What operating systems will run on an Athlon 64?

[MournSanity]

I'm wondering. I haven't heard of a 64-bit windows OS, but I have heard of 64-bit Linux I believe. So what kind of software will I be running on my Athlon 64 rig next year?

 

[Electrode]

There will be a 64-bit version of Windows XP released at the same time as the Athlon 64. You can also run a 32-bit OS on it.

(not-all-that-related note: I will have my dualie Opteron rig up and running by the end of next month. It will be running a 64-bit build of Linux from Scratch.)

 

[drag]

Linux can run on it, they have at least got the kernel to work and everybody is preparing for the next generation of proccessors. As soon as they start selling servers version based on AMD's 64bit chips I'd bet they would have at least a version of Redhat and/or SuSE that have been tested on it. I pretty sure that the kernel can handle the 64-bit mode, and a lot of the programs can. The rest will have to be ran in 32 bit mode. Anything more detailed then that I am not sure.

I went to SuSE's website and they have versions of their enterprise server support that they are advertising right now amd64 and optimtron, and itanium. (whatever that is suppose to mean) People are currently working on ports for Debian. IBM has a itanium server that runs linux (but nobody is buying then, windows or linux, I guess IBM wants to push it's own power4-based servers instead, and nobody wants to pay that much for a 800mhz server.) But that isn't amd64.

Server companies that are going to be putting low-end server editions running linux (dell, ibm, hp etc) are putting a lot of work into linux in order to stay competitive. And AMD isn't being stingy on the technical details to other developers and are providing virtual machine programs for developers to test on, so that I suppose most major development efforts have versions that are already designed to work on amd64. Since most everything is written in C and that is a easily portable language, all most things need is to have the compiler working for a new platform (GCC already has support as does the linux kernel) to be ported, then all they (like this is easy, or something, not) have to do is work out any bugs that may crop up and maybe do some adjustments in dealing with memory and other performance tweaking. Since linux already has versions that run on other 64-bit platforms for years and years such as the Alpha proccessor I don't see any thing that will stop linux from running on AMD64 stuff.

 

[Nothinman]

As soon as they start selling servers version based on AMD's 64bit chips I'd bet they would have at least a version of Redhat and/or SuSE that have been tested on it.

They've already started selling Opteron servers and they usually do come with RedHat or SuSe Enterprise Mega 64-bit server or whatever it's called.

I pretty sure that the kernel can handle the 64-bit mode, and a lot of the programs can. The rest will have to be ran in 32 bit mode. Anything more detailed then that I am not sure.

Linux and it's supporting software have been running on 64-bit (alpha, sparc64, PA-RISC, etc) architectures for quite a while now. Opteron will make testing on a 64-bit system easier as not everyone has access to an Alpha, but a large portion of the dirty work has been taken care of already.

 

[Armitage]

There's a difference between "run on" and "fully utilize".
Because opterons (and Athlon64) can run 32 bit apps transparently, I think most OS will do fine as is. The published opteron benchmarks are with 32bit windows. But to fully utilize the system, it needs to run as a 64 bit app. AFAIK, only Linux (and maybe *BSD?) can do that now.

 

[MournSanity]

Awesome, so I wont have to immediatly buy a new XP, and I could run linux on it. Great.

 

[Nothinman]

so I wont have to immediatly buy a new XP

You won't, but everything will be limited to a 32-bit address space, the same as you're current CPU.

 

[Armitage]

If you're not going to use the 64 bit aspects of the CPU, there probably isn't much point to going with a 64 bit CPU. `cept bragging rights I guess.

 

[Nothinman]

The 1M cache might be worth it, I know adding 2M L3 cache to my 600Mhz Alpha made a noticable difference.

 

[Armitage]

What happened to the large cache xeons? You used to be able to get up to 2MB I think, on the old PII or PIII Xeons?

I think the x86-64 has more registers also ... if the 32 bit compilers can be bothered to get smart enough to use em.

 

[jhu]

i think a freebsd for x86-64 just came out a few weeks ago

 

[Nothinman]

What happened to the large cache xeons? You used to be able to get up to 2MB I think, on the old PII or PIII Xeons?

They still exist IIRC, and the prices are probably comparable.

I think the x86-64 has more registers also ... if the 32 bit compilers can be bothered to get smart enough to use em.

I don't think it's possible for apps running in 32-bit mode to use them, that would break compatibility.

 

[Armitage]

Originally posted by: Nothinman

What happened to the large cache xeons? You used to be able to get up to 2MB I think, on the old PII or PIII Xeons?

They still exist IIRC, and the prices are probably comparable.

Picewatch & Pricescan don't list em, and it wasn't an option when I priced out a new Xeon cluster with several vendors.

Ok, here's the answer ... got to the source. Larger caches (up to 3MB) are only available on XeonMP ... for 4 or 8 way systems:

Xeon
Xeon MP

Wonder what the prices are like for those?

I think the x86-64 has more registers also ... if the 32 bit compilers can be bothered to get smart enough to use em.

I don't think it's possible for apps running in 32-bit mode to use them, that would break compatibility.

 

[Bremen]

Originally posted by: ergeorge
What happened to the large cache xeons? You used to be able to get up to 2MB I think, on the old PII or PIII Xeons?

I think the x86-64 has more registers also ... if the 32 bit compilers can be bothered to get smart enough to use em.

The orginal large cache Xeons were when Intel was still using slot packaging, so adding extra cache was easy. However when they went back to sockets having offboard cache was no longer possible so it had to be onboard, and that is a heck of a lot more expensive to produce (yields would plummet through the floor).

 

[FreakyGuy]

WOOSH

--- everything went over my head ----

no laymen speak

 

[Sunner]

Wonder what the prices are like for those?

Don't know how much the price differs between Sweden and the US, but we have to shell out around $5.000 for a 2 GHz 2MB Xeon MP or around $2.500 for a 1.5 GHz 1 MB version.

 

[Chaotic42]

Originally posted by: Bremen

Originally posted by: ergeorge
What happened to the large cache xeons? You used to be able to get up to 2MB I think, on the old PII or PIII Xeons?

I think the x86-64 has more registers also ... if the 32 bit compilers can be bothered to get smart enough to use em.

The orginal large cache Xeons were when Intel was still using slot packaging, so adding extra cache was easy. However when they went back to sockets having offboard cache was no longer possible so it had to be onboard, and that is a heck of a lot more expensive to produce (yields would plummet through the floor).

Wouldn't it be possible to make a cache socket? Have 512K on board, then have a socket next to the CPU where you could add more memory in.

 

[Nothinman]

Wouldn't it be possible to make a cache socket? Have 512K on board, then have a socket next to the CPU where you could add more memory in.

Like the old L3 cache design, back when you could buy L3 upgrade modules.

 

[Sunner]

Originally posted by: Chaotic42

Originally posted by: Bremen

Originally posted by: ergeorge
What happened to the large cache xeons? You used to be able to get up to 2MB I think, on the old PII or PIII Xeons?

I think the x86-64 has more registers also ... if the 32 bit compilers can be bothered to get smart enough to use em.

The orginal large cache Xeons were when Intel was still using slot packaging, so adding extra cache was easy. However when they went back to sockets having offboard cache was no longer possible so it had to be onboard, and that is a heck of a lot more expensive to produce (yields would plummet through the floor).

Wouldn't it be possible to make a cache socket? Have 512K on board, then have a socket next to the CPU where you could add more memory in.

Sure, but that's vastly much slower, both in terms of bandwidth and latency.

 

[Nothinman]

Sure, but that's vastly much slower, both in terms of bandwidth and latency.

Yes, but it's cheaper and still faster than main memory. It depends on what they're going for, like I said when I added 2M of L3 cache to my Alpha via an expansion slot like he mentioned it made a noticable difference in some things (i.e. kernel compiles too marginally less time) but not much in others because of the difference in bottlenecks.

 

[Sunner]

Originally posted by: Nothinman

Sure, but that's vastly much slower, both in terms of bandwidth and latency.

Yes, but it's cheaper and still faster than main memory. It depends on what they're going for, like I said when I added 2M of L3 cache to my Alpha via an expansion slot like he mentioned it made a noticable difference in some things (i.e. kernel compiles too marginally less time) but not much in others because of the difference in bottlenecks.

Yeah, the US-III's are a good example of this as well, with their rather massive 8 MB off die L2.

Seems like the trend is to move to on-die L2 caches in the 1-2 MB range along with off die caches with far more varied sizes.
Just my completely amateurish observation though