Intel Nehalem Photo debut

Printable View

Show 100 post(s) from this thread on one page

02-15-2008, 12:53 PM
Don_Dan

Quote:

Originally Posted by Donnie27

OK Gender Specific my bad too. We have Germans here who can go into a lot of Detail like Kids being neutral, ship, bridge being feminine and so on, it affects their articles like; is, a, an and etc.. it came from from old English like the and thou, Die, Der, Den and so on. Maybe a Processor would be Male while the Whole computer might might be Female?

hehe, correct, we have three different articles "der", "die" and "das" ( male, female, neutral ).
ship is neutral in German by the way.
So yes, every word, or better substantive, has one of the three articles, which change when it's plural etc...

Sorry for OT... :-)
02-15-2008, 12:56 PM
onewingedangel

Quote:

Originally Posted by Blauhung

http://sunnytalkstech.blogspot.com/2...ls-leaked.html
sums everything up nicely

The biggest question I have is what exactly does the tylersburg chipset contain - surely its more than just a pci-e controller/switch? Otherwise the XE will just be gaining an extra memory channel, and more than x16 pcie channels by moving the pcie controller offboard.

Edit: After further thought the market positioning clicked, see posts below
02-15-2008, 01:09 PM
Shintai

Quote:

Originally Posted by onewingedangel

The biggest question I have is what exactly does the tylersburg chipset contain - surely its more than just a pci-e controller/switch? Otherwise the XE will just be gaining an extra memory channel, and more than x16 pcie channels by moving the pcie controller offboard.

Whilst I have no problem with intel badge engineering their single socket xeon chip (which itself is a xeon DP with one quickpath channel disabled) to create an XE for the desktop, I fail to see how it will really perform much better than the mainstream chips, which will have plenty of bandwidth from dual channel ddr3, and dual pci-e2 x8 for sli/crossfire. All your really gaining is quad sli/crossfire. It really is athlon fx on socket 940 all over again.

Its nice to see intel differentiating the xeons from the regular desktop line up though - even if it means that single socket xeons won't be able to use mainstream desktop boards like the current single socket ones can.

Tylersburg is nothing else than a PCIe switch with QPI links. Also XE is targetted permantly for DP. Not singlesocket anymore.

SLI/CF is already a small niche market. So is this. Just like Skulltrail, 4x4 and so on.

AMD and Intel is going more and more platform oriented. The onboard PCIe for performance down to value is another signal of this. Even value also got an ondie IGP class GPU.

Flexibility is simply gone from anything but XE. And it will be the same from AMD in near future. Its just a matter of time before we sit with complete system on a chip.
02-15-2008, 01:13 PM
onewingedangel

I thought bloomfield is a single socket gainestown (one of the QPI links disabled)?

If XE was going to be multi socket it would use gainestown, not bloomfield.

Edit: Just realised that Bloomfield is not just going to be XE desktop, but midrange up as well.

So theres 4 nehelam cores:

1: Xeon MP - Beckton

2: Xeon/Desktop - DP - Gainsestown/SP - Bloomfield ( 1 QPI lnk disabled)

3: Value/Mobile - Lynnfield/Clarksfield

4: Integrated GPU - Havendale/Auburndale

Lynnfield threw me a bit, as its a quad core so doesn't seem much lower than the full fat core, but obviously the greater integration is more for the notebook market, but also means that value boards ban be made cheaper and smaller (one less chip on board). But Bloomfield is the chip most full sized/enthusiast boards are likely to use. Easier to think of the value desktop boards using the notebook chips (3 or 4) rather than the standard desktop ones (2)
02-15-2008, 01:22 PM
Shintai

Quote:

Originally Posted by onewingedangel

I thought bloomfield is a single socket gainestown (one of the QPI links disabled)?

If XE was going to be multi socket it would use gainestown, not bloomfield.

There is a difference on DP and MP.

Bloomsfield=2 CPUs, Gainestown=4-? CPUs. Just like 5300 and 7300 Xeons today.
02-15-2008, 01:24 PM
Shintai

Quote:

Originally Posted by THE JEW (RaVeN)

Oh, and I think 3 different sockets is cr@ppy.......no matter how the Intel PR tries to spin it.

But there is only 2 sockets....
02-15-2008, 01:32 PM
onewingedangel

Quote:

Originally Posted by Shintai

There is a difference on DP and MP.

Bloomsfield=2 CPUs, Gainestown=4-? CPUs. Just like 5300 and 7300 Xeons today.

http://sunnytalkstech.blogspot.com/2...ls-leaked.html

The link says that Gainstown is the DP chip, Bloomfield single socket, with beckton the MP chip.

Quote:

Originally Posted by Shintai

But there is only 2 sockets....

3 sockets

1: (socket 1567) Xeon MP (4 channel fb dimm, 4 qpi links)

2: (Socket 1366) Xeon DP/UP/performance desktop (3 channel ram, 1 or 2 qpi links)

3: (socket 1160) Notebook/value desktop (2 channel ram, onboard pci-e, with or without gpu) [no northbridge required]

Which is actually more logical and simpler than the current situation where the same dies using different sockets for notebook(merom), desktop(conroe) and xeon(woodcrest) markets.
02-15-2008, 01:35 PM
Donnie27

Quote:

Originally Posted by Don_Dan

hehe, correct, we have three different articles "der", "die" and "das" ( male, female, neutral ).
ship is neutral in German by the way.
So yes, every word, or better substantive, has one of the three articles, which change when it's plural etc...

Sorry for OT... :-)

Sorry on my part too! Thanks.

The only strange news I've heard or seen said Intel is not finished or completed plans for how many differrent models of Nehalem. Now if I could only find the link.
02-15-2008, 02:14 PM
LordEC911

So when did IMC on all Nelahem become official information?
There were quite a few different articles and such about IMC for only server and enthusiast.
02-15-2008, 02:26 PM
Shintai

Quote:

Originally Posted by onewingedangel

http://sunnytalkstech.blogspot.com/2...ls-leaked.html

The link says that Gainstown is the DP chip, Bloomfield single socket, with beckton the MP chip.

3 sockets

1: (socket 1567) Xeon MP (4 channel fb dimm, 4 qpi links)

2: (Socket 1366) Xeon DP/UP/performance desktop (3 channel ram, 1 or 2 qpi links)

3: (socket 1160) Notebook/value desktop (2 channel ram, onboard pci-e, with or without gpu) [no northbridge required]

Which is actually more logical and simpler than the current situation where the same dies using different sockets for notebook(merom), desktop(conroe) and xeon(woodcrest) markets.

Beckton is an EX chip. And I avoided it since basicly noone here will ever have one. Just like Tigerton is not LGA771 either.

Also 1366 is for Extreme/Servers only. No performance desktop.
1160 contains all the we usually refer to. To compare with today, Q9450, Q6700, E8x00 series etc is 1160. Only the QX series would be 1366 along with servers.
02-15-2008, 02:26 PM
Helmore

Quote:

Originally Posted by LordEC911

So when did IMC on all Nelahem become official information?
There were quite a few different articles and such about IMC for only server and enthusiast.

It has pretty much alway been like this and that's why I can't figure where you came up with this.....(sorry, I don't intent to be a bit blunt)
02-15-2008, 03:04 PM
lior307

The End of this year is gonna be very interesting
02-15-2008, 03:07 PM
THE JEW (RaVeN)

Quote:

Originally Posted by Shintai

But there is only 2 sockets....

I would have still liked them to combine 1567 and 1366 into one socket. This was done for Socket F, 604, 940, etc. That way if, by chance, you had a Beckton board but could only afford Gainestown procs initially, you could just buy a set and then upgrade later (ala Opteron 2XX vs. 8XX series).

Unlikely, granted, due to the difference in architecture, but some of us still dream of the old K7 days (where buying one cheap chip could net you a mobile, desktop, or server chip with just a defogger kit).
02-15-2008, 03:15 PM
onewingedangel

They use different memory types, and with the MP having more ram channels and twice as many qpi channels you simply need a huge ammount of pins - and an expensive socket and many layered motherboard to pair with it. The DP and MP sockets are so divergent theres no way to combine them.

Be interesting to see if Itanium ends up on s1567 though.
02-15-2008, 04:02 PM
Blauhung

Quote:

Originally Posted by Shintai

There is a difference on DP and MP.

Bloomsfield=2 CPUs, Gainestown=4-? CPUs. Just like 5300 and 7300 Xeons today.

Gainstown is DP, Bloomfield is UP.

Tylersburg is needed for scaling up the number of sockets and linking them all to the southbridge. It will also be used with the MP platforms and itanium. Think of it as the hub that keeps keeps all the extra sockets running.

Bloomfield's performance bonus will come from a few things. First off the boost in bandwidth from the 3 channel memory. Second the ability to use 2x16 PCI-e 2.0. Third, it probably stands a better chance at overclocking due to the fact that the PCI lanes are not onboard.
02-15-2008, 04:19 PM
onewingedangel

Is bloomfield UP xeon and XE desktop only, or is the crossover to Lynnfield/Clarksfield further down the product mix?
02-15-2008, 07:32 PM
Blauhung

Quote:

Originally Posted by onewingedangel

Is bloomfield UP xeon and XE desktop only, or is the crossover to Lynnfield/Clarksfield further down the product mix?

yup,

although there might be some not extreme desktop versons for those that want the 2x12 pci, i have no clue though
02-15-2008, 08:21 PM
adamsleath

more varieties of chocolate @ the store makes consuming more fun.
02-15-2008, 10:31 PM
schnulli

Quote:

Originally Posted by Helmore

It has pretty much alway been like this and that's why I can't figure where you came up with this.....(sorry, I don't intent to be a bit blunt)

hmm, no. in the end i think they just got confused with the terms IMC and CSI. so for now we can assume that IMC is a feature of the whole new nehalem family, but what about CSI?

Quote:

CSI will be used as the internal fabric for almost all future Intel systems starting with Tukwila, an Itanium processor and Nehalem, an enhanced derivative of the Core microarchitecture, slated for 2008. Not only will CSI be the cache coherent fabric between processors, but versions will be used to connect I/O chips and other devices in Intel based systems.

http://www.realworldtech.com/page.cf...WT082807020032

EDIT: that pcwatch diagram is new for me (my bad): still DMI for desktop? no tri-channel? meh...

Quote:

Originally Posted by Blauhung

Bloomfield's performance bonus will come from a few things.[...]Third, it probably stands a better chance at overclocking due to the fact that the PCI lanes are not onboard.

what did u mean by that? CSI :confused:
02-16-2008, 12:19 AM
Blauhung

Quote:

Originally Posted by schnulli

what did u mean by that? CSI :confused:

Well, as you raise the speed of a Lynnfield or Havendale, you are also playing with the on die PCI-e controler as well in terms of heat and possably clock speed. It's just one more component that is being flexed that is not ment to be.

This kinda just ocured to me, but i don't know what lever we really have to push the mainstream stuff anymore anyway. There is no more FSB in Lynnfield or Havendale, so all the cpu frequencies are most likely taken directly from the clock generator. At least with Bloomfield, the QPI has a set frequency and the CPU probably uses a multiplier off that frequency to set it's final speed.

I have no clue how the chips will work as far as clock generation, but I believe the mainstream stuff will have very few levers to overclock if any. Overclocking may just become limited on Intel platforms to the extreme side of the line.
02-16-2008, 01:15 AM
Kingcarcas

Quote:

Originally Posted by adamsleath

more varieties of chocolate @ the store makes consuming more fun.

Will Farrell speaks the truth:)
02-16-2008, 01:50 AM
schnulli

Quote:

Originally Posted by Blauhung

[...] At least with Bloomfield [...]

duh, now i got it.

but the question is do we have to go with that registered/ecc shiet? i really cannot stand the idea of the castrated bloomfield platform for the mainstream, but reg. memory isn't that great too...and does it help overclocking? :shrug:
02-16-2008, 05:56 AM
adamsleath

betta to be alone than in bad company? or is my italian sbaliato?
02-16-2008, 10:30 AM
Blauhung

Quote:

Originally Posted by schnulli

duh, now i got it.

but the question is do we have to go with that registered/ecc shiet? i really cannot stand the idea of the castrated bloomfield platform for the mainstream, but reg. memory isn't that great too...and does it help overclocking? :shrug:

it's not required. The 975x chipsets and a bunch others all have the option to turn it on or off. ECC isn't really castrated at all either.
02-16-2008, 12:52 PM
Shintai

Quote:

Originally Posted by schnulli

hmm, no. in the end i think they just got confused with the terms IMC and CSI. so for now we can assume that IMC is a feature of the whole new nehalem family, but what about CSI?

CSI is irrelevant for LGA1160 chips. Whatever they use CSI or a new name for the DMI to the southbridge doesnt matter. The old DMI was basicly a modified PCIe link at 2GB/sec.

Unlike A64, Nehalem desktops chips dont need a fast HT/CSI link to a PCIe northbridge host. Because the bandwidth demanding PCIe is ondie on the CPU.

The only thing left is SATA, sound, USB, network and such on the Ibexpeak. But I expect it to be a CSI link. Or a DMIv2(PCIe 2.0 (4GB/sec)) link. Since the value will also have to be able to carry the signals from the GPU for further distribution.

Show 100 post(s) from this thread on one page