backtop


Print 34 comment(s) - last by hstewarth.. on Sep 2 at 10:17 AM

Netburst still has life left in it

Intel has officially released its 65nm Tulsa core Xeon 7100 series processors. The dual-core processors are based on Intel’s Netburst architecture and will drop into existing LGA771 motherboards. Intel is catering the Xeon 7100 series to multi-processor servers. The new Xeon 7100 series has new features such as up to 16MB of shared L3 cache and Intel Virtualization Technology. DailyTech previously reported details of Intel’s Tulsa.

Eight Tulsa core Xeon 7100 models are available with varying cache sizes of 16MB, 8MB and 4MB. Xeon 7100 models also have varying front-side bus speeds of 800 MHz and 667 MHz. Two different models with different thermal data power ratings of 95W and 150W are available. The Xeon 7100M models will have 150 watt TDPs while the 7100N models will have 95 watt thermal envelopes.

Intel Xeon 7100 series processors are available from 2.5 GHz to 3.4 GHz. Pricing starts at $856 for the lower Xeon 7110N and tops out at $1,980 for the top of the line Xeon 7140M model. Servers based on Intel Xeon 7100 series processors are expected from the likes of Dell, HP and IBM.  



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

16MB = awful latency
By ForumMaster on 8/30/2006 6:15:08 AM , Rating: 2
even with the shared cache, 16MB will have awful latency. cool though that Intel took another page from AMD with a CPU that drops into the "old" socket of current Xeons. Intel really likes to bring new sockets with every CPU generation.




RE: 16MB = awful latency
By Furen on 8/30/2006 8:25:40 AM , Rating: 2
Awful latency? Much better than hitting the northbridge for data/cache coherency. While L3 latency is indeed worse than L2 latency (and L1, for that matter) it's a whole lot better than going off-chip.


RE: 16MB = awful latency
By Chillin1248 on 8/30/2006 8:29:37 AM , Rating: 3
Also what do you mean borrowed from AMD?

You don't call AMD releasing Sockets 754, 940, 939, AM2 within... Three or four years of each other pretty rapid? How long has socket 478 and LGA-775 been around for?


RE: 16MB = awful latency
By Goty on 8/30/2006 8:32:18 AM , Rating: 2
LGA-775 has been around for maybe two years, I think. And you forgot s479 in there, too.


RE: 16MB = awful latency
By Thorburn on 8/30/2006 10:47:49 AM , Rating: 2
On the desktop Intel have had 3 sockets since the introduction of the P4 in 2001, Socket 423, 478 and LGA775.

The issue has been new processors using the same socket but requiring a new chipset (e.g. 945/955 for dual cores, 915/925 not being supported).
Socket 479 was a mobile socket, not one designed for use in desktop.

AMD have had 462, 754, 939, 940, AM2 in the same time frame.


RE: 16MB = awful latency
By mino on 8/30/2006 1:46:39 PM , Rating: 2
Yeah,
Intel
PIII generation 2 desktop physical slots/sockets -> 4 different electrical specs
P4 generation 3 desktop physical sockets -> 7! different electrical specs
Core generation 1 desktop physical socket -> 1 electrical spec

AMD
K7 generation 2 desktop physical slots/sockets -> 3 different electrical specs
K8 generation 2 desktop physical sockets -> 2! different electrical specs
1 server/FX physical socket - borrowed form server space
K8 DDR2 gener. 1 desktop physical socket -> 1 electrical spec

sumarized:
Intel: 6 sockets/12 platforms
AMD: 5(6)sockets/6(7) platforms

s940 was never a real desktop socket so it should not be considered(940 custemers got pretty long upgrade path on Opteron line inmstead FX's)


RE: 16MB = awful latency
By mino on 8/30/2006 1:49:22 PM , Rating: 2
Sorry for (un)readability.
I forgot that the editing system will screw the spacing :(


RE: 16MB = awful latency
By Griswold on 8/30/2006 1:29:41 PM , Rating: 2
Lets see.

754 was the budget socket
939 was the performance socket
940 was, initially and only for a short time, the socket for A64 and served Opterons for years, like 939.
AM2 came to unify all this, at least in the 1p segment regardless of type.

They didnt change sockets around randomly.


RE: 16MB = awful latency
By DallasTexas on 8/30/2006 8:43:40 AM , Rating: 2
"..even with the shared cache, 16MB will have awful latency. cool though that Intel took another page from AMD with a CPU that drops..blah blah..."

Ah yes, the old 'but it needs an integrated memory controller' argument. Isn't that getting old?
Never mind Tulsa is breaking world records on 4-socket database performance. 32 CPu scalable, drop in, lowest performance per watt for this line of product. Nahh.. gotta have that integrated memory controller says the myspace engineer from Daily Tech. This forum is a treat.


RE: 16MB = awful latency
By mino on 8/30/2006 1:18:19 PM , Rating: 2
Agreed, with IBM chipset. This HUGE chipset has cache-coherency tables(~64MB!) integrated into the NB.
With Intel's 8500 chipset Tulsa gets smashed by Opteron.
Also remember that some DB tests for kids are asume that CPU's with >2MB cache do not exist...

Without IMC(or that wonderfull IBM's chipset) Tulsa has no chance to compete in serious workloads(10GB+).

That is a fact, no pun intended.


RE: 16MB = awful latency
By DallasTexas on 8/30/2006 8:44:37 AM , Rating: 1
"..even with the shared cache, 16MB will have awful latency. cool though that Intel took another page from AMD with a CPU that drops..blah blah..."

Ah yes, the old 'but it needs an integrated memory controller' argument. Isn't that getting old?
Never mind Tulsa is breaking world records on 4-socket database performance. 32 CPu scalable, drop in, HIGHEST performance per watt for this line of product. Nahh.. gotta have that integrated memory controller says the myspace engineer from Daily Tech. This forum is a treat.


RE: 16MB = awful latency
By Phynaz on 8/30/2006 9:49:39 AM , Rating: 1
Seen this? In the real world where these chips are used they are squashing anything else out there.

http://www50.sap.com/benchmarkdata/sd2tier.asp

Who cares about what the latency may be when it's the fastest thing around?


RE: 16MB = awful latency
By Fenixgoon on 8/30/2006 10:03:38 AM , Rating: 2
if you bothered to actually READ the system setups, you'd notice that the Tulsa 8CPU/16core setup has roughly double the score of a 4CPU/8Core AMD Opteron Rig (not to mention that Tulsa one has 4x the memory). Your "squashes everything" theory is total crap, when it's really an expected jump.

Plus, is an opty 2.6 the fastest one available?


RE: 16MB = awful latency
By hstewarth on 8/30/2006 10:26:25 AM , Rating: 2
That funny you also see that a Dual Woodcrest system 5160 will be 4 cpu AMD 2.2. Just imagine when Xeon MP are based off Core 2 technology. I believe its early next year - think of how fast it went beent 50xx and 51xx series.


RE: 16MB = awful latency
By RMSe17 on 8/30/2006 11:11:32 AM , Rating: 2
I dont see why anyone would get a xeon that is not a woodcrest...

MacPro's have had dual woodcrest xeon in them for few weeks already.


RE: 16MB = awful latency
By Phynaz on 8/30/2006 11:41:50 AM , Rating: 2
quote:
if you bothered to actually READ the system setups


Here let me spell it out for you, since YOU are the person that can't read.

Xeon 7140 16 core - 335330 LPH
Xeon 7140 8 core - 213000 LPH
Opteron 885 8 core - 170330 LPH

I'll even help you with the math - that's 25% faster for an equivalent system.

Your comprehension is what's "crap".


RE: 16MB = awful latency
By Fenixgoon on 8/30/2006 12:53:10 PM , Rating: 2
so how do 8 opterons with 16 cores total do? oh wait, that isn't in the benchmark.


RE: 16MB = awful latency
By mino on 8/30/2006 1:32:33 PM , Rating: 2
You've got it a bit screwed, I'll try to clear it up a bit:

IMB 3950/Xeon 7140/IMB chipset/DDR2 - 8way/16core - 335330 LPH
HP DL580/Xeon 7140/Intel chipset/FBD - 8way/16core - 213000 LPH
HP DL585/Opteron 885/no chipset!/DDR - 4way/8core - 170330 LPH

*HP DL585/Opteron 8220/no chipset/DDR2- 4way/8core - 210000 - my estimate


RE: 16MB = awful latency
By Phynaz on 8/30/2006 1:41:10 PM , Rating: 2
Jesus people can't read.

Here's a cut and paste.

HP ProLiant DL580 G4, 4 processors / 8 cores / 16 threads, Intel XEON 7140M 3.4 GHz, 16 KB L1 cache and 1 MB L2 cache per core, 16 MB L3 cache per processor

That system processes 213000 LPH.


RE: 16MB = awful latency
By Phynaz on 8/30/2006 1:42:09 PM , Rating: 2
That was for mino btw.


RE: 16MB = awful latency
By mino on 8/30/2006 1:51:18 PM , Rating: 2
What about this ?

HP ProLiant DL580 G4, 4 processors / 8 cores / 16 threads, Intel XEON 7140M 3.4 GHz, 16 KB L1 cache and 1 MB L2 cache per core, 16 MB L3 cache per processor


RE: 16MB = awful latency
By mino on 8/30/2006 1:53:00 PM , Rating: 2
LOL, just read after myself - I take that back, that "4" was in another line a I missed it.

My mess!


RE: 16MB = awful latency
By mino on 8/30/2006 1:56:17 PM , Rating: 2
To sum it up.

4-way 7140 is comparable to 4-way 8220 on 32GB configs.

Sounds like a tie for Tulsa - nice result indeed.


RE: 16MB = awful latency
By Phynaz on 8/30/2006 2:17:12 PM , Rating: 2
quote:
4-way 7140 is comparable to 4-way 8220 on 32GB configs


Ummmm....It's 25% faster.



RE: 16MB = awful latency
By mino on 8/30/2006 4:52:06 PM , Rating: 2
No, 885 with DDR is 20% slower.
8220 should be on par. 5% slower at worst.


RE: 16MB = awful latency
By Phynaz on 8/30/2006 1:50:27 PM , Rating: 2
quote:
so how do 8 opterons with 16 cores total do? oh wait, that isn't in the benchmark.


How much effort do you put into being so dense?


RE: 16MB = awful latency
By mino on 8/30/2006 1:22:13 PM , Rating: 2
As I said above.

The winner of those tests is NOT the Tulsa!

It is the IBM chipset !!!


RE: 16MB = awful latency
By Assimilator87 on 8/30/2006 3:07:00 PM , Rating: 2
I don't see anything about chipsets in that table of benchmark scores. Would you mind pointing it out for me?

BTW, this is the first news post where every reply has been for a single topic/thread. Wierd.


RE: 16MB = awful latency
By mino on 8/30/2006 4:56:33 PM , Rating: 2
IBM uses mostly its own chipsets, in high-end machines pretty mauch only its own. they are superior to Intel's 8500 series.
HP uses Intel 8500series - because no other chipset (except IBM's) is available.
You can verify this on product pages of the systems listed.
Yeah, that thread thingie is funny.


RE: 16MB = awful latency
By hstewarth on 8/30/2006 9:13:16 PM , Rating: 2
Like AMD making CPU's that drop into Intel sockets - I actually had one of those AMD 25mhz 286 - and I believe my manufactor didn't even tell me it was from AMD.

I don't believe the 16MB cache has bad latency.. Intel has inteligent caching system - the whol idea of large shared cache is reduce expensive access to main memory which is lot slower than the cache memory. Since a lot of time application are moving though loops - the cpu doesn't even have to touch the memory controller.


RE: 16MB = awful latency
By hstewarth on 8/30/2006 9:15:57 PM , Rating: 2
Oops - it actually was 25Mhz 386 cpu. But I believe that AMD also had 286.

I also had a 133Mhz 586 chip that was a drop in with Intel 486 system.


"And boy have we patented it!" -- Steve Jobs, Macworld 2007

Related Articles
16MB of L3 Cache: Intel's "Tulsa"
May 28, 2006, 7:12 PM













botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki