Slashdot Mirror
What Happened To SMP For AMD processors?
Christopher Cashell asks: "Does anyone know what is going on with AMD and support for multiple processors (SMP)? I love AMD CPU's, but I've also come to love dual processor machines. Ever since the Athlon was still an 'in progress' chip code named the K7, and AMD stated that the CPU would support SMP, I've been drooling over the idea. Now, especially, I would love to have a dual CPU Duron box. Has anyone heard anything? I couldn't find anything on AMD's site about it. As I understand it, the CPU supports SMP, so it's just a chipset issue, right? Is AMD working on a SMP chipset? And if not, are any of the other big mobo/chipset manufacturers considering it?"
There have been LOTS of problems with AMD chips. Actually, more precisely, the problems have mostly been with the chipsets, which AMD themselves didn't want to make -- they were really hoping the third-party market would do that for them, but nobody stepped up to bat with a really good chipset for the Athlon until very recently.
Quite some time before the Athlon came out, I myself bought a K6-2/300 (which I still have, happily chugging away as a Linux server) and had all kinds of hassles with the Aladdin V chipset and various and sundry cards. I eventually ended up buying a BX-chipset board instead, and was much much happier with it. Even now Linux doesn't run as well as it should, as it doesn't seem to have any support for that IDE chipset and leaves all the drives in non-DMA mode. It doesn't do much work so that's okay for me, but it's a bit annoying.
With Athlon machines, the biggest problem was simply inadequate power supplies. Those chips suck power like nothing else before them (I think the new P4 will suck even more!) and if you put a nice fast GeForce DDR (another power hog) in there, many motherboards and/or power supplies were simply overwhelmed by the demand. Your purchase of that PCPAC power supply was probably the best money you spent in that machine, and may have saved you lots of trouble.
There have also been AGP driver issues with some of the Athlon chipsets, though I haven't yet owned one and don't know the details. I CAN tell you that the compatibility problems have been severe enough that I held off buying an Athlon. It sounds like the KT133 chipset has it pretty well together, finally, but I will probably hold out a bit longer and go SMP when those ship. I haven't done an SMP machine yet for myself, though I have wanted to for a long time.
Oh, another thought: it sounds like you had pretty good luck with your system, but remember that you weren't running fast 3-d graphics and/or Win9X either -- video drivers have been especially problematic. You were running against 'old' standards that are very well documented and easily testable. A lot of people are buying Athlons to game with, and gaming taxes a system harder than almost anything else you can do with it. AGP appears to be something of an evolving standard, too, so there are all sorts of niggling little details that differ from chipset to chipset, and can cause weird behavior that you would never see on a BSD-based server.
My $0.02.
Tyan is almost done with their SMP Athlon board. It should be released soon.
-- Ted tsikora@powerusersbbs.com
Should be possible with Athlon, too. You can leave CPU sockets without CPU in Intel SMP configurations and the remaining CPUs are used properly. However, you have to put in termination dummies in order to not degrade the CPU bus signal quality. Since AMD uses the Alpha way for SMP, it won't have a CPU bus but separate ports on the chipset for every CPU, so you may not need terminators for those.
In fact, they are SMP limited by the chipsets: If a chipset existed, you could run a box with 1000 Athlon processors
Actually, my understanding was that limitations in the bus protocol limited SMP machines to 14 processors, as with Alpha machines. For more than that, you use a hierarchical scheme or clustering.
It is not the CPU but the Motherboard that can't support SMP. Maybe the MB manufacturers are waiting on a reference board or chipset from AMD? Does anyone know if the Irongate chipset is the holdback or is it simply the Motherboard configuration? (I have a feeling that the AMD motherboard chipset is not ready)
--------------------
Kyle at HardOCP posted up a link to an interesting synopsis of an AMD seminar. AMD confirmed that VIA would be supplying an SMP chipset which will be able to run two Socket A CPUs.
These should be available this fall. As we all know; VIA has no problem introducing a new chipset to the motherboard manufacturers and no doubt will have no problem getting boards made with their chipset. As many of us know, the same Irongate chipset that many of us use for Slot A Athlons was perfectly capable of running the Socket A CPUs as well and was always available to motherboard manufacturers. Despite this, manufacturers still opted to use a VIA chipset instead and delay availability of their boards because of this.
AMD uses DEC's EV6 protocal, same as alphas, which is why there is noise about alhp mb/ athlon compatibility.
EV6 is an insanely scalable (1000's of processors) architecture (as compared to intels apic -- like 8 procs or something )
The kernel might not need that much work as smp alpha stuff is already in place.
Old truckers never die, they just get a new peterbilt
Plan to see SMP as production exceeds demand or at least keeps abreast of it. Also it has been widely reported that AMD is concentrating on introducing the "Sledgehammer" with an SMP MB chipset almost from introduction.
"Science is about ego as much as it is about discovery and truth " - I said it, so sue me.
The boot 'ROM' code is probably in Flash, so could be blasted with x86 code. The peripherals which have boot code have x86 code already, as the Alpha supports x86 emulation even at the lowest levels. FatPhil
Also FatPhil on SoylentNews, id 863
Can someone who has knowledge about operating systems, especially process cheduling, comment that a bit? Can users be sure that the most CPU hungry thread/process is run on the fastest CPU?
--
By the way: AMD has no plans to cripple it's Duron processors.
p ost_id=15000265 - all Durons will be SMP capable, and, AFAIK, they should take full advantage of the 133MHz / PC266 DDR motherboards.
http://www.aceshardware.com/Spades/read_news.php?
The Alpha motherboard is close, but probably won't win the cigar.
Motherboards include ROMs, for things like setup, BIOS and booting up in the first place. Alpha uses a different instruction set, so a 386-instruction chip like Athlon wouldn't read it.
The mobo vendor could probably do an AMD port of the board. API is unlikely to since their job is to sell Alpha processors, not mobos. Somebody else with an Alpha SMP mobo would be a better candidate.
The reason SMP is still unavailable for the Athlon line of CPUs is that the current chipsets have no way of supporting it.
The first Chipsets capable of supporting SMP, AMD's 760MP(a SMP-enabled version of the 760, due out very late this year or early next year) and the 770 chipset, which is expected to have support of up to 4 CPUs, and due out early next year.
As far as VIA and otehr third-party chipset manufacturers, they are still awaitin a chipset from AMD, before they can begin making their own SMP chipsets.
AMD's plans for a more advanced (4 and 8-way)chipset also had to be canceled when the company they were working with (forgot the name at the moment) decided to leave the server chipset business, leaving AMD to work on the chipsets on its own.
Stop over-analyzing your analizations
You are correct in regards to the processor's support for SMP. The current crop of Athlons (including the Athlon classic) are SMP capable.
p ost_id=10000214
In fact, they are SMP limited by the chipsets: If a chipset existed, you could run a box with 1000 Athlon processors - of course, designing such a beast would be impossible...
At any rate, Ace's Hardware has been covering AMD's products fairly diligently. They've posted several articles about the 760MP (The SMP capable Athlon chipset.) One good example is available here: http://www.aceshardware.com/Spades/read_news.php?
From what I understand, the 760MP should be finished between December and January, and on store shelves late Q1 2000.
From what I read awhile back, the next VIA chipset will have SMP for Thunderbird AND Duron. It will also have DDR support.
I bet that not long after this chipset hits that companies like Abit have dual socket A motherboards, and for an attractive price. Especially considering that 700 MHz Durons, which are in some respects (FPU) better than a 700 Mhz P3, can be had for under $90...
It sucks how slow SMP for Athlon has been coming, but I think when it does hit, Intel will lose a lot of face. The "Mustang" core Athlon is supposed to be the next stage, a chip that competes with Intel's way WAY overpriced Xeon line.
In 2000 America, is a non-lawyer truly free?
As others have said, the first SMP chipset for the K7 family is the 760MP, due out in December or January. AMD has already released the single-processor 760 chipset--which looks to be the first DDR-capable PC chipset--to motherboard manufacturers; motherboards based on the 760 should be showing up at the end of this month.
;) the market for SMP boxes is primarily business servers. This presents two issues:
.13um process--in about 8 months. The 512kb, 1Mb and 2Mb L2 cache Xeons won't move above 800MHz or so in that time. Meanwhile, the new P4 chips are *not* SMP capable (or at least there will be no SMP chipsets available for them). Itanium is a joke and will likely never be launched. Now, Foster, the "P4 Xeon" will be released, possibly as soon as January, but the large-cache versions of Foster won't be out until Q2.
It has been pointed out that one of the major reasons AMD has taken so long to get SMP going is that they already sell all of their processors anyways, and already have commitments for all the K7 chips they can make through the end of the year. This is true, but misses a more important point: with some few exceptions (i.e. nerds like us
1) Many if not most Intel SMP boxes are currently stuffed with Xeons. Thus, AMD has been perfectly content waiting for the release of their upcoming "Mustang" core tweak, featuring up to 1MB L2 cache--due out...you guessed it...in December or January--before rolling out the 760MP. Conversely, the big-L2 Mustangs without an SMP-capable chipset are dead-in-the-water.
2) By and large, business still clings to the notion of AMD as a cut-rate unreliable chip company. Despite the fact that knowledgable consumers have switched over to the cheaper faster Athlon in droves, Intel still has a nearly complete monopoly over the x86 business market. This impression of AMD as the "cheapo generic brand" persists despite Intel racking up delay after delay, errata after errata, recall after recall, embarrassment after embarrassment (i820, i840, 1.13GHz P3, Itanium) in the past year and a half. AMD knows that if they release the 760MP and it runs into one rumor of one conflict with one obscure 3D graphics card no business machine would ever contain anyways, their foray into the high-margin world of business computing is over before it began. (Never mind that Intel can release the i840, their new workstation-quality top-of-the-line chipset, with an "errata" which rendered it unusable with ECC memory!) Thus they are being very careful, and rightly so, with their validation process on this one.
Interestingly enough, if they get their act together (and purchasing departments take their heads out of their asses), AMD has a major market opportunity on their hands here. The Coppermine Xeon (i.e. a plain-old Coppermine P3 with $200 tacked on to the price) is incapable of scaling past 1GHz until Intel moves to a
That leaves AMD with a quarter as the sole supplier of GHz+ large-cache multiprocessing x86 CPUs. Will that be enough to get them into the lucrative enterprise market?? Hard to say. After all, you never get fired for buying Intel...
I have read on internet (www.aceshardware.com I think) that you can use existing alpha mb to put K7 into it; word is the K7 uses the Slot A (DEC Alpha) interface and so is supposed to support alpha style SMP. So the K7 should have nice SMP, scable up to 32 processors.
If you wanted just dual processors, this implies A UP2000 or suchlike would do the job, but I can't say for sure.
Here's a qoute from Paul Jakma
Interestingly the new *Alpha* 21264 UP1000 motherboard uses the AMD Irongate chipset.. they also have a dual 21264 UP2000 board based on a DEC chipset. So it seems K7/Alpha chipsets are interchangeable, so then K7 SMP is probably possible using the DEC chipset.
And here's a qoute from Acehardware.com:
Alpha:Slot-A:Slot-B:Athlon KH Yeap Wednesday,
June 23, 1999 (10:00 AM EST)
Ok, at the ongoing PC Expo in New York, Alpha
Processor Inc. is demo'ing its new
Alpha-21264 750 Mhz, which is expected to
come out in July. More interestingly a 1 Ghz
versions of the processor, which runs under
regular air-cool condition, is also demo'ed
along with a Slot-A motherboard, UP1000, and
a Slot-B motherboard, UP2000. For further
details check out this News.com report.
Now, a lot of people have been wondering
about the possibility of running a K7 on an
Alpha Slot-A or Slot-B motherboard. According
to Alpha, yes, this is possible. To make
things even more interesting Alpha's new
Slot-A motherboard, UP1000, uses a chipset
that is a hybrid between AMD's very own K7
chipset, Irongate (AMD-751), and ALI's
M1543C!! PC Watch Japan has a great shot of
this UP1000 motherboard. Also appears on PC
Watch is a photo of the Slot-B UP2000 and a
photo of the 1 Ghz Slot-B Alpha processor.
Special thanks to Daiki for this wonderful
tips.
So, you could try an Alpha dual-processor Motherboard but I can't give you any garuntees.
Michael
...another comment from Michael Tandy.
"Goodness me, how unlike the FBI to abuse the trust of the American public." -- The Onion