations), hot swap damn-near-everything, upwards of 72 processors and 288 GB of RAM...
It all adds up to a fucking badass machine that smacks the piss out of any PC on the planet when it comes to getting its job done. Don't compare apples to oranges. The applications these machines are designed for do not include Quake 3. The benchmarks you have memorized don't mean a damn thing in this realm, so go back home.
Getting back to the article, I'm glad to see SGI coming out with a new CPU. I still see a few SGIs in the wild now and again. If they lock down Irix a bit more security wise and expand their target market, they might be a decent competitor for Sun within the next 10 years. I don't see them winning any shining star awards right off the bat, but if they are persistant they'll do alright in the long run.
You really cannot compare a 700Mhz MIPS chip to a 3000Mhz x86 p4.
You must remember, the R16000 is 64-bit, not 32-bit.
Also, it has 4000k of L2 cache, not 256k or 512k.
Also, out-of-order instruction execution, x86 chips can't do this.
you are trying to compare two things that are completely different.
Alcohol & calculus don't mix. Never drink & derive.
I worked all summer in an all-SGI shop.. And I call tell you how far behind they are. The place where I work is specialized in HPC, so when they started in 1992, SGI was probably a pretty good choice, but now for workstation, I wouldnt say its overkill, I would actually say that its underkill. We made a benchmark comparing an SGI Origin and a linux Ahtlon cluster, the athlon needed only two nodes to beat the origin and with all 16 nodes where about 10 times faster... SGIs are just overpriced, for 99.999% (that's 5 nines) PCs can do the job and even do it better and especially do it much cheaper. So their workstation market is being destroyed from under them.
On the other end, their HPC (super-computers) is being attacked from above. On that sector, price is not really a problem, its just pure performance. And there too they are being beaten, SGI just does not have the research power that
NEC or IBM can have. So they are starting to be pretty much behind, so they become not only more expensive (which does not really matter), but more importantly much slower...
Also on the workstation market, their desktop SUCKS, its just a pain to use. They are still stuck in the pre-win95 era... It might have been good compared to win3.1 or twm, but it just is not in the same world as GNOME, KDE, WinXP or MacOSX.
Also, their other strengh where there graphics board, they invented modern 3D hardware. And for a long time the roadmap for the PC 3d hardware was simple, they just had to do what SGI already had, but we have now passed a point where the PC hardware has actually more features then the SGI stuff. The only difference now between the pro and game markets are the amount of ram/cache and those "pro" cards exist on PCs. They do cost $ 2000-3000, but they are nowhere near the cost of the SGI workstation that includes them...
SGI has no future. They have been losing money for years. I have been thinking for quite a while that they where a good target for an acquisition, but now that MSFT has bought much of their patents. It might be cheaper to wait for them to go bankrupt and to pick up the pieces. They where in a fast playing game and they have gotten slow.....
Wha?
You, sir, are almost completely uninformed. The R16000 is an R10000 variant, just like the R12000 and R14000 before it. It is not a vector processor, and has no vector units. The R16000 is, furthermore, a desktop processor in its own right, because it's currently being used in the Fuel workstation.
Incidentally, SGI divested itself of Cray some time ago. Cray was bought by a company called Tera Computing, which then changed its name to Cray. They're building the SV2 vector supercomputer now, using their own processors, and they also have an arrangement with NEC to market the SX-6 in the United States with a Cray logo, but that's strictly a resale agreement.
I write in my journal
Instead of everybody saying "GHz doesn't matter, dummy" why doesn't somebody quote some real benchmarks? I poked around on the web a bit and all the benchmarks I can find either (1) are out of date, or (2) show Alpha, Intel and AMD blowing everybody else out of the water.
In my experience SGI's are slow but are extremely scalable. With IA32-based machines you'd be lucky to get 4 CPU's sharing memory, unlike the 64+ you get from SGI. Very good for scientific codes but not so hot for applications that are either not parallelizable at all, or embarassingly parallelizable such as Seti@Home or ray-tracing a feature film.
You must remember, the R16000 is 64-bit, not 32-bit.
For the record, the R10000 series can run either 32-bit or 64-bit code. All other things being equal, the 32-bit version of a program will run faster than the 64-bit version; you can fit more 32-bit ints into cache at once than 64-bit ints, so the 64-bit version of a program generally suffers more cache misses than its 32-bit counterpart.
On an SGI box, you don't compile for 64-bit unless you absolutely have to address more than 2 GB of virtual memory.
Also, it has 4000k of L2 cache, not 256k or 512k.
That's pretty puny for an SGI. The processors they use in the Origin servers have typically been equipped with 8 MB of secondary cache; the 4 MB version must be just for the workstations, to keep costs manageable.
you are trying to compare two things that are completely different.
On this point, however, you're 100% correct.
I write in my journal
I'm sorry to disapoint you.. but I have no problem agreing with you that the clock speed is not all... But its still important... On our CFD (Computation Flow Dynamics) the kind of thing that SGI super-computers are made to handle.. Our el-cheapo AMD Athlon based cluster kicks the ass of pretty much every SGI in the data-center where it is.. and I think it even kicks the ass of the NEC... So yes, I'm sorry but 3Ghz is more than 4 times 0.7ghz and it does heck a difference.. And if you look at operation per dollar, there is not even a comparison... And I wont tell you how much their OS sucks.. the latest Irix versions feels like linux for 8 years ago (I mean the userspace stuff, I dont know much about their kernel...)..
Nice troll, and if it isn't, wow. Everything in your post is false. the 700Mhz MIPS certainly does stand a chance against other processors and I would love to have one, but as for price/performance, it is probably a very poor option.
The N64 did well as a system, and had far more power than the playstation. The playstation just did incredibly well.
Hollywood is a city, not a company. I am assuming you are talking about 3D and compositing visual effects studios, of which a few are near Hollywood, California. They aren't going to BSD, they are going to Linux, not just for rendering, but for workstations. Irix is unix and it makes it a very flexible choice for an OS. Because Linux is so similiar, it is also a flexible and powerful.
This Wiki Feeds You TV and Anime - vidwiki.org
Before evryone assumes that this thing is fast here some numbers to keep in mind:
OK there are no numbers for 16K but here the numbers for 600Mhz 14K
SPECint2000 500
SPECfp2000 529
For comparison
UltraSPARC III Cu 1.015GHz
SPECint2000 576
SPECfp2000 775
AMD XP 2800
SPECint2000 913
SPECfp2000 843
INTEL P4 2.8
SPECint2000 1040
SPECfp2000 1048
Oh? Quick, everyone with Radeon 9700 PRO graphics boards in your PCs, make sure you have them in tower cases, or something!
For reference, the ATI specs page states:
I guess SGI might refer to actual output precision, i.e. the RAMDAC D/A-converters... In that case, it seems they still have the edge, since the ATI boards only have 10 bits per component. Still, I think that's of lesser value than the actual precision image operations are performed at.
main(O){10<putchar(4^--O?77-(15&5128 >>4*O):10)&&main(2+O);}
i dont need a MIPS history lesson. I didn't "forget" any of those CPUs. The R8000 was almost non-existant across SGI's product line.
While it was the first implementation of MIPS4, and it was an FP monster, and had a huge TLB for the time, it really wasn't so hot as a general purpose CPU.
A far as "true 64 bit" in the R4000, which version of IRIX ran on R4k with 64 bit pointers ? 6.2 and 6.5 certainly don't on my IP22.
When the R3k came out it was the first real example of commercially FAST and successful RISC design. It was used in multiple machines from multiple companies. SGI didn't "really" up the ante again until R10k, which was their first offering that was superpipelined and superscalar.
Finally, regarding SGI and clustering:
SGI is not price-competitive with shared-nothing clusters of PCs or Alphas. Nor is it trying to be. You probably know what the O2k/O3k systems are good at and how they differ from any other system being sold today, othewise you wouldn't have responded to me. I think my statement is valid --- the SGI big iron solves problems that shared nothing clusters CANT. Furthermore, they're so much more expensive than shared nothings that if you need shared nothing and buy origin, you're silly.
My opinions are my own, and do not necessarily represent those of my employer.
The R16000 has Out-of-order instruction execution? Sweet! So what was SGI's plan when they made this?
1.???
2.Profit!
3.Build new processor.
I wish there was some there was some way that I could be outside playing basketball, in the rain, and not get wet.
Two way systems are not data center solutions that IBM, Sun, and SGI are competing for with this kind of hardware.
Even if they were, you're ignoring the fact that you cannot physically pack as many CPUs with Intel or AMD as with MIPS, Power4, or Sparc into a chassis. Part of the reason they are clocked slower is because you need to balance heat management with performance density when you're dealing with the big servers.
These boxes are about aggregate compute and storage power per dollar, not about whether the individual CPU cores smoke. The only place you see these cores as singletons is workstations (Single-cored "servers" are usually just the same or similar motherboards as a workstation, but in a case that has a beefier power supply and room for a useful number of hot-swap cages.)
You try and pack 32 Intel cores at 3GHz into a chassis that will handle the same number of MIPS cores, and the only thing you're going to get is voltage underflow from an overloaded power supply. Beef up the power supply, and within minutes you're going to be getting that wonderful whiff of frying, overheated electronics.
Raw performance of a core is only one factor in engineering a complete server. Anyone who claims otherwise has clearly not been involved with the hardware end of this industry.
I do not fail; I succeed at finding out what does not work.
The R16000 has many significant architectural and memory-related improvements over the R14000A. However, you are correct in that it's not the speed demon that some folks are making it out to be.
But... keep in mind that it consumes far less than 20 watts of energy (and thus gives off little heat) and will eventually find itself packed in with other CPUs into Origin servers/supercomputers. The CPU bricks for the O3900, for example, have 16 CPUs in just 4U of rack space.
SGI's ccNUMA MIPS/IRIX machines are typically used for tasks that are severely I/O bound, that is, their strong point is chugging thru massive amounts of data where raw per-node CPU power is important, but not the largest factor. Somewhat like a mainframe, but with less redundancy and more CPU power.
and iirc you windows can't officially have higher than 32bit(24bit+8bit alpha/wasted/whatever) currently.
the matrox card that does 48(? or was it 42), is actually using a dirty hack of some sort to get that depth on windows..
world was created 5 seconds before this post as it is.
SGI is still in trouble. I love the company, their concepts, their hot rod machines and the supercool names they come out with..... BUT they are in trouble. And Linux is one of the primary reasons for SGI getting into trouble. A large number of design studios seem to have succumbed to the temptation of a cheaper yet stable machine (i.e Linux Boxes). As some other slashdotters pointed out, these guys are using Macs for the artwork and Linux boxes for the actual bull work. I wonder if SGI can reconquer their old customers and charm even more people.
|/________
|\A|ALYS|
Have you read the SPECint and SPECfp results posted above? The Pentium4 runs at 6 (six) times this cpu's speed, yet only scores twice. Talk about good cpu design.
You should also keep in mind that SGI has some ass kicking technology when it comes to cpu and memory interconnect. NUMAFlex makes it possible to have a penalty as little as 1.5 vs 1 for memory accesses outside the local ram banks. Now try doing that with commodity x86 hardware. For problems that aren't easily broken down in small parts and, that have huge datasets, nothing touches SGI.
Kudos to the SGI engineers for their great job.
A long time SGI fan :)
A far as "true 64 bit" in the R4000, which version of IRIX ran on R4k with 64 bit pointers ? 6.2 and 6.5 certainly don't on my IP22.
64-bit support was first supported with IRIX 6.0 running atop R4000 and up.
However, certain platforms do not support 64-bit pointers. IP12/IP20 (Indigo) IP22 (Indy/Indigo2), and IP32 (O2) are among those that don't. This is due to memory contraints and other assorted issues.
Most, if not all, Onyx and Challenge (L and XL only) machines support 64-bit pointers with IRIX 6.0 and up.
Onyx2, Origin, Octane, and Fuel certainly do.
4MB L2 cache => *huge* die => low yield => huge cost. Yeap, it's that simple.
The Raven