If you are a normal desktop user or even a power user with plans to run at over 4GHz, the vanilla LGA-1155 Sandy Bridge platform
is good enough. You get some of the fastest CPUs on the market today
paired with reasonably priced motherboards and the ability to use Quick Sync to transcode video...er...quickly. If that's not enough, Intel launched a higher end platform last month: the LGA-2011 Sandy Bridge E.
Take a regular Sandy Bridge, add PCIe 3.0 support,
increase the number of PCIe lanes that branch off of the CPU (from 16
to 40 lanes), double the number of memory channels (4 x 64-bit DDR3
memory controllers) and you've got Sandy Bridge E and its LGA-2011/X79 platform.
SNB-E is currently available in two forms: a $999 6-core Extreme
Edition part (Core i7 3960X) and a $555 6-core unlocked version (Core i7
3930K). Neither is exactly cheap but if you need the PCIe lanes, core
count and memory bandwidth, they are your only ticket.
LGA-2011 SNB-E (left) vs. LGA-1155 SNB (right)
Sandy Bridge E is a fairly niche platform to begin with, but what about
the niche within the niche (extremeception?) of users who just need the
LGA-2011 platform but not necessarily a 6-core behemoth? For those
users, there's the Core i7 3820:
LGA-2011 SNB-E (left) vs. LGA-1155 SNB (right)
The Core i7 3820 is the first (and only) quad-core Sandy Bridge E CPU.
It's so new that the chip isn't even available for sale nor will it be
until early 2012. Unlike the relationship between the 3960X and 3930K,
the 3820 is an entirely new die.
The chip features four Sandy Bridge cores and is paired with a 10MB L3
cache, giving it a slight advantage over the highest end LGA-1155 Sandy
Bridge CPUs on the market today. The result is a die that's bigger than a
regular SNB but significantly smaller than a 6-core SNB-E:
There's a small improvement in base clock frequency over the fastest SNB
(3.6GHz vs. 3.5GHz) but max turbo remains unchanged at 3.9GHz. For
single threaded applications the 3820 should be just as fast as a Core
i7 2700K or a 3960X. The same is true for if all heavily threaded
workloads, at least when compared to other quad-core parts (the
3960/3930 still maintain an advantage because of their two additional
cores).
There's no on-die GPU, no heatsink/fan in the box and this isn't a
fully unlocked part so the 3820 is actually cheaper than most of the
high-end Sandy Bridge CPUs: it's priced at $285. If it weren't for the
fact that you'll still need to spend over $200 on a motherboard I would
say that the 3820 is a steal.
Overclocking
Let's start with the basics. All Sandy Bridge CPUs are clock locked by
default, they can't operate at frequencies other than what they're sold
at. There are two exceptions to this rule. All Sandy Bridge CPUs that
support Turbo Boost are partially unlocked. Not only can they turbo up
to frequencies that are higher than their default clock, but they can
also be overclocked to frequencies even higher than their turbo speeds.
By default, all Turbo enabled Sandy Bridge CPUs can be set to run at up
to four bins (4 * bclk or 400MHz by default) higher than their standard
turbo frequencies:
The second exception is any of the K or X-series SKUs. If your Sandy
Bridge model number ends in a K or X, then the chip is fully unlocked
and can be overclocked up to 5.7GHz using only clock multiplier
adjustments.
The Core i7 3960X and 3930K fall into the latter category, which makes
overclocking them a breeze. The 3820 on the other hand is partially
unlocked, which means that we can run it at up to 4.3GHz using
multiplier adjustments alone. Note that the 4.3GHz limit only applies to
the 1-core active state, in the worst case scenario of all cores active
with no room to turbo the highest operating frequency of the chip when
overclocked would be 4.0GHz.
The max 3820 overclock without touching bclk settings
While these aren't bad targets, they're not all that exciting either.
Thankfully Sandy Bridge E makes it even easier to overclock through the
use of a few higher bclk frequencies. By default Sandy Bridge uses a
100MHz bclk, but SNB-E allows for 125MHz, 166MHz and 250MHz options as
well. The 166/250MHz settings are a bit too aggressive, but the 125MHz
bclk setting proved to be the perfect companion for the 3820.
Intel's DX79SI makes it extremely easy to overclock, especially with
the latest BIOS update. There are built in overclocking profiles for
each bclk setting that you can choose from. I simply selected the 1.25x
(125MHz) profile option and then went through the list of target
frequencies until I found one that seemed promising. I also went in and
tweaked some of the settings myself to get a bit more clock speed
(4.63GHz is the fastest profile this board allows by default) and
improve stability. In the end I was fairly pleased with what the 3820
could do: 4.75GHz with Intel's RTS2011LC closed loop cooling system.
Now we're talking
I couldn't get the 3820 as stable as I would like at 4.88GHz and 5GHz was unfortunately out of the reach of my sample. I can't really complain about 4.75GHz from a $285 chip though, especially without resorting to anything too exotic from a cooling standpoint. Overclocking is also extremely effortless thanks to the new bclk options on SNB-E. Although the Core i7 3820 isn't an unlocked part, that doesn't limit how far (or how easily) it can be overclocked. A big part of the ease of overclocking is due to how good of a job Intel did on the DX79SI BIOS options, but from what we've seen the third party boards also do a decent job of simplifying the process.
The Test
To keep the review length manageable we're presenting a subset of our results here. For all benchmark results and even more comparisons be sure to use our performance comparison tool: Bench.| Motherboard: | ASUS P8Z68-V Pro (Intel Z68) ASUS Crosshair V Formula (AMD 990FX) Intel DX79SI (Intel X79) |
| Hard Disk: | Intel X25-M SSD (80GB) Crucial RealSSD C300 |
| Memory: | 4 x 4GB G.Skill Ripjaws X DDR3-1600 9-9-9-20 |
| Video Card: | ATI Radeon HD 5870 (Windows 7) |
| Video Drivers: | AMD Catalyst 11.10 Beta (Windows 7) |
| Desktop Resolution: | 1920 x 1200 |
| OS: | Windows 7 x64 |
Windows 7 Application Performance
3dsmax 9
Today's desktop processors are more than fast enough to do
professional level 3D rendering at home. To look at performance under
3dsmax we ran the SPECapc 3dsmax 8 benchmark (only the CPU rendering
tests) under 3dsmax 9 SP1. The results reported are the rendering
composite scores.
Cinebench 11.5
Created by the Cinema 4D folks we have Cinebench, a popular 3D
rendering benchmark that gives us both single and multi-threaded 3D
rendering results.
With only a 100MHz clock speed advantage over a 2600K when running in
single core turbo mode, the 3820 isn't much faster than the 2600K in
our single threaded Cinebench test. The additional L3 cache doesn't have
much of an impact here, although I suspect that has more to do with
this particular workload rather than a general statement about the 3820.
Let's look at multithreaded perf:
The performance gap increases to 5% once we ramp up thread count. The
extra performance is mostly due to clock speed here, although you'll
see later on that there are some applications that definitely appreciate
the larger L3 cache.
7-Zip Benchmark
While Cinebench shows us multithreaded floating point performance,
the 7-zip benchmark gives us an indication of multithreaded integer
performance:
The 7-zip benchmark gives us a good example of what the SNB-E
platform can offer given the right workload. Here we see an 8.6%
performance advantage, despite a much smaller clock speed advantage. The
added L3 cache helps out a bit here, although obviously there's a huge
gap between the 3820 and its hexa-core brethren.
PAR2 Benchmark
Par2 is an application used for reconstructing downloaded archives.
It can generate parity data from a given archive and later use it to
recover the archive
Chuchusoft took the source code of par2cmdline 0.4 and parallelized
it using Intel’s Threading Building Blocks 2.1. The result is a version
of par2cmdline that can spawn multiple threads to repair par2 archives.
For this test we took a 708MB archive, corrupted nearly 60MB of it, and
used the multithreaded par2cmdline to recover it. The scores reported
are the repair and recover time in seconds.
In tests that have more of an IO influence the difference between the
3820 and the 2600K is negligible, it will take higher clock speeds and
more cores to really separate SNB-E from the vanilla SNB systems.
TrueCrypt Benchmark
TrueCrypt is a very popular encryption package that offers full
AES-NI support. The application also features a built-in encryption
benchmark that we can use to measure CPU performance:
Encryption speed once again scales with core count and clock speeds, the additional L3 cache doesn't do much in this benchmark.
x264 HD 3.03 Benchmark
Graysky's x264 HD test uses x264 to encode a 4Mbps 720p MPEG-2
source. The focus here is on quality rather than speed, thus the
benchmark uses a 2-pass encode and reports the average frame rate in
each pass.
We see a slight advantage over the 2600K in our x264 HD benchmark,
however video transcoding doesn't benefit all that much from the small
gains the 3820 offers. Most client users would be better off with the
Quick Sync enabled 2600K, and the serious video professionals will want
to invest in a six-core 3930K at the minimum.
Compile Chromium Test
You guys asked for it and finally I have something I feel is a good
software build test. Using Visual Studio 2008 I'm compiling Chromium.
It's a pretty huge project that takes over forty minutes to compile from
the command line on the Core i3 2100. But the results are repeatable
and the compile process will stress all 12 threads at 100% for almost
the entire time on a 980X so it works for me.
Again we see a step function improvement when moving from four to six
cores in our compile test, but no change between the 2600K and 3820. If
you're building a dev workstation you're going to either want to save
money and grab a 2600K or move to six cores for better performance. It
is worth mentioning however that if you need eight DIMM slots the 3820
might be a better option than the 2600K, allowing you to outfit your
workstation with insane amounts of memory.
Excel Monte Carlo
Our Monte Carlo simulation test is CPU bound but the 3820 shows a marginal improvement over the 2600K.
SYSMark 2007 & 2012
Although not the best indication of overall system performance, the
SYSMark suites do give us a good idea of lighter workloads than we're
used to testing. SYSMark 2007 is a better indication of low thread count
performance, although 2012 isn't tremendously better in that regard.
In 2007 we see mild gains over the 2600K, although 2012 shows a much
bigger gap between the 3820 and the 2500K due to the former's support
for 8 threads vs. 4.
Gaming Performance
Gaming performance across the board echoes what we've already seen a lot of - the 3820 shows marginal gains over the 2600K.
Civilization V
Civ V's lateGameView benchmark presents us with two separate scores:
average frame rate for the entire test as well as a no-render score that
only looks at CPU performance.
Crysis: Warhead
Dawn of War II
DiRT 3
We ran two DiRT 3 benchmarks to get an idea for CPU bound and GPU bound performance. First the CPU bound settings:
Metro 2033
Starcraft 2
World of Warcraft
Power Consumption
Power consumption at idle is a bit higher than the LGA-1155 options,
but that's largely negligible since we're talking about two different
platforms here. Power draw under load is slightly higher than the 2600K
and a lot lower than the 3960X for obvious reasons.
Final Words
I see three reasons why you'd want the Core i7 3820:
1. You need PCIe 3.0 today and/or you need more PCIe lanes than a Core i7 2600K can provide,
2. You need tons of memory bandwidth for a particular application,
3. You want a 2600K but you need a platform that can support more memory (32GB+).
If you fall into any of those categories, the 3820 gets the job done.
It's easily as fast as the fastest LGA-1155 Sandy Bridge without adding
significant power consumption or really being limited on the
overclocking side either. The 3820 admittedly targets a niche, but it
does so without any real trade offs. If you land outside of the 3820's
niche however, you're better served by the 2500/2600K at a lower total
platform cost or a 3930K/3960X if you're running a heavily threaded
workload and can use the extra cores.
What About Ivy?
By the time the 3820 is available for purchase early next year, Ivy Bridge
will be just about a quarter away. For desktop users Ivy Bridge is
really only going to bring lower power consumption and a better
integrated GPU. If you're seriously considering anything in the SNB-E
family, the latter isn't going to matter and the former will be of
arguable value. I do expect that we'll see a drop-in upgrade path to
IVB-E at some point in early 2013 if you're concerned about platform
longevity, although Intel hasn't officially committed to such a thing.
It's pretty safe to say that you'll be on your own after IVB-E however,
Haswell should be a fairly large departure from IVB-E in a lot of
senses.
For everyone else, if you need a desktop system today - the LGA-1155
Sandy Bridge is still a viable option. There's always something better
around the corner but I have no issues recommending either that you buy
now or you wait for IVB. If you can wait, you'll be getting a cooler CPU
with better integrated graphics and faster Quick Sync.
source:www.anandtech.com
