Athlon MP Reviewed

RendEr writes "At The Tech Report, there's a review of AMD's latest multiprocessor chip, the Athlon MP 1900+. Watching this thing smoke through Linux kernel complies is a beautiful thing. Combined with AMD's new 760MPX chipset, these chips could help usher in a new era of cheap dual-processor desktop systems. "

Comments

Is it me?

[professortomoe]

Or does anyone else think AMD's new naming convention is nifty. I'm getting sick of everyone bragging about their ghzs.

Re:Is it me?

[Sobrique]

AMD, Intel and Sun have all reached the conclusion that 'Mhz don't matter'. As a measure of processor speed it's only of a little relevance. The UltraSPARC III+ at 1015 Mhz is going to hit benchmarks comparable to a 2Ghs P4. Got to wonder about 'Athlon XP' though. I mean, did Windows beat them to it? And couldn't they think of anything less of an 'XP' cache (sic) in? Athlon RF (really fast). Athlon BI (Better than Intel). Athlon Turbo. :)

Re:Is it me?

[_Ash_]

Well, the term XP sells. AMD knows that. There are still much more users running windows than anything else. And all (or at least most) of these users want Windows XP. So if they're going to buy a new PC and they see Athlon XP, they will get an urge to buy it. It's pure marketing technique.

Re:Is it me?

[gazbo]

It's all marketing, though. Everyone knows that although AMD deny it, the XP nomenclature is a direct cash-in on Windows. And you're right, all processor manufacturers know that MHz are not a good way to compare chips with different architectures, but despite being told this over and over, Joe Public seems to be swayed by the big numbers. That is why Intel put so much effort into making high MHz over all else, and AMD are (effectively) lying about their MHz.

This is also why IIRC AMD will not approve any bios that lists the true speed of an AMD XP chip.

Re:Is it me?

[stoffel]

The great megahertz myth...

See http://www.apple.com/g4/myth/ for a simple explination that hertz is not everything...

Re:Is it me?

[nocent]

Or does anyone else think AMD's new naming convention is nifty. I'm getting sick of everyone bragging about their ghzs.

i agree with you regarding mhz is not everything. but what a coincidence. when AMD broke through the 1ghz barrier, they couldn't wait to brag about it. now that they're getting left behind mhz-wise by Intel and having trouble reaching 2ghz, suddenly, "clock speed is not an accurate indication of performance". if apple and amd had 3ghz chips, you can be sure they would be hyping that up like crazy.

p.s. it's nothing new. AMD and others used to do that eons ago with their old chips which led to tremendous confusion amongst users.

Re:Is it me?

[Hamshrew]

Mine lists the true speed(Tiger MP).
Actually, I'm not sure if it lists it on bootup. But it definitely lists it in the BIOS Setup screen.

Re:Is it me?

[mrfiddlehead]

All they're doing is using Intel's chip speeds as an external comparison. So AMD's 1900+ runs at 1600Mhz but compares to a P4 1900MHz system. We're not going to get away from the MHz comparisons that easily, especially since it still remains the single most identifiable factor.

Psychologically it's a good move for AMD. Even though I know that their 1600MHz chip is faster than Intel's 1900MHz chip (or equivalent) I would still feel a bit, disempowered ... groan. And for those who do not know they are surely going to be suckered into Intel's super MHz sales pitch.

Ya gotta roll with the punches ...

Re:Is it me?

[JanneM]

FLOPS is just as bad; it only measures FPU performance, so it's only remotely accurate for math intensive apps; also, just as for CPU speed, it totally ignores how much work you actually accomplish during a cycle. We might as well go with BOGOMips - at least we know those aren't accurate...
/Janne

Re:Is it me?

[Gannoc]

AMD, Intel and Sun have all reached the conclusion that 'Mhz don't matter'.

Wrong. Intel archetected their entire P4 line around having higher clock speeds in order to fool the public. This isn't a rumor, I worked there at the time.

They realized that when someone went to buy a computer, performance didn't matter as much as a big number. Consumers think a P3 running at 800mhz is much faster than a P3 running at 700mhz, and don't even consider stuff like video card, memory, and disk speed, much less differences in chip archetectures.

I mean, there's a reason why P4s performed so bad. Its not like they were all done making them and realized "Oh no! This thing isn't performing that well! Its barely better than a P3!" they knew it as they started to design it.

_AMD_ realized that they couldn't win the numbers battle, and renamed their chips to compensate. I hope it works better than when Cyrix did it.

Re:Is it me?

[nomadic]

Since when does Joe Public compare chips of different architectures? The naming convention is used by the non-computer literate, because traditionally it's been accurate; if you had a Pentium 233, and got a Pentium 333, it was faster. Seemed to work just then. Of course adding Athlons screwed it up a little, but it's still a good, general rule-of-thumb for comparing chip generations.

Re:Is it me?

[roystgnr]

Since when does Joe Public compare chips of different architectures?

Since as long as I can remember. I have friends who never played games who saved a bundle on Cyrix chips with weak FPUs... and others who had to avoid those like the plague.

Since right now, especially. It's always been the case that Intel chips could be outsped by random ultraexpensive workstation CPUs at half the clock rate, but now we've got 2Ghz machines being outbenchmarked by *completely compatible* CPUs at 3/4 of the clockrate.

I think AMD's marketing here is shamefully deceptive, but they really did need to do something, and Apple's "public education" attempts about the growing irrelevance of MHz didn't seem to work very well.

Re:Is it me?

[nomadic]

Since as long as I can remember. I have friends who never played games who saved a bundle on Cyrix chips with weak FPUs... and others who had to avoid those like the plague.

They bought the chips separately? Then they're not Joe Publics. I'm referring to the vast majority of computer owners, who only think about clock speed when they're buying a new computer every four years.

I think AMD's marketing here is shamefully deceptive, but they really did need to do something, and Apple's "public education" attempts about the growing irrelevance of MHz didn't seem to work very well.

It didn't help matters that Apple computers aren't really that fast; don't get me wrong, I own a mac and I love it, but the PPC just doesn't outclass x86s like Apple claims, at least in real-world applications.

Re:Is it me?

[roystgnr]

It didn't help matters that Apple computers aren't really that fast; don't get me wrong, I own a mac and I love it, but the PPC just doesn't outclass x86s like Apple claims, at least in real-world applications.

True; the "400Mhz G4 beats 800Mhz Pentium" tests all depended on vectorized assembly stuff, but I don't think there's a benchmark at which a G4 wouldn't have blown away an Intel chip at the same clock speed. I suppose that's part of the problem: "Our computers are slower than theirs, but not nearly as much as you'd think!" isn't a great marketing gimmick.

Re:Is it me?

[rnturn]

``If I designed a processor for AMD I would name it the Athlon ROLA-ESX.''

I must need more sugar and/or caffiene in my system. It took me a while to undo the big/little-endian translation before I got the joke.

Sweet...

[NightWhistler]

"Sorry kids, no Christmas presents this year... Daddy's gonna buy a dual Athlon!" ;-)

Athlon MP

[GigsVT]

We just set up an AthlonMP 1600+ server using the Tyan Tiger board, and I have to say, Intel is going to have some serious competition in the server market.

This thing is incredible. With our RAID streaming 30-40meg/sec writes, and 100-130meg/sec reads, the Athlons barely break a sweat, sitting at 2X25% utilization, in the same situation where Dual 933 coppermine Intel chips maxed out at 2X100%.

The main reason we hadn't gone with AMD sooner in a server is because of the lack of a 64bit PCI board that didn't require special power connectors.

The Tyan Tiger was a godsend. In all, it, two 1600+s, 1gig DDR ram and a dual 160 SCSI card cost about 25% less than the Supermicro P3TDE6, 1GB RAM, and two 933 coppermine PIIIs (on board dual SCSI).

The Tyan board does have less 64 bit PCI slots, and also doesn't support 64bit 66Mhz PCI, but we didn't have any cards that supported that either. It does have four 64bit slots, and that was enough for us.

One thing I don't understand about the Tyan is why they didn't just make all the slots 64bit PCI. It is fully forward and backward compatible.

As a former die-hard Intel guy, I have to say AMD is finally a contender in the server market.

MP 1900+ same as XP 1900+

[Phosphor3k]

The two are the exact same chip, excepting AMD's "SMP certification". So, if you want an even cheaper duel cpu solution, (without the warrenty of course) go for an 760MPX board with 2 XPs.

Re:MP 1900+ same as XP 1900+

[Chazmati]

I've heard there are 760MPX boards coming out (with 66 MHz support for the 64-bit PCI slots; is there anything else?) but I haven't seen any yet. Anyone see any non-sketchy details anywhere? The timing I've read about was mid-November. Anand Tech probably had the best article with pictures of some boards. Here's a link to his Preview from Comdex.

Re:MP 1900+ same as XP 1900+

[red_dragon]

There are some additional, albeit minor, differences between the MP and XP; IIRC, the MP's hardware data prefetch is optimised for SMP configurations. You can put any Athlon on a 760MP(x) board (even a Duron), but the MP will be a bit faster because of this. However, beyond this, yes, they're pretty much identical.

Re:MP 1900+ same as XP 1900+

[Chaostrophy]

Not the same!

They have a few treaks for better memory performance in multi processor situations, have a look at the http://www.anandtech.com/ benchmarks. They do not seem to have benifit in single cpu systems.

Yes, unlike Intel, AMDs multi cpu version of their chip has real design differences, not pinout and cache changes.

Of course, all socket A chips are good for SMP use.

Re:MP 1900+ same as XP 1900+

[Sloppy]

Of course, all socket A chips are good for SMP use.

Excuse me, you look like someone who might have a clue. :-) Do you know if you can mix and match different types? Could I use one of these new MP 1900+ with an "old" 1 GHz Thunderbird on the same motherboard?

Re:MP 1900+ same as XP 1900+

[psamuels]

The two are the exact same chip, excepting AMD's "SMP certification". So, if you want an even cheaper duel cpu solution, (without the warrenty of course) go for an 760MPX board with 2 XPs.

Yeah, and get the 1800+ rated chips instead of 1900+ and just overclock 'em.

Or, if overclocking makes you nervous (as it should), using XP chips instead of MP chips should make you nervous as well, for exactly the same reason.

I believe that very recent releases of the Linux kernel include a check to see if you're using Athlon XP chips for SMP. (Actually I don't know if the check made it in, but Dave Jones was at least playing with it.) This information is included in stack dumps, so you can cut and paste it into your bug report email and the kernel developers will know your machine is out of spec. Apparently there have been some odd bug reports in the past that might point to XP + SMP issues.

Re:MP 1900+ same as XP 1900+

[Chaostrophy]

http://www.anandtech.com/showdoc.html?i=1483&p=13

1.2Ghz MP vs regular 1.2Ghz Athlon (and many other AMD and Intel systems) single and dual, on a database benchmark, lower numbers are better.

2 MP 12.2
2 normal 16.0
1 MP 19.3
1 normal 19.8

Re:MP 1900+ same as XP 1900+

[Chaostrophy]

and I've never had searches work at Anandtech.

: (

Uses

[jaavaaguru]

Apart from really fast kernel compiles and stuff like that, what's the benefit of such a machine? Messing around with the usual windows stuff, my AMD K6-450 is about as good as my friend's Intel P4. When it comes to playaing a game at the same time as an MP3, I can see where the MP becomes useful, but as far as I'm aware not many games are written to take advantage of multiprocessor, and although Windows supports it now, it doesn't make best use of it going by reviews I've seen in the past (the one I'm thinking of was mentioned earlies this week, but I can't remember the article title - it showed the MP being 7% faster under an mpeg decode and game playing benchmark test). Lets hope the Althon MP encourages people to write code that is suited to a multiprocessor environment.

Re:Uses

[keefebert]

It is really suited for the server market. It is true that you don't need a dual processor system at home, that is why you can't go to Best Buy and pick one up. They are suited for server environments and men with small -um, I mean- power-hungry users.

Re:Uses

[jedrek]

I can't belive that. I just switched from a dual celeron 450 to a p3-866 and there's a huge rise in the 'feeling of speed', for lack of a better term. Everything goes faster. I burn CDs, play MP3, have a few apps (photoshop, flash, illustrator) in the background and play RTCW and it doesn't even break a sweat. Gotta love it.

Re:Uses

[MtViewGuy]

Actually, nowadays a fast way to improve your system performance is simple: get as much RAM as you can afford and install a faster hard drive.

ATX-motherboard systems that use the Intel 440LX or newer chipsets immediately benefit by going to 256 MB of RAM and using the latest UDMA-100/133 hard drives. It's possible to get performance increases of 100% or more because 1) you dramatically reduce the need to use swap space on your hard drive, and 2) data is read and written to the hard drive much faster.

You can also increase performance with a newer graphics card, but that affects games for the most past.

Re:Uses

[Amazing Quantum Man]

Who needs to feel lucky on Google? NVidia makes it fairly easy to find Linux drivers. On their home page, under "Download Drivers", there's an item for 'Linux'. Not particularly hard to find, and (amazingly) it's in the logical place, too!

Re:smokin!

[Anonymous Coward]

Oh come on... "heat issues"? Yeah, it smokes if you rip off the heatsink, what a surprise...

Reliability & Compatibility

[Detritus]

OK, it's very fast. That's nice. How reliable and compatible is the system? Those are my top priorities, esp. for a server. How well does it run with some random version of Linux or *BSD?

Re:Reliability & Compatibility

[zsazsa]

How reliable and compatible is the system?

Given proper cooling - very reliable. The thermal diode in the MP/XP line improves this reliability even more. (Which brings up the question - do these boards fully support the diode?)

How well does it run with some random version of Linux or *BSD?

Perfectly.
The onboard stuff on the Tyan boards is quite standard: Adaptec AIC7xxx SCSI, 3com 3c59x Ehernet.

Ian

Re:Reliability & Compatibility

[GigsVT]

I had trouble getting lmsensors set up at all on a Tyan TigerMP. I gave up after a while, so it might be possible with some difficulty.

The finger test works for me. They get pretty warm, but with stock coolers and sufficient case air flow, they don't get too hot.

Re:Reliability & Compatibility

[LordNimon]

It's not very stable for me. I can't get it to run cpuburn for more than 10 minutes before hanging, and this is with just dual Athlon 1.2s on a Tiger. A friend of mine had the same board running 1.8's, and he had to return it because it wouldn't last more than a week before locking up.

Re:Reliability & Compatibility

[whovian]

I haven't yet gotten around to trying it, but this user seems to have had some success.

Text:

From: otheos (otheos.at@clara.net)
Subject: Re: TigerMP (2460) and lm_sensors, anybody care?
Newsgroups: alt.comp.periphs.mainboard.tyan
View: Complete Thread (5 articles) | Original Format
Date: 2001-12-05 17:43:47 PST

Was not very simple but finally got it.

You need:
/sbin/modprobe i2c-dev
/sbin/modprobe i2c-amd756
/sbin/modprobe w83781d init=0

Make sure you use init=0 otherwise the system will freeze. Also getting the
new lm_sensors from the CVS and patching/recompiling the kernel works
better (didn't mention anything though).

Now when you run sensors you'll get 77C for the CPU's. but if you change
the sensor type at /etc/sensors.conf under the:

chip "w83782d-*" "w83783s-*" "w83627hf-*"

#like this:

set sensor1 2
set sensor2 2
set sensor3 2

and then run sensors -s (to read the changes) and then sensors again,
you'll see all fans and CPU temps + the Northbridge readings. Voltages are
more complex (apart for CPUs) as the calculation method needs tuning.

Re:Reliability & Compatibility

[Hast]

You need to get bigger / more fans. A friend of mine had problems with his Athlon because it didn't have proper cooling. (The fan was rated at the specific speed, it just didn't do the trick anyways.)

A new CPU fan later and his system is running nicely.

complies

[snatchitup]

Well, I'm glad this chip complies with the Linux kernel. Does it comply with any others?

Yeah, but the systems are pretty had to find

[MurrayTodd]

I've thought for a while that the new Athlon MP systems would make great desktops--especially with what a dog the P4 is. The funny thing is that almost nobody actually makes an Athlon multiprocessor desktop. A few places make servers. Otherwise you're *almost* required to build a system from scratch. (It's doable, but still a pain.)

One online company that did have decent looking systems was Xi Computing, in case you're interested.

Better naming system

[Hougaard]

Simply call it the:

"AMD Athlon Faster than the last one we released"

Re:smokin!

[linzeal]

I want to see tom's Hardware do an expose on the space shuttle and take off the heat shield during re-entry. That is something that I'm sure worries the astronauts to death.

Dual Processing...

[Trillian_Angel]

... dare I commit a computer junkie sin, and ask what is *truly* the point in running dual processors, if when you are already running a decent (1 gig +) board and chipset... the difference is only in a few seconds? I haven't been using *nix systems for very long (I just started running Debian recently) but the first thing I noticed after switching from windows was that the sheer processing power from my 1 gig chipset (with my measly 128 ram) is already fast enough that compiling the kernel isn't a long wait at all... I'm almost afraid to know how fast it would have compiled if I had been running a dual chipset.

Now, don't get me wrong, I'd *love* to be able to use numerous copies of photoshop (if I had windows and if the version of windows I had been using even used both chips... afaik win 98 would get real cranky about 2 processing chips and just use the one), but I've never had the oppurtunity to put that to the test because I've never been able to afford a dual chipset. (so if you know how windows does with dual processors, I'd be delighted to find out.)

One thing I am wondering though, is during the test, they used the Duron chips verses the athlons... its downright obvious which chipsets going to win... If i am not mistaken, the Duron was AMD's reply to the Celeron... a "cheap" chip that lacks as much sheer power as the Athlons. (And if I'm not mistaken, the Duron gives the Celeron a good run for its money).

Now if I had the spare... howver much money it would cost for the chips and board (and some newram to go with it) I would probably buy it in a heartbeat, though I don't know what good it would *really* do a computer user like me... the most cpu intensive program I use is the gimp. (One thing I miss about windows is ps6.0...*sigh*)

What I would like to see is tests of the athlon verses intels latest prize... that would be a competition worth snickering at. I'd like to see intel get blown out of the water.

Re:Dual Processing...

[mr-spam-uk]

We use dual processor machines here as WORKSTATIONS.
Sure I can't see the point if we are talking about having a DP machine at home for your personal use, but in industry time = money. The amount of time our artists spend waiting for photoshop to filter an image or our programmers have to sit and wait for a compile to run IS an issue for us. That and maybe, if they have to wait a while we'd like them to be able to get on with something else in the mean time.
I'm a programmer, I quit frequently have 2 or 3 compiles running at the same time, as I'm hooked up to test hardware downloading a buld and testing/debugging it.

Re:Dual Processing...

[Pengo]

Yup, I was thinking the same thing to myself not too long ago. I told myself when I bought my athlong 700 I wouldn't upgrade until things hit 1+ ghz. (I know, a while ago).. then , when that time came.. I decided to wait until things hit 2+ghz. hehehe... I finally broke down and bought a athlon 1800+ w/MB + Ram and upgraded my system.. but it wasn't anything crazy... and cost less than 500 dollars for top-notch upgrade.

Now, I probably won't do anything until the CPU speeds reach 4Ghz and when I do that it will be dual processor.

I guess what I would really like to see if a PDA that could double as a workstation via docking station :) Now let the new age of rat-racing begin!

Re:Dual Processing...

[tswinzig]

... dare I commit a computer junkie sin, and ask what is *truly* the point in running dual processors, if when you are already running a decent (1 gig +) board and chipset... the difference is only in a few seconds?

If you were running something like BeOS, you would not need to ask this question.

Re:Dual Processing...

[__aaahtg7394]

personally, i use dual machines because i write software (well, i did... now i'm at a nice crappy support job taking a breather). it's nice to be able to push the machine and not have it get latent. i can compile the source code with gcc's -j3, listen to MP3s, and still surf the web comfortably on my dual celeron 400 (BP6).

in short, MP is really nice if you're a multi-tasking user that runs lots of intensive processing.

Re:Dual Processing...

[MeerCat]

dare I commit a computer junkie sin, and ask what is *truly* the point in running dual processors

If you're developing multi-threaded code (and yes, I know the debate, but for some tasks it is an easier way to make a single-cpu app more responsive) then you haven't really tested for deadlocks and synch problems until you've run it on a multi-cpu machine. Thats my excuse to the wife for buying a dual Athlon machine anyway...

T

Re:Dual Processing...

[Graymalkin]

How the fuck can you put optimized, Linux, and old days in the same sentence? For a while you had a kernel that didn't work very well and then when it did work well it didn't do a whole bunch. Now you've got a kernel with a shitload of modules that let you do all sorts of stuff. However these modules are configurable. Don't have a RAID device? Don't load the damn module. The increase in the size of the tar.gz you download is mostly from new drivers available. It's quaint to have a 10k kernel but for a PC you'd be much better off with something that actually supported your kilobuck hardware investment. QNX easily fits on a floppy with no hard drive needed because in such a compact state it doesn't have any sort of drivers that could even manage the driver for a hard drive let alone handle the intricacies of a file system. Microsoft and now Linux distros have learned that you need to provide an OS with a very broad range of support. You end up with alot of shit you don't think you need, usually because you don't even fucking know you need it. Microsoft's OS installs have gotten larger because there is a shitload of stuff they've had to add to make sure their system can run everything they say it can. Same with Linux distros, sure you can fit a VERY sparse system onto a floppy but can you actually use it? Hell no. Your whole rant about bloated code is complete crap and I doubt you've ever done much coding yourself. Computers have a very limited set of things they can do naturally, basic arithmatic and they can move stuff. In order to do something complex you need to adjust these limited operations into something meaningful, and have it work in a variety of circumstance. You can start by writing a very straight forward function. It works in small programs that don't do a whole bunch, it is "elegant". Then you use the same function in something more complex, everything goes batshit. Turns out your elegant function isn't so elegant when your system is low on memory or when called fifty umpteen million times. Then sometimes you need to add some abstraction on top of your elegant function to make it run in series with other functions, under low memory conditions, on multiple processors, being called fifty umpteen trillion times. Now apply this to any v.1 piece of code of N complexity. The size of the final compiled binary gets to be pretty big as you iron out bugs and kinks and add a feature or two your intended audience can't live without. Don't get into rewriting all of your code from scratch, that thread was shit to tell a couple days ago. Oh yeah, Quake3's engine is fully multi-threaded and runs even better on a dual processor than it does a single. The framerate doesn't skyrocket but you notice it a bit more with 12 bots set on higher difficulty levels.

Re:Dual Processing...

[Amazing Quantum Man]

Funny.... I ran an SCO ODT2 box for 7 years straight. Only downtime was to add more disk.

Austin Wintower433E (back when Austin was American made, not IPC). Tower was built like a tank. 32MB RAM, Adaptec 1742 EISA SCSI Controller. SCO ODT 2.0. Started with 1GB Fujitsu disk, added a 4GB Micropolis later.

Memory

[Anonymous Coward]

For those with large memory requirements, the Athlon MP using the Tyan boards can only go up to 3.5 GB of RAM (reliably, that is ... there are memory corruption issues with 4 GB), whereas both the Tualatin and Xeon have motherboards that can take 4 GB of RAM. Right now, this is the only thing that is a disadvantage for the Athlon systems (and the only thing that precludes my company from wholeheartedly jumping onto the Ahtlon bandwagon). As noted in the article, memory for Xeon systems is quite expensive, making a fully populated Xeon system significantly more expensive than an Athlon or Tualatin system.

Re:Memory

[RelliK]

Well gee, according to the spec, Tiger MP supports 3GB of RAM. If you are trying to cram 4GB of RAM in it you are going above the spec, so do it at your own risk. But anyway, I don't even see it as a problem. Since RDRAM is so expensive, I doubt anyone would use 4GB of it (or anywhere near). Oh, and the largest memory module I've seen is 512MB. 4 of them give you only 2GB (the boards have only 4 slots).

Really Funny Joke!

[squaretorus]

Would the cache-less version of your proposed BI be the "Athlon BI-curious"

ho ho ho

Slashdot?

[TheLink]

I figure it would be pretty useful for running something like Slashdot.

Or for running something that's being slashdotted ;).

For instance I've webapps that do about 30 pages/sec on a single processor PIII-550 (db+app on same server :( ). That's far from slashdot loads (>>60hits/s).

So if the load ever goes up, a Athlon MP 1800+ and 266MHz DDR RAM server would come in handy ;).

Definitely be useful for servers. I'll need to be reassured about thermal safety because our airconditioning isn't comfortably reliable. That said, AMD seems to be moving in the right directions, and it shouldn't be a big worry.

DB servers with Gigs of DDR RAM will kick ass. When you can do full table scans at 266MHz, who cares about the huge second level caches Sun boasts about.

Cheerio,
Link.

MP / XP

[shut_up_man]

I've been looking at upgrading to dual Athlons for the last while, and was considering running XPs in one of the new MPX motherboards, rather than paying extra for the MP Athlons. Everything I'd read pointed at them working just as well, so way pay more?

Then I see in the Bapco Sysmark test that the dual Duron setup hung in the same place each time - this is the first real evidence I've seen that running non-MP CPUs might be a bad idea... good to know.

audio/video

[lumpenprole]

A lot of posts have been asking what the point of MP boards are. I can tell you that MP boards run a/v editing programs a hell of a lot faster than single boards.

This ties in to my desire to get rid of my stupid windows box forever. Has anyone out there tried setting one of these up to do audio or video in linux? If so, I'd love to hear about it, becuase this seems to me to be the last frontier as far as home user linux apps.

What about motherboards?

[Console]

The Tiger is not overly expensive (for a Tyan), but it still almost eats up the price benefit for a dual Athlon system compared to a dual P3 one. When will someone other than the Mercedes of motherboard manufacturers release an MP-capable Athlon board?

Not that I can afford one before christmas anyway, but I'll be standing in line outside the shop immediately after newyears, and they'd better be stocked then! Orls! Heh...

Re:What about motherboards?

[hattig]

ASUS, MSI, EPoX, etc are all releasing 760MPX based motherboards within the month. These will all have 2 64-bit/66MHz PCI slots and 3/4 standard PCI slots. Some will have network on-board. Some will have 6-channel CMedia audio. Some will have ATA-RAID. Some might even have SCSI. All will support 4GB registered DDR memory, or 2GB unregistered DDR.

There are plenty of places that have listed these - TomsHardware, XBitLabs, AMDZone, HardOCP, etc etc.

Re:What about motherboards?

[dhamsaic]

When will someone other than the Mercedes of motherboard manufacturers release an MP-capable Athlon board?

Today, actually. Be looking for it on Tom's Hardware, AnandTech, etc.

Re:"AMD's Latest Multiprocessor chip?"

[nusuth]

All athlons, including the first ones, are mp capable if you build a chipset to support them. So surrent athlon mps are fourth generation of mp capable chips amd has produced.

large models

[hawk]

My models typically run for hours, if not days. Even so, they're smaller than what I want to do, due to the time I wait. Dual processors means bigger models, better papers, . . .

hawk, power junkie

Many Uses, In Fact

[BRock97]

Apart from really fast kernel compiles and stuff like that, what's the benefit of such a machine?

There are quite a few uses beyond gaming, in fact. Here are just a couple that made me invest in a dual Athlon 1600+:

• Adobe After Effects. The programs is SMP aware, and it shows. Rendering can now be done and the system doesn't go to complete pot. Saves a great deal of time in the long run.
• Blender. Not that it is SMP aware, but I can now let it render and still use the system. Another time saver.
• Adobe Premiere. When you go to output your finished product, the system is still usable.

Sure, these are pretty specialized, but for what I do, it is awesome. Plus, VC++ just works well with it. Now, don't get me wrong, there are some downsides to SMP, specially from a Linux driver standpoint. Both Soundblaster and nVidia had major issues with SMP and their respective products. Issues that have been fixed, and they were not major, as I just changed out the video card to something else and didn't have sound, so I was able to still use my Linux desktop. Just my $.02.

Uses you can sink your teeth into

[yoink!]

Some well designed profession Digital Audio Recording and Mastering software like Sequoia from Sek'd and it's big little brother Samplitude (Under Pro Audio) have amazing support for dual CPU's in a way which isn't considered in most of these reviews. These programs basically emulate a physical mixer, aux effect busses, tracks of audio and MIDI data (999 of them), and effects plug-in. When you run these applications on a dual-cpu system they allow you to assign certain functions to each cpu independantly. For example if you want to run all your effects off one cpu, that leaves the other free to do any mixer automation and file allocation (important in non-linear audio production) that world normally cause hiccups in the effected audio stream. Quite nice really and for those who need it, worth the expense.

Games on a server-type system

[Maddog_Delphi97]

How well does a system like this or a system with more than one processor (basically an SMP configuration) work with games? Assuming, of course, that the operating system is Windows 2000 or Windows XP. Linux has a few games, but the selection's not as good compared to Windows.

The reason why I'm asking is because I've heard that there are issues with games like Black & White not working correctly unless you "force" B&W to run only one processor (B&W may not be thread-safe could be a possible reason) I know other games like Quake 2 & Quake 3 Arena actually work quite well with the second processor, but doesn't "scale" to any processors than that...

Not too long ago, I seriously considered getting an SMP system like this... and decided that games probably wouldn't be too compatible with them.

Re:Games on a server-type system

[hughk]

I have 2*450MHz PIIIs, yeah, I know this is slow but I can't afford to upgrade at the moment. The graphics card is a Riva TNT2 but the audio is over a Creative Soundblaster Live. The O/S is Win 2K both pro and server. B&W runs but ocassionally crashes, always though on the audio, thus confirming what someone else has been saying here.

Re:smokin!

[sammy baby]

This may gradually become less of a problem. Some heatsink manufacturers (read - the ones making really huge blocks) are starting to abandon conventional ZIF socket clips in favor of bolting the damn thing directly to the mobo. The new Alpha PAL8045 is one such example.

Advantage - the thing is much more secure, and screwing the heatsink on and off can be a damn site easier than wrestling with those stupid clips. Disadvantage: not all mobos have holes drilled in the appropriate places, and taking a power tool to a motherboard right next to the CPU is not my idea of a relaxing pasttime.

760 MPX was released today, along with the 1900+

[spiro_killglance]

See the AMD press release,

http://biz.yahoo.com/bw/011212/110489_1.html

Asus, ABit, Gigabyte and MSI are all promising
760MPX boards.

Micon's 8MB cache SMP AMD chipset?

[Chaostrophy]

Anyone heard any new rumors about this?

Summary

[4of12]

So, although I'm a great fan of AMD's price performance ratio for the Athlon, I get from the review that there is precious little perceptible difference in performance for the three highest speed grades of Athlon.

And that memory starvation is occuring.

While RAMBUS has galled everyone by their legal tactics, I think there is a fundamental need for more memory BW for the higher clocked Athlons.

The P4, while over-hyped in the consumer marketplace for MHz, does show that memory speed helps for some apps once the memory transfer gets established.

The Athlon's great integer performance and the apparent lower latency of DDR are nice, but they don't seem to be enough at these speeds.

Apparenty room on the board

[DABANSHEE]

I remember reading somewhere that the reason why they didn't make all slots 64bit was because they couldn't afford giving up on the extra real estate

Love mine

[ZoneGray]

Highly recommended setup... just replaced a 2xP3-550 board with a 2xMP1600. It screams. Board, 2 CPUs, GF3-Ti200, 512MB ECC... under $1K and it worked the first time.

Going from a dual to a single cpu.

[BrookHarty]

I just migrated from a dual p3-800 to a single amd 1800. I regret it now. I/O is still a bitch on single CPUs, Copy a large file in the background and open another explorer window, and explorer waits. Programs load quicker, but the snappy task switching isnt there. Couple tasks using all the cpu and you feel it.

Thou on the bright side, with a gf3 ti500, im getting 100fps in tribes2 at 1024x768 with all display options set to max. 3DMark2001 actually runs in the 30+ fps in the highres demos. I dont do CAD or Modeling but so far, The cpu+gfx card
combo just tribes2 playable, its been collecting dust for a few months now.

Re:Going from a dual to a single cpu.

[TheGratefulNet]

Copy a large file in the background and open another explorer window, and explorer waits.

blaming the hardware for what is a M$ limitation?

Re:Athlon vs. Alpha

[Amazing Quantum Man]

In theory, if your dual Alpha uses the EV6, it should be possible with reflashed ROMS and a socket adapter, to drop the Athlons into the Alpha board. Remember, Athlon uses the Alpha EV6 bus.

The funny part is that AMD licenced the bus from Compaq/DEC and now Intel owns it.

Re:Athlon vs. Alpha

[Cadderly]

The other way around is something that is done by Samsung. They use the AMD 760 and 750 chipset on boards for Alpha CPUs. Samsung Alpha Boards

Re:SMP Forever

[green pizza]

I miss it. I had dual 466's @ 550. Then I fried the board and got a PIII 600. Some things were faster but it wasn't as "thick". That's the work I used to describe it.

Speed versus Torque. Like comparing a hot motorcycle to an 18-wheeler. We see the same thing on our Silicon Graphics Octane workstations at work -- some have faster CPUs, some have dual CPUs.

Ehh...

[bani]

There's no MPX benchmarks because there are no MPX boards available yet.

And the link you posted doesnt even have any MP benchmarks...

PCI-64 is not PCI-32 Compatible

[Cadderly]

I don't think that the PCI64 slots are compatible with standard PCI cards... look at this picture of a MSI board: MSI as you can see the polarity lock is reversed on the PCI-64 slots.