PDA

View Full Version : News: Intel demonstrates quad-core server cpu


Cooper
04-16-2006, 04:40 AM
At IDF Taiwan Intel demonstrated server quad-core CPU - Clovertown. 2.0GHz and 4MB L2 shared cache on each core hits 1723 points in Cinebench 9.5 where`s same CPU with only one core enabled hits only 362 points - 4.67 times more than theoretical increase !(another proof about Intel`s FSB/cache size hunger) Clovertown and maybe Woodcrest will have double shared L2 cache of total 8MB to store data within both cores and reduce FSB utilization. Intel also plans to introduce multi-core solution - Tigerton - aimed at multi-processor servers where`s Clovertown will be only for dual-socket ones.

http://img130.imageshack.us/img130/8017/clovertown3rsgeulfkkfnx0kb.jpg

Source (http://xbitlabs.com/news/cpu/display/20060415072338.html)

BitpowerPM
04-16-2006, 05:35 AM
This result is very good for 2,00 GHz "Clovertown" prototype samples.
A Merom (2 Cores) shows 605 points @ 2 GHz clockspeed - the Clovertown DP setup (8 Cores) reach this score * 3.

An over 75% efficency with only core increases @ 2 GHz presamples is good enough.
No x86 dual-processor(!) system has that power. Opteron 2xx not, too !

Clovertown is targeted @ 2,33-2,66 GHz in early 2007. :)

nn_step
04-16-2006, 12:08 PM
Hyperthreading + Quad Core != Interesting
Two Dies == FSB strangle

Tulatin
04-16-2006, 12:36 PM
If intel ever wants to really pull ahead in the processing race, maybe they could increase the frigging FSB for once...

maratus
04-16-2006, 12:52 PM
FSB should be huge bottleneck IMO ...
But I hope I'm wrong!

2fink
04-16-2006, 01:10 PM
wait and see when everyone can buy one to test it!

Ailleur
04-16-2006, 01:21 PM
Hyperthreading + Quad Core != Interesting
Two Dies == FSB strangle

So, benchmarked am2 showed to be lacking, am2 == good
Unbenchmarked intel xeon, xeon != good.

I like this logic

nn_step
04-16-2006, 01:29 PM
So, benchmarked am2 showed to be lacking, am2 == good
Unbenchmarked intel xeon, xeon != good.

I like this logic
didn't say AM2 is good
saying that the FSB is going to kill that Quad's performance...

ORCBEAST
04-16-2006, 01:35 PM
Hyperthreading + Quad Core != Interesting
Two Dies == FSB strangle


Hyperthreading ? Thought Presler was the last to use that techonology.


Isnt this the Two Dual Core chips ? Or is that Kentsfield

nn_step
04-16-2006, 01:42 PM
Hyperthreading ? Thought Presler was the last to use that techonology.


Isnt this the Two Dual Core chips ? Or is that Kentsfield
Look at the picture, it has 8.
Yes it is.. thus the shared FSB is going to create a Boxing match between the two dual cores..

BitpowerPM
04-16-2006, 01:45 PM
Clovertown has the same theoretical bandwidth per core than Merom (10,4 GB/sec / 4 = 2,7 GB/sec)

A quad-core Opteron @ Dual-Channel registered DDR2-533 (there won't be any high registered(!) DDR2 memory with high capacity > 1 gig) will have only 2,1 GB/sec per core. :slapass:

nn_step
04-16-2006, 01:53 PM
Clovertown has the same theoretical bandwidth per core than Merom (10,4 GB/sec / 4 = 2,7 GB/sec)

A quad-core Opteron @ Dual-Channel registered DDR2-533 (there won't be any high registered(!) DDR2 memory with high capacity > 1 gig) will have only 2,1 GB/sec per core. :slapass:
:slap: Quad Core Opterons WIll NOT Be using DDR2-533 :slap:
a MIN of DDR2-667 will be used, DDR2-800 is suggested and will most likely used. but that number is COmpletely accurate for Intel's Quad Core and its pathetic FSB

mr_knowitall15
04-16-2006, 01:59 PM
you mean they wont use DDR3? :p:

2fink
04-16-2006, 02:18 PM
you mean they wont use DDR3? :p:

owned :slapass:

BitpowerPM
04-16-2006, 02:22 PM
:slap: Quad Core Opterons WIll NOT Be using DDR2-533 :slap:
a MIN of DDR2-667 will be used, DDR2-800 is suggested and will most likely used. but that number is COmpletely accurate for Intel's Quad Core and its pathetic FSB
Do you find any Registered ECC DDR2-800 >= 1 GB on the market ? :D

http://www.micron.com/products/modules/ddr2sdram/partlist.aspx?pincount=240-pin&version=Fully%20Buffered&package=DIMM

Micron is sampling 4 GB FB-DIMM @ DDR2-667 speed - so a Bensley Plattform System can easily scale up to 128 GB quad-channel DDR2-667 (32 GB / 8*4 GB modules per channel) memory.

Starscream
04-16-2006, 02:45 PM
Do you find any Registered ECC DDR2-800 >= 1 GB on the market ? :D

http://www.micron.com/products/modules/ddr2sdram/partlist.aspx?pincount=240-pin&version=Fully%20Buffered&package=DIMM

Micron is sampling 4 GB FB-DIMM @ DDR2-667 speed - so a Bensley Plattform System can easily scale up to 128 GB quad-channel DDR2-667 (32 GB / 8*4 GB modules per channel) memory.


that it doesnt exist yet doesnt mean it nvr will. it simply means there isnt yet any demand for it.

but when the launch of a Quad core Opteron wich uses DDR2 comes near (wich will benefit from it) a demand will be created and RAM wich meets the specs will b made and launched.

Absolute_0
04-16-2006, 03:05 PM
Very impressive! I was under the impression that AMD would be the first to get to quad core, because the AMD64 architecture handles multiple cores better than Intels. Is this not the case?

Starscream
04-16-2006, 03:12 PM
Very impressive! I was under the impression that AMD would be the first to get to quad core, because the AMD64 architecture handles multiple cores better than Intels. Is this not the case?


because of HTT etc AMD should b better with Quad cores.

but that Intel shows these quad cores of doesnt mean they will have em for sale before AMD does.

BitpowerPM
04-16-2006, 03:15 PM
Clovertown (two Woodcrest/package) is only the first but nice case study about Quad-Core and powerful enough. Tigerton/Whitefield (Xeon MP) with Common System Interface (kind of ring enabled HTT) will be the first real Quad-Core solution.

nn_step
04-16-2006, 04:53 PM
Very impressive! I was under the impression that AMD would be the first to get to quad core, because the AMD64 architecture handles multiple cores better than Intels. Is this not the case?
Look at dual cores.. all they did was put two dies in a package, requires NO research and it could come out EXTREMELY fast. In contrast AMD designed their Dual core to effectively share resources. It wasn't for almost a year later before Intel released A TRUE dual core solution.. not the Improvised piece of Crap that they launched and called Dual core...
Just because it has 4 cores doesn't mean it is a real Quad core set up..
It is just a boxing match on the FSB.

mesyn191
04-16-2006, 06:05 PM
Guys, current Opteron based systems can beat this score. Its very very good though for a single socket system. AFAIK not much Intel can do to improve thier FSB, they need to completely redo it, which is what CSI was supposed to do but that has been held back to early 2008 IIRC.

coldpower27
04-16-2006, 06:25 PM
This isn't a bad showing though as Clovertown is designed to be placed in the Xeon Dual Porcessor arena. At only 2.0GHZ it's delivering these scores fairly impressive.

I don't think Clovertown will much FSB problem as 1333/4 is 333 per core which is same as Merom which won't kill it.

Tigerton Xeon MP, may have a problem though if it's limited to Dual Independent Bus, it would be better for them to move to CSI soon at that front.

JumpingJack
04-16-2006, 09:38 PM
This isn't a bad showing though as Clovertown is designed to be placed in the Xeon Dual Porcessor arena. At only 2.0GHZ it's delivering these scores fairly impressive.

I don't think Clovertown will much FSB problem as 1333/4 is 333 per core which is same as Merom which won't kill it.

Tigerton Xeon MP, may have a problem though if it's limited to Dual Independent Bus, it would be better for them to move to CSI soon at that front.

There are going to be a few products between now and then that up the ante in the 4 way and 8 way space, of which, specialized chipsets are going to help open up the bandwidth to the processor, Dual FSB, Snoop filtering (?? not sure if it will make it though), 1333 MHz, FB-DIMMS.... some interesting reads:

http://www.realworldtech.com/page.cfm?ArticleID=RWT110805135916

http://www.realworldtech.com/page.cfm?ArticleID=RWT021906030756

Tulsa will be a power hog, though not sure of it's acceptance, Bensley should provide a stop gap to Clovertown and Woodcrest will help ease over to the new uArch. But CSI or some new interconnect will certainly be needed, for Intel the sooner the better. Anything they muck around with on FSB are just stop-gaps to buy time.

onewingedangel
04-17-2006, 01:00 AM
just to clear up some missunderstanding above intels 'core' architecture does not have hyperthreading, the eight threads are from two quad core chips in a dual socket system.

Cooper
04-17-2006, 11:13 AM
just to clear up some missunderstanding above intels 'core' architecture does not have hyperthreading, the eight threads are from two quad core chips in a dual socket system.

Exactly.

BitpowerPM
04-17-2006, 11:17 AM
Guys, current Opteron based systems can beat this score.
link ?

With four(!) Opteron 8xx ok, but not with 2xx setups.

We are talking here about systems with only 2(!) sockets, no more.

ORCBEAST
04-17-2006, 01:26 PM
Look at the picture, it has 8.
Yes it is.. thus the shared FSB is going to create a Boxing match between the two dual cores..

Yea my wireless internet wasnt working. It was so slow the picture never came up.

Explains all ^__^
I read Presler was the LAST HT processor using that techonology. But have no way of backing it up.


Hmm my friend says its got no HT.
And I Quote:

Dual sockets, 2x 4 cores = 8 cores

Merom, Conroe, Kentsfield all dont use HT.