facebook rss twitter

Review: Intel Core i7 3960X Extreme Edition CPU

by Tarinder Sandhu on 14 November 2011, 09:27 3.5

Tags: Intel (NASDAQ:INTC)

Quick Link: HEXUS.net/qa72p

Add to My Vault: x

Final thoughts and rating

Utterly predictably, the Intel Core i7 3960X CPU becomes the fastest consumer chip on the market, wrestling the title away from the last-generation Core i7 990X. The margin isn't as large as you may expect in some benchmarks, but design similarities mean the fancy-pants Core i7 3960X only shows the previous champ a clean pair of performance heels in apps where the Sandy Bridge architecture, upon which Core i7 3960X's based, is plainly better - single-threaded programs being obvious examples.

Much like the chip it replaces as the chief, the performance microarchitecture, full of cores and cache, readily fits into many markets, including entry-level server, workstation and ultra-high-end desktop. This is a chip whose $999 pricing causes befuddlement to many readers, but it's manifestly not for them. This is technology reserved for those where the financial burden is a largely secondary concern.

But the £775 layout isn't all. Core i7 3960X requires a new motherboard, a 'snip' at £200-plus, and, perhaps, more DDR3 memory to fill those four channels. Transitioning to a 3960X backbone won't leave any change out of a grand though any purchaser will be sure to bask in the refulgent glow of owning the fastest system around, more so if overclocked.

Sage advice would be to opt for a Core i5 2500K and overclock it to 4.8GHz - it's plenty fast enough for practically all everyday tasks. Heck, even the similar Core i7 3930K would be a better bet, all things considered. Yet those who contemplate ultimate LGA2011/X79 goodness care little for the safe, conservative play. Instead, and you know who you are, beg, borrow or steal £1,000 for the platform; there ain't nothing better on the horizon for a long while yet.

Is sheer performance enough? Clearly impressed as we are by the speed, the release actually leaves us a touch deflated. Intel has a rich history of innovating at every turn, bringing out new features and benefits from launch to launch, but there's no genuinely compelling reason to consider this platform other than insane speed. What's more, this isn't even the full SNB-E core - two cores and 5MB of L3 cache are purposely disabled. Perhaps it's merely the standard set by regular Sandy Bridge almost a year ago, which impressed us greatly, that this release gives us a feeling of 'good, yes, but not great.'

Bottom line: the fastest consumer CPU this side of Intel's research labs, created by giving mainstream Sandy Bridge performance steroids. It's not cheap, but it was never going to be, was it?

The Good

Fastest consumer chip in the world
Sandy Bridge architecture's potential extended

The Bad

Not a huge leap over 990X in many common apps
Expensive
Needs additional outlay for X79 motherboard

HEXUS Rating

3.5/5
Intel Core i7 3960X Sandy Bridge Extreme

HEXUS Awards


Intel Core i7 3960X Sandy Bridge Extreme

HEXUS Where2Buy

Please see here.

HEXUS Right2Reply

At HEXUS, we invite the companies whose products we test to comment on our articles. If any company representatives for the products reviewed choose to respond, we'll publish their commentary here verbatim.



HEXUS Forums :: 11 Comments

Login with Forum Account

Don't have an account? Register today!
I don't understand something. Page 3 with the chipset diagram you state:

“Chip-to-northbridge bandwidth drops by about 20 per cent as Intel switches from QPI to DMI 2.0”

That seems quite incongruous. ‘Northbridge’ in X58 just contained the connection to the PCI-E cards. In X79 these are direct to CPU and bandwidth is correspondingly higher.

CPU to PCH (storage etc.) in X58 was via a 2GB/s DMI link. In X79 it's a 20GB/s DMI link.

So ‘northbridge’ to ‘northbridge’ comparison X79 is 40GB/s vs X58's 25GB/s (both bi-directional), and a 60% increase. A PCH to PCH comparison X79 is 20GB/s vs X58's 2GB/s, a 1000% increase.

So where does Hexus get its 20% decrease in bandwidth figure from?

Otherwise, great review :)
Nope, you're right, I was inadvertendly comparing QPI to DMI v2.0. That's all well and fine from a specification standpoint but that's not how, as you've correctly pointed out, the chipset's topography works.
Will the multi-processor Xeon varients feature QPI or DMI for the processor-to-processor links?

You would think there are enough PCIe lanes on each chip to do away with QPI entirely - as you could use half each chips PCIe lanes in dual-processor systems and three quarters in 4-way systems for chip-to-chip communication without sacraficing any lanes for external communication - but I imagine drop-in compatability with s1567 would be possible if it features QPI even if the PCIe lanes would then go to waste.
Michael H;2180857
Will the multi-processor Xeon varients feature QPI or DMI for the processor-to-processor links?
2x QPI. Servers still have need of PCI-E slots for storage accelerators and GPGPU.
kalniel
2x QPI. Servers still have need of PCI-E slots for storage accelerators and GPGPU.

80 PCIe lanes avaliable in a dual-socket system though, 160 in a quad - which would largely be going to waste - QPI does have the advantage of a simpler layout for motherboard.