Which X58 mobo for 32 nm Westmere?

[Tzupy]

After flirting with the idea of a low-power system for some time I have decided that for my work I need the full power of Nehalem.
I plan to keep an i7 920 reasonably cool and quiet with a Mugen2, and I hope to upgrade to the 32 nm Westmere in about 18 months.
I'd like to know which X58 motherboard guarantees compatibility with the 32 nm future CPUs (Gulftown, 6 cores, LGA1366).
Currently I plan to get the Gigabyte EX58-UD4. Is there a future compatibilty reason (with 32 nm CPUs) to get the UD4P or the UD5 instead?
Thank you in advance for the answers.

[Scoop]

That's just the thing. It's not quaranteed that Westmeres will work on any of the current X58 boards.

[shleepy]

I'm going to bet that you'd need a BIOS update when the new CPU's come out, at the very least, with any of today's options. And I think it's a bit early for companies to be marketing that they will be compatible with something that's a fairly long time off (in the world of computer upgrades), so it's hard to say what will be compatible and what won't.

[Mats]

A new D0 CPU stepping will show up in march, look here.

Core to core transactions and multithreaded applications will get the largest gain over C stepping, cores are equal clock for clock as C stepping.

These will show up as stepping 5.

D stepping runs at lower thermals, much lower from what I have seen.

920 is the first announced, not sure why but other models will also arrive as D stepping.

D stepping is capable of a much higher clock frequency, it will get easier to find a 5Ghz chip and we should see a lot more around 5.5Ghz on average.

920 C0 Stepping: SLBCH

920 D0 Stepping: SLBED

[Tzupy]

Thank you for the answers, so there's no way to tell now if the 32 nm Westmeres will be supported with just a BIOS upgrade.
A particular thanks to Mats, I knew the D0 is coming, but didn't know about the improvements. I'm interested in multithreaded performance (OpenMP).
Is there any confirmation that the thermals are so much lower? Or is it just a waiting game?

[Riffer]

Looking at the state of BIOSes these days, you will be lucky to get one that works properly with current CPU's and chipsets, much less those in the future.

More likely any board purchased today will ship with a dicey BIOS, go through a couple of fixes for obvious problems, then be abandoned when the next set of boards come out.

[Mats]

Is there any confirmation that the thermals are so much lower?

Not really, but since it's FUGGER's words, I believe him.
Besides, you can wait a month, right?

[Tzupy]

I'm fed up with waiting! I already waited for the price drop on the Q9650 that I eyed 2 months ago, but now it's out of stock everywhere in my city.
I agree that it's too risky to get the D0 in one month, I could end up with a mobo that couldn't even POST with the new stepping, and I wouldn't be able to update the BIOS.
Probably the wisest move now would be to go with a Q9550 that's widely available and wait several months for the D0 AND motherboards that properly suport it.

[Mats]

Probably the wisest move now would be to go with a Q9550 that's widely available and wait several months for the D0 AND motherboards that properly suport it.

Well that's the most expensive solution if you're asking me.
Buying a C0 is a better idea, forget about D0, and wait for 32 nm and hope that it works in your board.

But I hightly doubt that a new stepping wouldn't POST in the current motherboards.

[Aris]

Its already been stated that the new westmeres will use the same x58 northbridge, but intel has had a habbit of forcing users to upgrade motherboards when they release new processors. So i wouldnt hold your breath for future compatability.

Generally if your going with intel, you need a new motherboard with every new CPU unless your only upgrading clock speeds within the same current model line-up.

[Scoop]

Its already been stated that the new westmeres will use the same x58 northbridge, but intel has had a habbit of forcing users to upgrade motherboards when they release new processors. So i wouldnt hold your breath for future compatability.

Generally if your going with intel, you need a new motherboard with every new CPU unless your only upgrading clock speeds within the same current model line-up.

I wouldn't exactly generalize it like that, Penryn worked with P35 boards with updated bios. If Westmere is to Nehalem what Penryn is to Merom, they should work, provided that the manufacturer decides to add the support for Westmere with a bios update and preferably keep updating that while supporting the new boards which will hit shelves with Westmere. I would wait it out and at least have the prices drop on DDR3.

I'm waiting for Sandy Bridge myself though

[AZBrandon]

I wouldn't exactly generalize it like that, Penryn worked with P35 boards with updated bios. If Westmere is to Nehalem what Penryn is to Merom, they should work, provided that the manufacturer decides to add the support for Westmere with a bios update and preferably keep updating that while supporting the new boards which will hit shelves with Westmere. I would wait it out and at least have the prices drop on DDR3.

I'm waiting for Sandy Bridge myself though

That's the whole "tick-tock" thing at Intel.

1) A new processor design is released for a given size (say, 65nm)
2) The same basic architecture is released, but smaller (say 45nm)
3) New architecture comes out, at the old size (45nm)
4) Die shrink of the step 3 architecture to 32nm

Wash, rinse, repeat. CPU #1 & 2 should be compatible and #3 & 4 should be compatible. Obviously when you're a big dog like Intel you can do whatever you want, but that's how it is "supposed" to work now. Two generations of CPU per socket.

[jessekopelman]

but that's how it is "supposed" to work now. Two generations of CPU per socket.

Not really "per socket." Look at how Socket 775 encompassed Prescott P4 through Pentium D through Conroe C2D through Yorkfield C2Q. That said, how many Prescott era MB work with a 45nm C2Q? That I think is really the point of Scoop's post. Socket 1336 could be around for many years, but that doesn't mean that a current Socket 1336 MB will automatically work with some future Socket 1336 CPU. It is all up to the continued support by the MB manufacturer. If you are going to take the risk, minimize it by going with a brand that has a good record for continuing to provide support for a long time.

[Mats]

but that's how it is "supposed" to work now. Two generations of CPU per socket.

Not really "per socket." Look at how Socket 775 encompassed Prescott P4 through Pentium D through Conroe C2D through Yorkfield C2Q.

Yes, per socket.
Prescott/Smithfield, Cedar Mill/Presler, Merom and Penryn for S775, that's four.
Prescott/Smithfield and Cedar Mill/Presler for S775, the older version, two.
Willamette, Northwood and Prescott for S478, three.
Yes, it happened to be the same pin count, but it's pretty much like two different sockets, except that the C2 boards can use older P4 CPU's, which is only a good thing.
The list above isn't as simple as it looks though, since there have been different bus speeds used. But still, even each bus speed have had two CPU generations.

That said, how many Prescott era MB work with a 45nm C2Q?

That's not two generations, that's four.
90 nm Prescott, 65 nm Cedar Mill, 65 nm Merom, 45 nm Penryn.

No matter how you look at it, all these sockets have had at least two CPU generations.
I think the only Intel socket this decade that supported only one CPU generation was S423.

[jessekopelman]

No matter how you look at it, all these sockets have had at least two CPU generations.
I think the only Intel socket this decade that supported only one CPU generation was S423.

I think you misunderstood me. I am not disagreeing that an Intel socket is good for at least 2 CPU generations. That is exactly why I brought up S775, which was the same socket for at least 4 generations (maybe more depending on whether you count C2Q as just part of the C2D family). My point is that socket is not the only determining factor. You need proper BIOS support too. Your old MB might have the right socket, but that doesn't mean it will support a newer CPU without a BIOS upgrade. Some vendors are better than others at providing these BIOS updates over the long periods of time.

[Mats]

Maybe I did. I just meant that the Prescott/C2Q motherboard is a bad example since there are no such boards from 2004 that works with C2Q, and it's not just a FSB or BIOS issue. They were never supposed to work together.

[jessekopelman]

Maybe I did. I just meant that the Prescott/C2Q motherboard is a bad example since there are no such boards from 2004 that works with C2Q, and it's not just a FSB or BIOS issue. They were never supposed to work together.

But I think that is what makes it an excellent example of my actual point -- that just having the right socket is not enough to ensure compatibility.

[Mats]

Maybe I did. I just meant that the Prescott/C2Q motherboard is a bad example since there are no such boards from 2004 that works with C2Q, and it's not just a FSB or BIOS issue. They were never supposed to work together.

But I think that is what makes it an excellent example of my actual point -- that just having the right socket is not enough to ensure compatibility.

Not really, you said something else first:

but that's how it is "supposed" to work now. Two generations of CPU per socket.

Not really "per socket." Look at how Socket 775 encompassed Prescott P4 through Pentium D through Conroe C2D through Yorkfield C2Q. That said, how many Prescott era MB work with a 45nm C2Q?

AZBrandon was right from the beginning. S775 does work with two CPU generations.

[Mats]

Tzupy: Have you bought it yet? The new stepping is launched today.
We'll probably see power draw numbers this week.

[Tzupy]

According to the Fudzilla article they won't be available in retail (in the USA probably) until the 25th of March, which means another 1-2 months for my country.
Link to article: http://www.fudzilla.com/index.php?optio ... &Itemid=35
I may be able to order the 920 D0 from an e-tailer like AmazonUK , but if I can't find one in 1 week I'll get the Q9550 instead.

[Mats]

How come you changed your mind for the Q9550, instead for the 920 C0? By doing that you ruin the chances of upgrading the CPU.
I don't know how expensive it is for you to order parts from Germany or Austria, but they often seem to have the lowest prices in Europe.

[Tzupy]

Because the Q9550 is widely available, including at the nearest computer shop (NOT on-line).
They also have the 920 C0, but don't have the motherboard and memory I want for the i7.
And I already have a spare Ninja2 and Skt 775 bolt-through kit. With the i7 I'd have to buy a new Mugen2, from a shop that's quite far.
It would only be worth the effort if I can get the new D0, with better thermals and improved multithreading performance.
For the next 10 days I'll probably be too busy to go shopping, but after that I'll have to sort it out.

[smilingcrow]

I don’t think anyone mentioned that the issue with compatibility is more of a hardware one with the VRM being one major problem area. The VRM design can change a lot even in the space of 1 or 2 years and it’s not an artificial thing as electrically the designs can change a lot from one generation of CPU to the next especially with a process shrink.

The major players tend to be good at providing BIOS support where possible so I think that’s a non issue if you choose one of them.

Unless the D0 stepping has BIOS values which are outside of the range of the C0 chips it should work fine with an older BIOS. The changes may be purely at the physical level of the manufacturing process so the BIOS won’t even care. Best do some research though.

[Tzupy]

Intel Gulftown 32nm 6-core *should* work with X58 boards, according to this Fudzilla article:
http://www.fudzilla.com/index.php?optio ... 3&Itemid=1

[smilingcrow]

*should* work with X58 boards

i.e. it's too early to tell.