The Sandy Bridge Previewby Anand Lal Shimpi on August 27, 2010 2:38 PM EST
Update: Be sure to read our Sandy Bridge Architecture Exposed article for more details on the design behind Intel's next-generation microprocessor architecture.
The mainstream quad-core market has been neglected ever since we got Lynnfield in 2009. Both the high end and low end markets saw a move to 32nm, but if you wanted a mainstream quad-core desktop processor the best you could get was a 45nm Lynnfield from Intel. Even quad-core Xeons got the 32nm treatment.
That's all going to change starting next year. This time it's the masses that get the upgrade first. While Nehalem launched with expensive motherboards and expensive processors, the next tock in Intel's architecture cadence is aimed right at the middle of the market. This time, the ultra high end users will have to wait - if you want affordable quad-core, if you want the successor to Lynnfield, Sandy Bridge is it.
Sandy Bridge is the next major architecture from Intel. What Intel likes to call a tock. The first tock was Conroe, then Nehalem and now SB. In between were the ticks - Penryn, Westmere and after SB we'll have Ivy Bridge, a 22nm shrink of Sandy.
Did I mention we have one?
While Intel is still a few weeks away from releasing Sandy Bridge performance numbers at IDF, we managed to spend some time with a very healthy sample and run it through a few of our tests to get a sneak peak at what's coming in Q1 2011.
The naming isn’t great. It’s an extension of what we have today. Intel is calling Sandy Bridge the 2nd generation Core i7, i5 and i3 processors. As a result, all of the model numbers have a 2 preceding them.
For example, today the fastest LGA-1156 processor is the Core i7 880. When Sandy Bridge launches early next year, the fastest LGA-1155 processor will be the Core i7 2600. The two indicates that it’s a 2nd generation Core i7, and the 600 is the model number.
|Sandy Bridge CPU Comparison|
|Base Frequency||L3 Cache||Cores/Threads||Max Single Core Turbo||Intel HD Graphics Frequency/Max Turbo||Unlocked||TDP|
|Intel Core i7 2600K||3.4GHz||8MB||4 / 8||3.8GHz||850 / 1350MHz||Y||95W|
|Intel Core i7 2600||3.4GHz||8MB||4 / 8||3.8GHz||850 / 1350MHz||N||95W|
|Intel Core i5 2500K||3.3GHz||6MB||4 / 4||3.7GHz||850 / 1100MHz||Y||95W|
|Intel Core i5 2500||3.3GHz||6MB||4 / 4||3.7GHz||850 / 1100MHz||N||95W|
|Intel Core i5 2400||3.1GHz||6MB||4 / 4||3.4GHz||850 / 1100MHz||N||95W|
|Intel Core i3 2120||3.3GHz||3MB||2 / 4||N/A||850 / 1100MHz||N||65W|
|Intel Core i3 2100||3.1GHz||3MB||2 / 4||N/A||850 / 1100MHz||N||65W|
The names can also have a letter after four digit model number. You’re already familiar with one: K denotes an unlocked SKU (similar to what we have today). There are two more: S and T. The S processors are performance optimized lifestyle SKUs, while the T are power optimized.
The S parts run at lower base frequencies than the non-S parts (e.g. a Core i7 2600 runs at 3.40GHz while a Core i7 2600S runs at 2.80GHz), however the max turbo frequency is the same for both (3.8GHz). GPU clocks remain the same but I’m not sure if they have the same number of execution units. All of the S parts run at 65W while the non-S parts are spec’d at 95W.
|Sandy Bridge CPU Comparison|
|Base Frequency||L3 Cache||Cores/Threads||Max Single Core Turbo||Intel HD Graphics Frequency/Max Turbo||TDP|
|Intel Core i7 2600S||2.8GHz||8MB||4 / 8||3.8GHz||850 / 1100MHz||65W|
|Intel Core i5 2500S||2.7GHz||6MB||4 / 4||3.7GHz||850 / 1100MHz||65W|
|Intel Core i5 2500T||2.3GHz||6MB||4 / 4||3.3GHz||650 / 1250MHz||45W|
|Intel Core i5 2400S||2.5GHz||6MB||4 / 4||3.3GHz||850 / 1100MHz||65W|
|Intel Core i5 2390T||2.7GHz||3MB||2 / 4||3.5GHz||650 / 1100MHz||35W|
|Intel Core i3 2100T||2.5GHz||3MB||2 / 4||N/A||650 / 1100MHz||35W|
The T parts run at even lower base frequencies and have lower max turbo frequencies. As a result, these parts have even lower TDPs (35W and 45W).
I suspect the S and T SKUs will be mostly used by OEMs to keep power down. Despite the confusion, I like the flexibility here. Presumably there will be a price premium for these lower wattage parts.
Post Your CommentPlease log in or sign up to comment.
View All Comments
tech6 - Friday, August 27, 2010 - linkMore speed with less power - it looks like a very competitive product. I really hope that AMD has something up their sleeve with Bulldozer and Bobcat to compete with Sandy Bridge.
killless - Friday, August 27, 2010 - link17% higher performance is just not exciting.
You need to give me 50% improvement at least to make me want to spend $1000 for new CPU/Motherboard/memory.
It really hasn't been all that exciting since Core 2 Quad...
tatertot - Friday, August 27, 2010 - linkI take it turbo was also disabled on the rest of the parts used to compare, right?
Anand Lal Shimpi - Friday, August 27, 2010 - linkTurbo was enabled on everything else - SB performance should be higher for final parts.
tatertot - Friday, August 27, 2010 - linkOh!
Well that puts the IPC gains of Sandy over Westmere at something like 20% then, considering the 880 turbos up to 3.73GHz on single-threaded work.
That's pretty impressive.
Drag0nFire - Friday, August 27, 2010 - linkI just want to say first of all, this totally made my Friday! I love previews of upcoming architectures!
Any news on the roadmap for the mobile variant of Sandy Bridge? Or do I have to wait til IDF?
Jamahl - Friday, August 27, 2010 - linkWhat system was this benchmarked on Anand?
Anand Lal Shimpi - Friday, August 27, 2010 - linkClarkdale - those charts were actually pulled from here, just with the SB numbers added:
We didn't have the system for long enough to rerun the tests with the 5450 on the H67 board. The 5450 is GPU bound at those resolutions/settings however:
Those numbers were generated with a Core i7 920.
Anand Lal Shimpi - Friday, August 27, 2010 - linkI just ran a sanity check on the Core i7 880 with the 5450, the numbers don't move by more than the normal test margins - the 5450 is totally GPU bound here.
ESetter - Friday, August 27, 2010 - linkDo you know if any of the benchmarks make use of AVX instructions? Sandy Bridge effectively doubles the maximum throughput for compute-intensive operations like SGEMM and DGEMM. While it might not translate to a 2x speedup in real-world applications, I imagine it should give a significant gain, at least in the HPC field.