In 2006 Intel introduced its tick-tock cadence for microprocessor releases. Every year would see the release of a new family of microprocessors as either a tick or a tock. Ticks would keep architectures relatively unchanged and focus on transitions to smaller manufacturing technologies, while tocks would keep fab process the same and revamp architecture. Sandy Bridge was the most recent tock, and arguably the biggest one since Intel started down this road.

At a high level the Sandy Bridge CPU architecture looked unchanged from prior iterations. Intel still put forth a 4-issue machine with a similar number of execution resources to prior designs. Looking a bit closer revealed that Intel completely redesigned the out-of-order execution engine in Sandy Bridge, while heavily modifying its front end. Sandy Bridge also introduced Intel's high performance ring bus, allowing access to L3 by all of the cores as well as Intel's new on-die GPU.

The Sandy Bridge GPU was particularly surprising. While it pales in comparison to the performance of the GPU in AMD's Llano, it does represent the first substantial effort by Intel in the GPU space. Alongside the integrated GPU was Intel's first hardware video transcoding engine: Quick Sync. In our initial review we found that Quick Sync was the best way to quickly transcode videos, beating out both AMD and NVIDIA GPU based implementations in our tests. Quick Sync adoption has been limited at best, which is unfortunate given how well the feature performed in our tests.

Sandy Bridge wasn't all rosy however. It was the first architecture that Intel shipped with overclocking disabled on certain parts. Any CPU without Turbo Boost enabled is effectively unoverclockable. Intel killed the low end overclocking market with Sandy Bridge.

The overclocking limits were a shame as Sandy Bridge spanned a wide range of price points. The low end Core i3-2100 was listed at $117 while the highest end Core i7-2600K came in at $317. While you can't claim that Sandy Bridge was overpriced at the high end, there's always room for improvement.

Despite abandoning Pentium as a high end brand with the 2006 release of Intel's Core 2 Duo, Intel has kept the label around for use on its value mainstream parts. Last year we saw only two Pentium branded Clarkdale parts: the G6950 and G6960. This year, powered by Sandy Bridge, the Pentium brand is a bit more active.

Processor Core Clock Cores / Threads L3 Cache Max Turbo Max Overclock Multiplier TDP Price
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
Intel Core i5 2400 3.1GHz 4 / 4 6MB 3.4GHz 38x 95W $184
Intel Core i5 2300 2.8GHz 4 / 4 6MB 3.1GHz 34x 95W $177
Intel Core i3 2120 3.3GHz 2 / 4 3MB N/A N/A 65W $138
Intel Core i3 2100 2.93GHz 2 / 4 3MB N/A N/A 65W $117
Intel Pentium G850 2.9GHz 2 / 2 3MB N/A N/A 65W $86
Intel Pentium G840 2.8GHz 2 / 2 3MB N/A N/A 65W $75
Intel Pentium G620 2.6GHz 2 / 2 3MB N/A N/A 65W $64
Intel Pentium G620T 2.2GHz 2 / 2 3MB N/A N/A 35W $70

The new Sandy Bridge based Pentiums fall into two lines at present: the G800 and G600. All SNB Pentiums have two cores (HT disabled) with 256KB L2 per core and a 3MB L3 cache. CPU core turbo is disabled across the entire Pentium line. From a performance standpoint, other than missing hyper threading and lower clocks - the Sandy Bridge Pentiums are very similar to Intel's Core i3.

Intel continues to separate the low end from the high end by limiting supported instructions. None of the Pentiums support AES-NI or VT-d. Other than higher clock speeds the 800 series only adds official DDR3-1333 support. The 600 series only officially supports up to DDR3-1066.

All standard Pentiums carry a 65W TDP. The Pentium G620T runs at a meager 2.2GHz and manages a 35W TDP. Regardless of thermal rating, the boxed SNB Pentiums come with an ultra low profile cooler:

These Pentium CPUs work in the same 6-series LGA-1155 motherboards as their Core i3/5/7 counterparts. The same rules apply here as well. If you want video out from the on-die GPU you need either an H-series or a Z-series chipset.

The Pentium GPU

When Intel moved its integrated graphics on-package with Clarkdale it dropped the GMA moniker and started calling it HD Graphics. When it introduced the Sandy Bridge Core i3/5/7, Intel added the 2000 and 3000 suffixes to the HD Graphics brand. With the Sandy Bridge Pentium, Intel has gone back to calling its on-die GPU "HD Graphics".

Despite the name, the Pentium's HD Graphics has nothing in common with Clarkdale's GPU. The GPU is still on-die and it features the same architecture as Intel's HD Graphics 2000 (6 EUs). Performance should be pretty similar as it even shares the same clock speeds as the HD 2000 (850MHz base, 1.1GHz turbo for most models). I ran a quick test to confirm that what Intel is selling as HD Graphics is really no different than the HD Graphics 2000 in 3D performance:

Intel HD vs 2000 vs 3000 - Crysis Warhead

All is well in the world.

Where the vanilla HD Graphics loses is in video features: Quick Sync, InTru 3D (Blu-ray 3D), Intel Insider (DRM support for web streaming of high bitrate HD video) and Clear Video HD (GPU accelerated post processing) are all gone. Thankfully you do still get hardware H.264 video acceleration and fully audio bitstreaming support (including TrueHD/DTS-HD MA).

Missing Quick Sync is a major blow, although as I mentioned earlier I'm very disappointed in the poor support for the feature outside of the initial launch applications. The rest of the features vary in importance. To someone building a basic HTPC, a Sandy Bridge Pentium will do just fine. Personally I never play anything in 3D, never use the Clear Video HD features and never use Intel Insider so I wouldn't notice the difference between a Sandy Bridge Pentium and a Core i5 for video playback.

The Matchup
Comments Locked


View All Comments

  • iwod - Tuesday, August 23, 2011 - link

    The sad thing is AMD is still using a CPU core then is two generation behind Intel. We will have to wait for Bulldozer + updated GPU for a decent low end CPU.

    However by the time Intel would have either Ivy or Haswell ready.

    AMD you need to work harder.
  • CeriseCogburn - Monday, June 25, 2012 - link

    iwod, you may have waited, and achieved: massive, epic failure, amd style...

    Since amd still has to be sold with endless lies, everyone knows the answer that few are willing to admit.

    I congratule on amd lying about bulldozer transistor count, fanning their fanboys to the outer limits of worshipful surrender.
  • Targon - Tuesday, August 23, 2011 - link

    Now that Firefox and IE9 support GPU acceleration, adding these to the CPU benchmarks SHOULD be seen as fairly important when talking about low to mid range processors/GPUs. I suspect that if running Firefox 6 with Flash(due to banner advertisements and such), you might see some interesting results on a per-core basis between the AMD A6 and Intel Pentium chips reviewed here.
  • fuzzymath10 - Tuesday, August 23, 2011 - link

    I'm pretty sure you need only DX10 for that. My laptop's x3100 IGP from 2008 supports h/w rendering because the driver is DX10. Flash video decoding is more restricted, but it's possible the new Intel IGPs will work since they have DXVA decoding of h.264.
  • Targon - Tuesday, August 23, 2011 - link

    The GPU acceleration obviously will be better with a faster GPU in browsers and Flash, and that helps level the playing field in this case. I would like to see how much it levels the playing field.
  • JarredWalton - Tuesday, August 23, 2011 - link

    If you have a good benchmark for measuring performance in IE9 and FF6, I'd love to hear it. Sure, I can open up a Flash video, but on any modern GPU (including Arrandale's IGP), it's not a problem. Even GMA 4500MHD could handle Flash video content with the latest drivers (helped by the CPU of course).

    The problem is coming up with a meaningful benchmark with browsers. Are we supposed to look at CPU usage while watching a video, or power usage? Do we go to some weird web page benchmark that stresses the GPU accelerated portions of the browser, even though 99% of web sites look like our site and don't benefit? I've got from Firefox 1 through 6, most versions of Chrome, every major Internet Explorer release, and I've even dabbled with Safari (sucks on Windows!) and Opera. Unless you're playing Flash games, I haven't ever felt that any of the browsers was "too slow"; mostly I stuck with FF for the extensions, but I've moved to Chrome now. Both still run fine whether I'm on IGP or dGPU.

    I guess my point is, sure, you can create a browser test that stresses GPUs/CPUs to the point where it's a benchmark, but is it actually useful data? If a web site is putting a 100% load on your CPU, GPU, or both, it's a poorly designed site for normal use.
  • knedle - Tuesday, August 23, 2011 - link

    I have bought few months ago 620T, compared to AMD I was previously using, power consumption is much smaller, same goes for noise, I can't hear it with stock cooling.
    My whole computer with high quality PSU, 3 HHDs (all power saving), a DVD and some basic ASUS motherboard takes only 60W in stress, and for things I do with this computer, it's far more superior than AMD.
    I should mention, that I use my computer as server for backups of my customers servers, so I need high IO rate without any bottlenecks, and AMD always gave me problems with that. For example, AMD had problems while I wanted to watch a movie, while 2 servers were sendbing backups to my computer and saving it to LVM on software RAID array.
  • koan00 - Tuesday, August 23, 2011 - link

    On page 3, the link to "Bench" is broken in the second sentence

    Also, the hardware setup lists a Corsair drive for the Hard Disk, but then 2 paragraphs down references the Intel XM25-M ?
  • Ryan Smith - Tuesday, August 23, 2011 - link

    The Bench link looks fine here. What are you seeing?
  • koan00 - Tuesday, August 23, 2011 - link

    It initially linked to " but it appears to be correct now. Thanks.

Log in

Don't have an account? Sign up now