Intel Core i7 3960X (Sandy Bridge E) Review: Keeping the High End Alive
by Anand Lal Shimpi on November 14, 2011 3:01 AM EST- Posted in
- CPUs
- Intel
- Core i7
- Sandy Bridge
- Sandy Bridge E
If you look carefully enough, you may notice that things are changing. It first became apparent shortly after the release of Nehalem. Intel bifurcated the performance desktop space by embracing a two-socket strategy, something we'd never seen from Intel and only once from AMD in the early Athlon 64 days (Socket-940 and Socket-754).
LGA-1366 came first, but by the time LGA-1156 arrived a year later it no longer made sense to recommend Intel's high-end Nehalem platform. Lynnfield was nearly as fast and the entire platform was more affordable.
When Sandy Bridge launched earlier this year, all we got was the mainstream desktop version. No one complained because it was fast enough, but we all knew an ultra high-end desktop part was in the works. A true successor to Nehalem's LGA-1366 platform for those who waited all this time.
Left to right: Sandy Bridge E, Gulftown, Sandy Bridge
After some delays, Sandy Bridge E is finally here. The platform is actually pretty simple to talk about. There's a new socket: LGA-2011, a new chipset Intel's X79 and of course the Sandy Bridge E CPU itself. We'll start at the CPU.
For the desktop, Sandy Bridge E is only available in 6-core configurations at launch. Early next year we'll see a quad-core version. I mention the desktop qualification because Sandy Bridge E is really a die harvested Sandy Bridge EP, Intel's next generation Xeon part:
If you look carefully at the die shot above, you'll notice that there are actually eight Sandy Bridge cores. The Xeon version will have all eight enabled, but the last two are fused off for SNB-E. The 32nm die is absolutely gigantic by desktop standards, measuring 20.8 mm x 20.9 mm (~435mm^2) Sandy Bridge E is bigger than most GPUs. It also has a ridiculous number of transistors: 2.27 billion.
Around a quarter of the die is dedicated just to the chip's massive L3 cache. Each cache slice has increased in size compared to Sandy Bridge. Instead of 2MB, Sandy Bridge E boasts 2.5MB cache slices. In its Xeon configuration that works out to 20MB of L3 cache, but for desktops it's only 15MB. That's just 1MB shy of how much system memory my old upgraded 386-SX/20 had.
CPU Specification Comparison | ||||||||
CPU | Manufacturing Process | Cores | Transistor Count | Die Size | ||||
AMD Bulldozer 8C | 32nm | 8 | 1.2B* | 315mm2 | ||||
AMD Thuban 6C | 45nm | 6 | 904M | 346mm2 | ||||
AMD Deneb 4C | 45nm | 4 | 758M | 258mm2 | ||||
Intel Gulftown 6C | 32nm | 6 | 1.17B | 240mm2 | ||||
Intel Sandy Bridge E (6C) | 32nm | 6 | 2.27B | 435mm2 | ||||
Intel Nehalem/Bloomfield 4C | 45nm | 4 | 731M | 263mm2 | ||||
Intel Sandy Bridge 4C | 32nm | 4 | 995M | 216mm2 | ||||
Intel Lynnfield 4C | 45nm | 4 | 774M | 296mm2 | ||||
Intel Clarkdale 2C | 32nm | 2 | 384M | 81mm2 | ||||
Intel Sandy Bridge 2C (GT1) | 32nm | 2 | 504M | 131mm2 | ||||
Intel Sandy Bridge 2C (GT2) | 32nm | 2 | 624M | 149mm2 |
Update: AMD originally told us Bulldozer was a 2B transistor chip. It has since told us that the 8C Bulldozer is actually 1.2B transistors. The die size is still accurate at 315mm2.
At the core level, Sandy Bridge E is no different than Sandy Bridge. It doesn't clock any higher, L1/L2 caches remain unchanged and per-core performance is identical to what Intel launched earlier this year.
The Lineup
Processor | Core Clock | Cores / Threads | L3 Cache | Max Turbo | Max Overclock Multiplier | TDP | Price |
Intel Core i7 3960X | 3.3GHz | 6 / 12 | 15MB | 3.9GHz | 57x | 130W | $990 |
Intel Core i7 3930K | 3.2GHz | 6 / 12 | 12MB | 3.8GHz | 57x | 130W | $555 |
Intel Core i7 3820 | 3.6GHz | 4 / 8 | 10MB | 3.9GHz | 43x | 130W | TBD |
Intel Core i7 2700K | 3.5GHz | 4 / 8 | 8MB | 3.9GHz | 57x | 95W | $332 |
Intel Core i7 2600K | 3.4GHz | 4 / 8 | 8MB | 3.8GHz | 57x | 95W | $317 |
Intel Core i7 2600 | 3.4GHz | 4 / 8 | 8MB | 3.8GHz | 42x | 95W | $294 |
Intel Core i5 2500K | 3.3GHz | 4 / 4 | 6MB | 3.7GHz | 57x | 95W | $216 |
Intel Core i5 2500 | 3.3GHz | 4 / 4 | 6MB | 3.7GHz | 41x | 95W | $205 |
Those of you buying today only have two options: the Core i7-3960X and the Core i7-3930K. Both have six fully unlocked cores, but the 3960X gives you a 15MB L3 cache vs. 12MB with the 3930K. You pay handsomely for that extra 3MB of L3. The 3960X goes for $990 in 1K unit quantities, while the 3930K sells for $555.
The 3960X has the same 3.9GHz max turbo frequency as the Core i7 2700K, that's with 1 - 2 cores active. With 5 - 6 cores active the max turbo drops to a respectable 3.6GHz. Unlike the old days of many vs. few core CPUs, there are no tradeoffs for performance when you buy a SNB-E. Thanks to power gating and turbo, you get pretty much the fastest possible clock speeds regardless of workload.
Early next year we'll see a Core i7 3820, priced around $300, with only 4 cores and a 10MB L3. The 3820 will only be partially unlocked (max OC multiplier = 4 bins above max turbo).
163 Comments
View All Comments
actionjksn - Monday, November 14, 2011 - link
I'm pretty sure the motherboard makers will add the extra ports, even though the controllers aren't built into the processor or chipset.just4U - Monday, November 14, 2011 - link
mmm double standards..hechacker1 - Monday, November 14, 2011 - link
No, not double standards.This chip does outclass it's competition (50 plus percent) in some cases that are highly threaded.
It actually uses all of those transistors to be a speed daemon. Bulldozer just doesn't, even with its 2 billion transistors.
Phylyp - Monday, November 14, 2011 - link
Does the 2+ billion transistor count reflect the 2 cores that are fused also, or only the active transistors?iceman-sven - Monday, November 14, 2011 - link
I was interested in SB-E and the X79 platform, but i will skip it and continue to use my i7 965X. Maybe I go for IB-E, but it is doubtful, when the Nvidia Kepler GPU is released. What I really want is Haswell-E on something like a EVGA Classified Super Record 2 (SR-2) class Motherboard.cearny - Monday, November 14, 2011 - link
Thanks for including the Chromium build time test :)For GCC people out there, why not a Kernel build time test in the future also?
DanNeely - Monday, November 14, 2011 - link
Actually why not do a chromium build in GCC to make the two numbers more directly comparable. Doing it this way will give a 'free' article on which compiler is better.ckryan - Monday, November 14, 2011 - link
What about the corresponding release of Intel's next SSD?We had speculated that since it missed it's initial window that it would have been released on the 14th with SB-E. I guess we were wrong again.
Anyone want to field this?
xpclient - Monday, November 14, 2011 - link
Why? You have got newer multicore specific benchmarks that prove otherwise wise guy? Then share them.xpclient - Monday, November 14, 2011 - link
Here's a Jan 2010 benchmark: http://www.infoworld.com/d/windows/windows-7s-kill... Fact: You would need 8 core machines before Windows 7 can outperform XP.