Sun Picks Athlon For Cobalt Servers 125
ncc74656 writes: "In this TechWeb article, AMD may have achieved one of its longtime goals of getting the Athlon into the server market. Sun's Cobalt division is set to unveil a single-processor Internet-appliance server next week that will use the Athlon. Since there's still no 760MP chipset, there won't be any MP Cobalt boxen for a while ... but not everybody needs MP, and this is still a step in the right direction."
Re:Jesus Christ... "boxen" ? (Score:1)
On the other hand, the fact that someone is close-minded enough to get hung up about a playful nonstandard plural is also quite amazing.
Something I once heard... (Score:1)
---
Re:Yeah, 10 watts makes the difference... (Score:2)
The "10 watt difference" suddenly is 537.6kwh/day, or 193536kwh/year.
Even if you look at that in terms of just how much the electricity itself costs, we'd be paying $27,095.04 more a year if we were using (the currently non-existent) dual athlon boxen.
On the right track... (Score:2)
Not an issue at this scale. (Score:4)
Not necessarily true... a saving of $100 per processor adds up when extended to a server farm, particularly when multiprocessor machines are (eventually!) added. given that in certain configurations the athlon outperforms the PIII Mhz per Mhz, that gains some more.
I think what is pushing the envelope here though is RDRAM - buying a Gig of SDRAM is much cheaper than RDRAM ($280 vs $1452, at a glance on pricewatch), as high-end Intel chipsets require it.
The best point about this happening is it may bring the Athlon in as the choice for the desktop machine as well, if the company sees it performs the same or better and is cheaper.
Secondly, assuming you have lots of servers in an enclosed area, heat is a big deal; you want good air condition system. A room full of Athlons is HOT. This further offset the "true" cost of using Athlon servers
I wouldn't agree. The only signs I've seen of Athlons generating lots of heat is when they're overclocked. A friend of mine runs his Thunderbird 850 at 950Mhz at 20 celsius (about 50F), with a decent cooling system.
Any company that is buying machines for use as servers is going to spend the extra $30 per machine to get a good cooling system, regardless of processor type. Those boxes come with 3 fan ports, a heat sink the size of a brick... usually a room with lots of servers in it is bloody cold because of all the airflow!
Fross
Hacker jargon (Score:2)
----
Sun just made the worst decision possible. (Score:1)
Mark my words, I will disconnect from a website/FTP server/game server if I find out that it's an AMD-driven Cobalt. Furthermore, Cobalt is Intel's color. They should've called this series "Jade". Either way, when this server series gets recalled because of the potential of spontaneously crashing due to the CPU's inherent incompatibilities, I'll be the only one who's laughing.
Cobalts have always used AMD chips! (Score:1)
Re:Sun just made the worst decision possible. (Score:2)
laggy API that only performs decently on $75,000 Sparc servers
>>>>>>>>>>
Umm, API's are only laggy if they're inherently badly designed (like X
92.815% x86 compatible CPU = disaster!!!
>>>>>>>>>>>>>>>
Yes, the Athlon is totally x86 uncompatible. Absolute and utter bullshit. I have yet to hear of an incompatibility problem with the Athlon CPU. Aside from the lack os SSE, the Athlon is 100% compatible. Maybe you'd care to point at some stories that show otherwise? Even rare x86 instructions like sysenter/sysexit are supported in the same way as on a P6. What *are* you talking about?
Re:Not an issue at this scale. (Score:2)
if you'd like a better comparison, using figures taken from around 3/4 of the way up the list in pricewatch (ie the top 25% most expensive):
1gig PC133 SDRAM: $800 (4x256M)
1gig RDRAM: $3200 (4x256M)
you pays your money, you makes your choice.
Fross
Thermodynamics anyone? (Score:1)
I wouldn't agree. The only signs I've seen of Athlons generating lots of heat is when they're overclocked. A friend of mine runs his Thunderbird 850 at 950Mhz at 20 celsius (about 50F), with a decent cooling system.
Any company that is buying machines for use as servers is going to spend the extra $30 per machine to get a good cooling system, regardless of processor type. Those boxes come with 3 fan ports, a heat sink the size of a brick... usually a room with lots of servers in it is bloody cold because of all the airflow!
Right, what you do to the *processor* does not matter 2 pinches of goat shit to the *server room*. In fact, it can theoretically make the situation worse ... bigger processor cooling system itself produces more heat for the real cooling system (the one that services the server room as a whole).
As for the stupid comment about airflow, this is meaningless. What matters is the rate of heat exchange out of the server room. What the airflow is in the absence of an efficient heat exchanger is a matter that rapidly becomes moot.
Re:Not an issue at this scale. (Score:2)
or, the Back Of the Envelope calculation, 20*2 + 30 = 70. That's at least one you can do in your head....
Sorry to quibble about this, but I hate it when people get simple conversions wrong.
Re:Sun just made the worst decision possible. (Score:1)
Re:beowulf possibilities? (Score:1)
Re:a difference that can power a whole cpu, heh (Score:1)
And why would you run a 20W lamp? Is that even enough to light a 3x3x8 closet?
Re:What goes around comes around (Score:1)
Re:Not an issue at this scale. (Score:1)
Re:Not an issue at this scale. (Score:2)
Mark Duell
Re:Clarification. (Score:2)
A) The PIII is faster at Quake than an Athlon
B) The PIII is more common than an Athlon
C) The PIII has better Visual C++ support
D) The PIII has multiprocessor capability (Quake III is SMP)
Take your pick. I have not heard a single comment from Carmack that he uses a PIII because of compatibility issues. And even if he does, it means jack-shit because he is probably just phobic from the whole K6 dabacle, (chips which actually *did* have compatibility problems!) If you are saying that programmers avoid AMD like the plague, I'd request that you get your head out of your ass and show me exactly where it says that!
Re:Yeah, 10 watts makes the difference... (Score:1)
Re:MP (Score:2)
Re:Cobalt Alternatives? (Score:2)
In fact, I recently helped a friend purchase a Cobalt Qube 3. He has no knowledge of, or really any interest in, Linux or any other kind of server operating system. But the Qube does what he wants out of the box. It's small, attractive, and has a wonderful web-based administration GUI. The fact that I could have built him the system for half the price is irrelevant - he would not have wanted that system, so it wasn't worth *any* amount of savings, because he wouldn't have used it.
I will say that I wish the markup wasn't quite so extreme. I've long wanted to buy my workstations and servers from someplace like VA or Penguin Computing, but it's just not worth the price markup. (Plus they rarely offer the exact components I want, namely the latest AMD CPus.)
not a good choice? (Score:2)
After all, what good is a server room filled with cool-running PII-350 mobile processors, when your application demands the performance of a 1GHz PIII, 850Mhz Athlon or 700Mhz Xeon?
Maybe I'm wrong, but don't most people purchase servers by determining their application needs, then finding a hardware platform that provides those needs? At least in my experience, the hardware cost isn't even considered as a factor, as you either need something, or you don't.
As for the MP support, Athlon hasn't released it yet, but the design kicks some Intel MP arse. Basically it uses Alpha-style 'only lock what each CPU needs locked' resource sharing instead of Intel-style 'lock fucking everything, we like contention' resource sharing.
Don't get me wrong, I love my Intel box, but servers might not be the right application for *this* generation of Intel chips. The door swings both ways.
--
"Don't trolls get tired?"
Re:Athlon not a good choice? (Score:1)
I don't think Sun wants multiprocessor support for Cobalt. If you want MP go and buy SPARC for your server platform.
Source for used Cobalt Raqs? (Score:2)
The MIPS boxes were fine. (Score:1)
When given enough RAM and having 32bit DMA turned on for the IDE disks, the Qube 2 (for instance) can easily saturate a 100baseT line. If anything, they could do with faster disk.
They ran exactly as I would expect a 250MHz CPU on a PCI bus and IDE drive to run.
Stanfford University Network? (Score:1)
Myth?
Re:OT: Posting at (Score:1)
Re:Cobalt Alternatives? (Score:5)
Say you've got $5000 to spend on a new server. You can get one of two 1U rackmounted servers:
Option 1: 450Mhz processor, 512MB RAM, two 5400RPM 30GB mirrored (RAID-1) drives, lots of free software.
Option 2: Two 1GHz processors, 1024MB RAM, two hot-swappable 10K RPM Ultra-160 SCSI 36GB mirrored (RAID-1) drives, and the identical free software.
Option 1 is a cobalt Raq. Option 2 is a Supermicro 6010L with 1GHz Pentium III processors, Supermicro certified memory, and IBM drives.
What it comes down to is that a $5000 Cobalt RaQ is a $1000 system with a $4000 name.
I'd say to build your own boxes -- I can't imagine anyone here would have trouble working where things go -- but if you don't want to do that, get systems from VA Linux or BSDI; as for software, take a look at webmin, there are very few server applications which do not have webmin plugins, and with webmin you can give restricted access to people as you see fit.
(OT: where did the bandwidth and server space come from, anyway?)
Re:Cobalt Alternatives? (Score:1)
As for the bandwidth, these days I do video work as a side job (editing, lining up talent for voiceovers, doing special effects, and even some 3D representations of complex events that were not originally videotaped... such as automobile crashes). Awhile back a client wanted to know if I could dump low-res previews of key scenes onto a website for review at any hour. Sounded like a good idea. My first attempt with ISDN was a horrible, slow failure. Moved to a DS1, and now have a DS3 & DS1. My ISP at the moment is a local Tier-2 provider, though it wouldn't cost much to change to a Tier-1. My bandwidth utilizaion is almost null most of the time, until a client (usually with a cable modem or some other high-speed connection) downloads a clip. Quite bursty, but I'm sure I could work something out.
Re:Jesus Christ... "boxen" ? (Score:1)
Re:Cobalt Alternatives? (Score:1)
It's a fairly old version of Linux but then, if you wanted the latest kernel, fastest CPU, highest performance get in and hack it system then you don't want a Cobalt. They are designed to be easy for inexperienced people to configure and run.
Re:AMD is great but missing SMP (Score:1)
shutdown on overheat (Score:2)
Re:Cobalt Alternatives? (Score:2)
Mean traffic scales linearly with the number of servers. Standard deviation scales with the square root of the number of servers. The greater sigma/mu is, the more important it is to share your pipe with other people.
Re:Internet-appliance server (Score:2)
An internet appliance server DOES NOT mean it will serve content for home use internet apps.
Internet appliance server are very simplified servers with a CPU, memory, a NIC, a SCSI contoler and an (optional) internal hard disk. Everything else is external to the unit, attached to it via SCSI bus. They are relativelly cheap (for a server) and if it fails you just have to change the CPU box, attach the external storage units to the new box and you're back online in no time. No need to remove a dozen of internal disks from the dead server to put in the new one. Other advantage of internet appliance servers is the low power comsumption and small heat production, thanks to the reduced number of internal components.
No, Cobalts are cheap! (Score:2)
Then you write a bunch of scripts to automate setting up virtual servers for e-mail, web, FTP, file services, disk quotas, databases and shell accounts.
Next, recompile Apache to support the usual hosting stuff like PHP, database access, IMAP, FrontPage extensions, ChiliASP, etc. And install and configure some webmail software and modify it to support the virtual hosting.
Okay, now that you've done all that, install and configure a remote management system that will let you administer the machines and accounts in bulk, and write or cobble together easy web interfaces to let customers manage mail accounts, permissions, access contro, etc.
Ready? Now get a tech writer to put together user documentation for your customers. And keep someone on staff to watch security lists and figure out how to deploy security patches and software upgrades safely without breaking any of your customers' sites.
Point is, if you want to use a Cobalt as a "regular" Linux box, it's overpriced and awkward, and not particularly cutting-edge. But if you're looking to provide web hosting for small-business and personal customers, it's mighty cheap and easy, since it's preconfigured for hosting, supported for hosting, comes with end-user documentation, and can be thrown into a rack and added to your hosting farm in a few minutes out of the box.
Re:Sun just made the worst decision possible. (Score:1)
Re:Sun wont Release a MP box. Duh! (Score:1)
There are machines in the middle, they just aren't marketed as such.
--Ben (who spends waaay too much time at Sun)
Re:Sun just made the worst decision possible. (Score:1)
either. I mean they work and stuff, but when it
comes down to placing a large load, then they
collapse, but randomly.
Intel machines though slower at some places, have
smoother ride. Like when you run programs,
after large loads, AMD CPU is sort of confused and
runs real slow, and then 5 sec later it all gets
into chunky pipe line.
Most code out there is designed for x86 by people
who have celerons of PIIIs Athlons are only recent
addition to x86 familiy. We had servers do kernel
oops on Athlon machines every 2 months or so.
And that is production quality kernel!
I'd say to AMD better to get your act together
or you will not see the high end market at all.
Problems like that are random... weird.
All hardware is respectable:
GeForce or TNT2 Ultra
A7V + AMD
IBM 75GXP type HDs
D-Link cards - very nice tulip like interfaces.
Go figure.
A classic struggle between GOOD and EVIL (Score:3)
We already know that Slashdot's readers overwhelmingly support AMD to holy-war proportions.
We should be able to moderate the stories actually post on the main page. The editors should be subject to karma. If you put up something stupid you get smacked.
Of course, as a result of editor karma-whoring every other story would show Linux pit against any other operating system or AMD pit against Intel.
There were probably more than one submission about this news but Cliff chose the one that showed significant bias.
I don't even really understand why there is such a hype about the Athlon chips. They are less expensive and may be technically superior. But that arguement doesn't cleanly follow the patterns slashdot readers follow. Take Linux for example, it is surely less expensive than other operating systems, but it is not even close to being technically better than all others. FreeBSD, technically better, lacks bleeding edge drivers, same cost as Linux, yet with almost 1/5 the user base.
Who the hell cares if they can get 150fps in quake or 147, I sure as hell don't. There are better things to argue about. We're about to innagurate a president who will only account for his actions in the past 20 years.
And on top of it all, my favorite lighter is almost empty.
Re:Not an issue at this scale. (Score:1)
Also, I notice you didn't respond at all to the point that Althons need more power (about twice as much as an equivalent pentuim, I think). That in itself is a reason not to use them, especially if your server farm is in california
Re:a difference that can power a whole cpu, heh (Score:2)
I don't consider half a 20W lamp to be much responsible for global warming. Do you?
Re:Yeah, 10 watts makes the difference... (Score:2)
In any case, if we have the 10 watts times 112 servers, then = $2,700/year, the difference in price between an amd 1.2 ghz t-bird and an p3 xeon w/ 256k cache 1ghz clock this morning is roughly $600-$280 = $320. 112 times $320 = $30,000+. Deperciated over the 2 years of expected lifetime of such a server this is $15,00 per year. The xeon farm you suggest (even for single processor configurations) is thus over $12,00 per year more expensive. Add in the fact that the 1.2 Ghz t-bird runs rings around the xeon and the p4 with today's code and the price/performance issue is still solidly in AMD's court. However, your point is well taken that one has to consider the entire cost of ownership for a given level of performance. Anyway, the power dissapation issues may be a thing of the past once AMD proceeds with SI-28 technology. They have produced 1.5 paliminos that are passively cooled. Kinda will throw the power argument on its ear after isonics irons out the problems with eagle-pritcher this quarter.
Intel may close the gap again and pull ahead at some point, but for a while AMD has the advantage. As it currently stands, Intel produces a distinctly second place product since Andy Groves retired.
Re:Yeah, 10 watts makes the difference... (Score:1)
Off by a factor of 10, which makes the annual cost $2,700, not $27,000.
Re:Yeah, 10 watts makes the difference... (Score:2)
Great.. (Score:1)
Re:As for SMP (Score:1)
Re:Not an issue at this scale. (Score:1)
Re:HOOHA (so fuck off) (Score:2)
Re:Clarification. (Score:1)
Are you slashdot-terminal's cousin or something?
Re:Cobalt Alternatives? (Score:2)
Somewhat of a side question or solution; are your needs simple enough for the much-less-expensive Cobalt Qube?
I'm considering getting a Qube, as it would help me deal with all the www/ftp/smb stuff I use for my work-at-home days.
If anyone has a Qube, lemme know if it supports hosting CVS? How about setting up ipsec tunneling? (I have a second box doing vpn/nat already.)
Re:a difference that can power a whole cpu, heh (Score:1)
Yes. Use more effecient bulbs (flourescent or LED) instead of extremely inneffecient incandescent.
Re:Not an issue at this scale. (Score:1)
Now weather those benchmarks play true to the server market, i dunno. But increased memory bandwidth would have to be a good thing.
Besides that, another thing the AMD has going for it is the 266Mhz (on >850mhz chips) FSB(ok, 133 DDR). This again gives it just a little more speed.
Re:why not use SPARC? (Score:1)
Re:As for SMP (Score:1)
Athlon not a good choice? (Score:5)
First, lets get this get of the way: I use an Athlon at home and a Pentium at work.
Many people (including myself) like Anthon because it is cheap and fast, more bangs for your bucks and all that; this is ideal for home user. However I see two things against using Athlon as a server: it uses more power and generates more heat.
Now in a serious server environment, it seems to me that money is probably not as big an issue, so this kind of negated one of Anthlon's main advantage over a Intel chip right off.
Secondly, assuming you have lots of servers in an enclosed area, heat is a big deal; you want good air condition system. A room full of Athlons is HOT. This further offset the "true" cost of using Athlon servers
Also since Athlons use more energy, it stands to reason that your UPS system will not last as long as a similar number of Pentiums if there is a problem. Now for servers, up time is very important (unless you run WinXX ;-), so this seems another strike against Athlon.
And finally like the article said (you did read it right?), Ahtlon don't have multi-processors support yet so that is another strike...
Don't get me wrong, I love my Athlon. But server may not be the right application for *this* generation of AMD chips. But AMD did promise support for multi-processors, low-energy chips soon, so there is still hope.
====
Re:Jesus Christ... "boxen" ? (Score:2)
Re:Cobalt Alternatives? (Score:1)
In Taiwan good QA people are fired for insubordination if they raise a peep about quality issues.
'Nuff said.
Just a clarification and some personal comments (Score:1)
Is this really what AMD wants? (Score:3)
low-end Internet-appliance server
I gues that's not exactly the Server market AMD wants too get their fingers in. It may be a lucrative but that's not ther server market Intel dominates. I guess there are already a lot of homegrown AMD based "Servers" in a lot of offices
Besides that it's not even a new customer for AMD. As the article states:
already supplies the K6-II processor for Cobalt's current one-rack Internet server
Must be a very big server if it takes a whole rack(Just kidding, I know it's one U)
Sun wont Release a MP box. Duh! (Score:2)
Re:Athlon not a good choice? (Score:2)
It's even possible that SUN intends to help them put one together (It's not like Sun's never done board/chipset design). I can see that as being a mutually beneficial transfer of technology
`ø,,ø!
Which OS will future Cobalts run? (Score:2)
Seeing how current Cobalts run Linux and Sun's more recently slightly GNU/Linux "friendly" stance, I would imagine they would stick with Linux... but anything can happen. Linux seems to be doing the job well, but I would almost like to see them try something with *BSD, just to see how well it works. Trying other things is a *good* thing.
heat & power Re:Athlon not a good choice? (Score:5)
Your observation that the Athlons use, on average, more power and radiate more heat than the comparable Intel processor is, from what data I've seen, absolutely correct. However, I don't think that they use more power or generate more heat than an Alpha processor (some of those guys get hot, as anyone who's ever not quite given an Alpha-attached heatsink enough time to cool down has discovered ;-) ). Alphas are fairly common in the server (and rackmount server) market. True, you probably wouldn't want to stick one in a 1U, but still... (I don't have any hard numbers, but I imagine that the Duron line is probably well suited to a 1U in terms of die size, power use, and heat profile, given a well-designed case at least.)
If there is anything that will hold the Duron/Athlon/Thunderbird line back from server-market acceptance (technical reason, not Intel-ism in IT depts.), it would be the comparatively small cache sizes. I.e. you'll probably want to use an UltraSparc-based solution with (2-8) megs of cache per chip (or some other "big cache" arch like Alpha) for the DB server[1], but everything else (www, mail, etc) is just fine with the "small" x86 machines, a domain in which ceteris parabis Athlon would win over Pentium by virtue of decreased cost for similar of greater functionality.
[1] of course there are many other factors that go into making a mid-to-high-end DB server, but I/O and backplane bandwidth do play an important role; having large chip caches and >32bit architectures helps this quite a bit (the other main area is of course disk, but that's outside the topic at the moment)
--
Fuck Censorship.
maybe sun could work on the cobalt software next (Score:2)
Making the admin interface somewhat less than DEATHLY slow would also be a big help. I can whip out three zones in webmin before I can pull up a zone in cobalt admin.
How about not using a drop-down box for selecting which zone to edit in the DNS configuration? Netscape on Unix -- you know, that OS that Cobalts use, and I hear Sun develops -- doesn't exactly deal gracefully with that.
Sorting domains in the virtual site list by TLD first was perhaps the most precise and logical thing, since you want to keep subdomains together
It would be great if the interface weren't so slow.
Speaking of webmin, I can middle-click on anything in the interface and get a popup. Cobalt's excessive use of frames and javascript make that quite impossible, screw you very much.
Did I mention the interface is slow?
--
Re:Source for used Cobalt Raqs? (Score:1)
Another consideration if you're planning on the used RaQ route is that cobalt charge for their OS restore CD (installed via NFS IIRC), and it is the only way to do a proper restore - the RaQ's have no bios, so you have no way of getting any alternative to work. TBH, you would be better looking some of the alternatives - it'll work out cheaper, and with better hardware in the long run...
As for SMP (Score:2)
If I have the coin to get a dual system of any x86 chip, I will. If I don't, I'm getting a Thunderbird. The DDR ram will give me some nice throughput (hmm, server potential there?) and the integer performance? I can definately be happy with that...
Here's the point: Why pay more for a SMP box if you don't need it? ftp.cdrom.com uses a single P3 for all the users it serves!
The Athlon will provide an excellent server for many.
---
Re:OT: Posting at (Score:2)
powerpc. (Score:1)
they dont
no one cares
Re:better than MIPS the mips-based RaQs for sure . (Score:1)
The MIPS processors have no L2 cache, so performance definitely suffered as a result. That, and lack of 3rd party application compatibility are the main reasons Cobalt dropped the MIPS CPU for the AMD K6.
MIPS interfaces were slow too because they were all static HTML files which had to be regenerated by CGIs everytime you changed some information. x86 boxes use PostgreSQL to maintain/mirror a lot of config info to speed dynamic page generation.
The new Sausalito [cobalt.com] API (available on Qube3, and all future RaQ/Qube products) is another huge step forward. Persistent connections, etc, make the average time for a new screen to appear in the 1-second-or-less range. Not to mention the API can be hooked into from PHP, Perl, or C++ for end-user customization possibilities...
Re:OT: Posting at (Score:1)
Well I'd use that box myself if all of my contents weren't so damn insightful and deserving of positive moderation...
--
Re:Cobalt Alternatives? (Score:1)
Re:Source for used Cobalt Raqs? (Score:1)
Re:Internet-appliance server (Score:1)
Cobalt used AMD processors previously (Score:3)
<ramt>
I will be extermely glad when AMD enters this market in a serious way. There has been a less than virtuous circle occuring with Dell/Intel commercial products. Since the company buys Intel servers, it buys Intel desktops (after all, we can only find the talent with the mind of a slime mold to maintain them you know and these people can only maintain one kind of one thing...). Both the desktops and the servers are horridly underpowered and overpriced. Further, while Dell machines are made to be assembled easily, upgrades are an oxymoron. It will be nice when the corporate market understands that it has a real choice.
</rant>
<disclosure>
I own and reccomend AMD stock
</disclosure>
Yeah, 10 watts makes the difference... (Score:2)
What goes around comes around (Score:2)
Re:Not an issue at this scale. (Score:2)
IIRC, it's Intel i840 chipsets that need RDRAM, and that's a Coppermine chipset, has been since sometime in 99. Yes there are SDRAM chipsets for the PIII too, but a server would use the highest-end it could go, supposedly. I can't recall what the P4 chipset is, but yes that uses RDRAM too.
Also, I notice you didn't respond at all to the point that Althons need more power (about twice as much as an equivalent pentuim, I think). That in itself is a reason not to use them, especially if your server farm is in california
Anyone who houses a server farm in california, between the power shortages/brownouts and the ridiculous Silicon Valley rates, is not playing with a full deck, to put it mildly. I don't think the 10W extra consumption of the Athlons is anything to break thew bank though
Fross
The 'Enhanced Screwdriver Shop' Killer (Score:2)
They won't run Linux on the boxes; they'll adapt Solarix x86 (fine-tuned so that it runs flawlessly, Solaris x86 in general suffers from being a product from a company that codes to their own hardware adapted to run on generic hardware). By doing so they will be able to market seamless Solaris solutions from top to bottom-end.
It doesn't bode well for Linux in the enterprise at all that Sun is doing this. In particular the 'Enhanced Screwdriver Shops' like VA are gonna be hurt.
Why would they care? (Score:2)
Servers vs workstations (Score:2)
I agree with those disadvantages, but I see them more as a strike against Athlon in workstations, and having negligable effect on servers.
Workstations outnumber servers by dozens (perhaps hundreds) to one, so the real place where you want to conserve power is in the 100 computers spread all over the building, not the two boxes in the server room.
And (I guess this depends on what you're serving) in my experience, servers are generally I/O bound, so lack of SMP isn't a biggie. OTOH, at the desks, is where people want major CPU power for a faster frame rate, compiles, raytraces, etc. (Or if they're running Windows, faster rebooting and paperclip animation. ;-)
So Athlon sounds pretty good for servers, to me. It's the high-end workstations and hacking machines and gaming machines where it fails.
---
Re:Athlon not a good choice? (Score:2)
Looking at their datasheets, it doesn't make sense to put an athlon in a sever box when a p3 would use such less power and so much more reliable.
Athlon (1.2) P3 (1.0)
Pmx 66 26 W
Ta 42 45 C
Tj 95 70 C
Vcc 1.75 1.7 V
c/s no-dp yet i840
First, 66 vs 26 Watts is a big difference, especially given city power grid issues on the west coast.
Second, Ta is the temperature the OEM must keep at 3mm above the die. 42 is _very_ hard to keep for a 66 W part and requires a real sol'n. Most sol'ns I've seen from OEMs are less then adequate, but they don't have to be too fancy because typical desktop PCs can be a low cost sol'n. A server part would need to spend way more $$$ for an AMD part than Intel.
3rd & 4th. Tj is the max temp the die can run at at that frequency. This number is usually increase to hit higher frequency targets. As Tj increase, electromigration becomes a huge issue. EMigration is the deteriation of the metal contacts on the die, which directly translates into MTBF. If intel can run 1GHz at a whopping 25C LESS than AMD, that says something about intel's reliability and reduced cost to cool.
(Side note: compare these temp/pwr numbers then compare the cooling sol'n seen in OEM boxes: they're usually the same for AMD/intel. Now ask yourself is Intel overconservative, or is AMD hoping their boxes last long enough for the next upgrade cycle?)
SO why did Sun go with Athlon? Don't know. Must be a political move to piss off Intel, esp. considering they plan to support IA64. Also odd b/c historically, Sun has always used Sparc processors and eschewed x86. Why the sudden change of heart? We'll see what kind of volume they do on this box. It may be just political move.
---
SPARC vs. AMD (Score:2)
The migration still may happen. The new RAQ most probably was already designed and was in pre-production prior to Sun buying Cobalt.
-E
SMP? LOADBALANCING! (Score:2)
What bothers me in all these discussions is the lack of load balancing talk. I work in a data center for a Verio large Internet hosting company. We use Intel and Sun based servers, with a couple rows of Cobalt MIPS bases systems as well. I've been a server SA for the goverment and large commercial companies. I hate to say it, but SMP is only really needed, least for Internet servers, on DB servers and huge mail servers (millions of messages an hour). Otherwise, for serving up webpages, ftp servers, or doing DNS, a single processor system will work fine. What I've seen needed more in serving webpages and doing FTP is good load balancing. I think AMD should be looking at this market with eyes wide open. Software and hardware based load balancing on cheapish servers right off the bat makes more sense then trying to shoe horn it onto servers that can barely keep up with the flow.
Just one nuts thoughts...
creature
Re:Sun wont Release a MP box. Duh! (Score:2)
Well, there is a large market segment (ah, help, I'm thinking like a marketing major! :-) ) in terms of performance demand between one x86 processor and a high-end Ultra Enterprose machine (rackmount uniprocessor server = ~$3000, maxed-out E10K = ~$2,000,000 (of course you do get an awesome amount of performance with the later, heh)). In other words, if a customer comes to you and says: "Well, I need something more beefy than your $$$ uniprocessor machine, but not as beefy as your $$$$$$ U.E. line..." you want to be able to sell to than need.
This price/performance continuum is the same thing you see done in many markets (e.g. Celeron/Px(x={2/!!!}/Xeon by Intel). It just makes good business sense.
--
Fuck Censorship.
Re:better than MIPS the mips-based RaQs for sure . (Score:2)
Re:Which OS will future Cobalts run? (Score:2)
If I had to put on a pointy-haired-boss mask and think like a Sun manager in charge of this, I'd speculate that the machines will come standard with a choice between a Linux (most likely RH or Debian; to look at other companies doing similar things like SGI) and Solaris x86. Unless I'm missing something these are pretty standard x86 machines, so other OS's would be easy to install (*BSD, NT4/Win2k (ugh)).
But who knows? Maybe they'll decide to throw us all a curveball and stick something like Darwin or QNX on there by default. :-)
--
Fuck Censorship.
Cobalt Alternatives? (Score:2)
Are there any great alternatives to the Cobalt line? Perhaps another similar piece of all-in-one hardware. Maybe one that is even better? Sure, setting up a Linux or BSD box could work, but there would be a steep learning curve for someone that wants a Cobalt. Maybe there's a Cobalt-like GPL (or even commercial) package for Linux? Any suggestions?
Re:Athlon not a good choice? (Score:2)
SMP is less of an issue because I don't think Cobalt offers a multi-processor configuration. If you're buying a multi-processour machine, you're less likely to want to use Cobalt's software, and Sun probably wants you to buy a Solaris server anyway.
Re:MP (Score:3)
a difference that can power a whole cpu, heh (Score:2)
10 watts is also way more than enough for embedded MIPS/ARM/Transmeta CPUs, too. Food for thought.
Re:heat & power Re:Athlon not a good choice? (Score:2)
Um, I hope you didn't mean "you wouldn't want to put an alpha in a 1U," because its already been done:
AlphaServer DS10L [compaq.com]
PIII chipsets and RAM types (Score:2)
Exactly right. The highest-end PIII chipset is the ServerWorks ServerSet III, which uses SDRAM.
I imagine that in the future there will be a version that supports the P4 bus and DDR SDRAM.
Re:MP (Score:2)
SMP is imporatant for big servers (Score:2)
It really depends on your application of choice. Some applications are I/O bound. Here another proc wouldn't help. But many applications are processor bound, especially at certain points. When your I/O is really fast, you start to really see the importance of MP. For example, there is a very big difference between your best SP I/O speed vs. your best MP I/O speed when you are using a RAID array with lots of cache connected through a 64-bit 66 Mhz PCI FibreChannel controller. If AMD wants to win the big server environments where the profit margins are higher, they need to support at least dual processor systems. Heck, Unisys has a 32x Intel box that they have licensed to Dell, Compaq, and HP.
Re:Which OS will future Cobalts run? (Score:2)
--
you are not what you own
News? (Score:4)
--
A message to you Rudy (Score:2)
Re:Cobalt Alternatives? (Score:2)
It depends what you're after - I've been adminning RaQ's for a while, simply because they're been available from hosting providers at relatively good prices. For what they're meant to do, they're great machines - powerful enough, dead simple for even a fairly clueless user, but reasonably flexible so as not to hold back the clued too much. On the other hand, there becomes a point that the RaQs are too frustrating, and you feel the need to move on. I'm currently setting up a cluster of Intel ISP1100 servers - PIII 800, 1GB ram, 2x36GB hdd, all in 1U of rackspace. They're extremely fast, and you can stick pretty much any OS you like on there. Along with something like webmin, or an inhouse frontend, they'd make a pretty good solution. They're also not that expensive for what they are...