Wednesday, April 21st 2010

Intel Sandy Bridge to Introduce New Sockets, Chipsets, Reorganize Platform Further

Intel plans a pair of new sockets for launch with its new processor architecture that succeeds Westmere, codenamed "Sandy Bridge", which are due for 2011. As part of its "tick-tock" product launch strategy, the company is currently transitioning between the 45 nm "tock" (Nehalem architecture), and 32 nm "tick" (Westmere architecture). In 2011, it will transition from the 32 nm "tick" (Westmere architecture), to the 32 nm "tock" (Sandy Bridge architecture). The company uses a "tick-tock" model of process development, where each processor architecture gets to be made in two successive manufacturing processes, while each process gets to build two succeeding architectures. It seems to have become clear that with Sandy Bridge, Intel will also switch to new socket designs, making existing motherboards obsolete then. Architecturally, Sandy Bridge will introduce new feature-sets that make the CPU more powerful, clock-to-clock, such as AVX - Advanced Vector Extensions, an evolution of the SSE instruction set, native AES engine which has been introduced with Westmere, and so on.

The present LGA-1156 package on which Intel builds value-through-performance processors including a bulk of mainstream processors, will be succeeded with the LGA-1155 package. Though similar, LGA-1155 and LGA-1156 are not inter-compatible, meaning that LGA-1155 processors will not work on existing LGA-1156 motherboards, and LGA-1156 processors will not work on LGA-1155 motherboards, either. For these processors, the arrangement of vital components is similar to the LGA-1156 package, except that every LGA-1155 processor - dual-core or quad-core - will feature an on-die display controller.

The die itself will be monolithic, where the northbridge component completely integrates with the processor component, leaving only the southbridge outside the package, on the motherboard. Currently, the "Clarkdale" and "Arrandale" dual-core processors have the processor and northbridge components on separate dies, but on the same package. LGA-1155 is also designated as socket H2 (LGA-1156 is H1), the package is identical in size to LGA-1156, but has a different pin layout and orientation notch.

Chipsets that drive the LGA-1156 platform include P67, H67, H61, and Q67, which will support features which were conceived prior to Ibex Peak platform's launch, but were shelved, such as ONFI NAND Flash "Braidwood", etc. USB 3.0 still isn't part of the feature-set, though native SATA 6 Gb/s support is on the cards.

The next big platform to succeed the LGA-1366, which caters to processors in the upper performance-though-enthusiast segments is the "Patsburg" platform, succeeding the existing "Tylersburg" based Intel X58, 5000 series chipsets. Here, Intel will introduce a massive new socket, the LGA-2011. The pin count is drastically increased for two reasons: the processor will have a 256-bit wide memory interface (quad-channel DDR3), and the northbridge component (currently X58 PCH) will be integrated completely into the processor package, upping the pin count with the PCI-Express and DMI pins. The on-die PCI-Express 2.0 root-complex will give out 32 lanes for graphics (unlike 16 lanes on the LGA-1155), and a DMI link to the so-called "Intel X68" chipset, which is relegated to being a Platform Controller Hub, just like the P55, or P67. The X68 could have a feature-set similar to the P67.Source: bit-tech.net
Add your own comment

119 Comments on Intel Sandy Bridge to Introduce New Sockets, Chipsets, Reorganize Platform Further

#1
HillBeast
by: Relayer
Not all processors come with the box. Most come inside the assembled PC. I've seen a professional animator who believed that his i7 920 had 8x 2.67GHz cores because there are 8 rendering threads and it's a 2.67GHz rated processor. He didn't build his workstation. He bought it though because the person selling it to him lead him to believe that while it wasn't really an 8 core processor, it was the same thing. I was just trying to avoid that type of misinformation. If it's that obvious though, carry on. :)
True.
Posted on Reply
#2
lukesky
[QUOTE

On applications that aren't heavily multithreaded, the architecture of Core i# is inferior to that of Core 2 and Phenom II.[/quote]No. Core i7 is faster in single threaded application. It has higher performance per clock than any other architecture. Look at how it fares in gaming compared to phenom and Core 2 Quads.
Posted on Reply
#3
DrPepper
The Doctor is in the house
by: lukesky
No. Core i7 is faster in single threaded application. It has higher performance per clock than any other architecture. Look at how it fares in gaming compared to phenom and Core 2 Quads.
Not much better. It does have a high instructions/clock rate 23.9 compared to 18.6 for a core 2 quad yorktown so yes it is faster but not by a huge margin.
Posted on Reply
#4
LAN_deRf_HA
by: lukesky


No. Core i7 is faster in single threaded application. It has higher performance per clock than any other architecture. Look at how it fares in gaming compared to phenom and Core 2 Quads.
Games? That's the last thing you should look at for a yorkfield vs i7 comparison. There's zero performance difference with single gpu setups, a slight edge only shows up for dual gpu situations. Random productivity benchmarks are where the difference is.

Look here at the bottom http://www.anandtech.com/bench/Product/45?vs=48

Two games are tied, two games where either or wins. That sums up the whole of the video game comparison... it's even.
Posted on Reply
#5
D007
All I see is marginal improvement for far to much cost.
This is a bad financial move for intel. Exactly like the ps3 was for sony.
Posted on Reply
#6
FordGT90Concept
"I go fast!1!11!1!"
by: HillBeast
I highly doubt Intel would be dumb enough to release a chip worse than it's predecessor after the fiasco they had with Netburst.
Nehalem practically is Netburst with a new dress. The only real difference is the pipelines are quite a bit shorter (the mistake in NetBurst was the assumption they could keep increasing the clockspeeds indefinitely).


Look at the most to least portion of the results here:
http://forums.techpowerup.com/showpost.php?p=1686222&postcount=119

4 times was exagerating :roll:, it was (as expected) almost twice as fast with hyperthreading enabled. No hyperthreading = very low scores except in the floating point department. It got hurt very badly in the integer department without Hyperthreading. That proves that over half of the ALU's are idle without two threads throwing work at it where the FPUs are mostly loaded with just a single thread giving it work.


As you can see, I haven't done any testing at only one thread yet.

by: lukesky
No. Core i7 is faster in single threaded application. It has higher performance per clock than any other architecture. Look at how it fares in gaming compared to phenom and Core 2 Quads.
Phenoms II win 50% of the time. Core 2 loses all the time because of the delay in reaching the RAM that is minimized in Phenom II and Core i#.
Posted on Reply
#7
Wile E
Power User
by: FordGT90Concept
Nehalem practically is Netburst with a new dress. The only real difference is the pipelines are quite a bit shorter (the mistake in NetBurst was the assumption they could keep increasing the clockspeeds indefinitely).


Look at the most to least portion of the results here:
http://forums.techpowerup.com/showpost.php?p=1686222&postcount=119

4 times was exagerating :roll:, it was (as expected) almost twice as fast with hyperthreading enabled. No hyperthreading = very low scores except in the floating point (double and single) department. It got hurt very badly in the integer department without Hyperthreading.
Numerous tests on i7 with HT disabled shows that it's faster clock for clock than both AMD and Core 2. It's all over the net. Just look at the 750's benchmarks.
Posted on Reply
#8
HillBeast
by: FordGT90Concept
As you can see, I haven't done any testing at only one thread yet.
That what I was talking about.
Also about the whole Core 2s being better than Core i7s in poorly threaded apps, it's simply not true. I wrote a program years ago for comparing P4s and it tests the CPUs per thread performance, and per MHz, Nehalem is faster than Core.
I never said it was better in multi threaded apps, but from your absolutely confusing chart all I can understand is that i7 works better when you turn on HT, which any smart person knows already. It does after all double the number of threads per core.
Posted on Reply
#9
FordGT90Concept
"I go fast!1!11!1!"
by: Wile E
Numerous tests on i7 with HT disabled shows that it's faster clock for clock than both AMD and Core 2. It's all over the net. Just look at the 750's benchmarks.
Edit: Clock for clock, Core 2 (Penryn) is about equal to Phenom II. Core i7 is clock for clock faster than Core 2 (and thus, Phenom II) in multithreading but I'm not certain about single threaded.


uint64: Core i7 920 < Phenom II 955
double: Core i7 920 > Phenom II 955 (not by much on the 4 thread test)

Core i7 920 is better at floating point operations than the Phenom II 955 but not at integer operations. If you are testing with an application that is heavy in floats, the Core i7 will come out on top (with hypthreading disabled). If the application is heavy on ints, Core i7 (hyperthreading disabled) will come in second.

Turning on hyperthreading and the workload the same (4 threads each), Phenom II 955 and Core i7 920 are very close (i7 on top) while the Core i7 920 slaughters the Phenom II 955 in the floating point area. Turn on hyperthreading and add 4 more threads and the Core i7 920 simply mocks the Phenom II 955.


by: HillBeast
I never said it was better in multi threaded apps, but from your absolutely confusing chart all I can understand is that i7 works better when you turn on HT, which any smart person knows already. It does after all double the number of threads per core.
Read the results at the bottom for analysis. The test was to compare HT to no HT, two processors versus one, and AMD versus Intel (they were about the same price when the benchmark was done).
Posted on Reply
#10
HillBeast
Right so here are some results from my testing program.

When I first wrote it, it was when P4s were all big and I made it to compare them. I first wrote it on my P4 2.4GHz Northwood and since then I have been using it on all the CPUs I have had since and here are some quick results:

code:

Arch. Processor Core Clock Score Points/MHz
Nehalem Intel Core i7 930 3800MHz 11704 3.08
Nehalem Intel Core i7 860 2860MHz 8780 3.07
Nehalem Intel Core i7 920 2793MHz 8520 3.05
Core Intel Core 2 Duo E8400 3000MHz 8677 2.89
Core Intel Core 2 Duo E6550 2333MHz 5577 2.39
Pentium M Intel Pentium M 1.86 GHz 1866MHz 3972 2.19
Pentium M Intel Celeron M 420 1600MHz 3478 2.17
K8 AMD Athlon 64 X2 4600+ 2400MHz 4748 1.98
K8 AMD Athlon 64 X2 4200+ 2200MHz 4348 1.98
K7 AMD Athlon XP Barton (Underclocked) 1722MHz 3137 1.81
K7 AMD Duron 800+ 800MHz 1330 1.66
Netburst Intel Pentium 4 540 3000MHz 3499 1.17
Netburst Intel Pentium 4 Northwood 2400MHz 2985 1.14



And if you look back at history these scores really do correlate well. K7 was faster than Netburst per clock, Pentium M was much faster than Netburst, Core was faster still. Nehalem gets the highest score therefore Nehalem is the best.

Anyone who doubts these findings are either a fanboy or a just plain stupid.
Posted on Reply
#11
FordGT90Concept
"I go fast!1!11!1!"
Wanna send me that app so I can try it?
Posted on Reply
#12
HillBeast
Here is my testing app. Ignore the Total Processor Score as it is inaccurate. It simply theorizes what the CPU could do if all the cores work perfectly together in multi threaded applications. I didn't have the know-how to write a multi threaded program back when I wrote it. The Per Thread Score is still accurate though.

Oh this is a rewrite of the original program for P4s.

Let me know what you get.
Posted on Reply
#14
FordGT90Concept
"I go fast!1!11!1!"
My Core i7 920 matches your score (within a few dozen points).

code:
Core Intel Xeon E5310 (dual processor) 1600MHz 4336 2.71


What type of calculations does this appication do? Lemme guess, floating point?


by: LAN_deRf_HA
I've never seen a review that showed Phenom II being even with core 2. Close in some things, ahead in a few select things, but lagging in the vast majority. http://www.anandtech.com/bench/Product/48?vs=88
According to that, Phenom II 965 BE and Core 2 Q9650 are well matched, Core 2 being slightly more efficient clock for clock.
Posted on Reply
#15
LAN_deRf_HA
I usually don't consider mismatched stock speed comparisons as phenom and core 2 chips have roughly equal clock speed potentials, so clock for clock is most relevant for enthusiasts.
Posted on Reply
#16
HillBeast
by: FordGT90Concept
What type of calculations does this appication do? Lemme guess, floating point?
I can't remember. It was a long time ago. I don't think so though. I'm pretty sure it was just simple arithmetic.
Posted on Reply
#17
FordGT90Concept
"I go fast!1!11!1!"
I haven't tested any Core 2s and you didn't list any Phenom IIs so it's hard to line up my multithreaded charts to your single-threaded chart. Still, note how close Core 2 (Penryn) is to Core i7 despite the major architectural changes (namely, moving the memory controllers to the chip).
Posted on Reply
#18
HillBeast
by: FordGT90Concept
I haven't tested any Core 2s and you didn't list any Phenom IIs so it's hard to line up my multithreaded charts to your single-threaded chart. Still, note how close Core 2 (Penryn) is to Core i7 despite the major architectural changes (namely, moving the memory controllers to the chip).
Yeah after my bad experience with my original Phenom, I promised myself to stay away from them. I was an AMD boy until I got my first Phenom and it really let me down.

I don't think my program is affected by memory much, hence the very low difference.
Posted on Reply
#19
FordGT90Concept
"I go fast!1!11!1!"
I predicted (back on Hardware Analysis) the Phenom was going to suck before it even had a name (it was referred to as K8L or K10 at the time). I avoided them like the plague when they did finally launch like 18 months late. I'm glad they finally got off their rubbish 65nm fab with the Phenom II and Athlon II processors though (they are decent for mainstream systems).
Posted on Reply
Add your own comment