Architecture & 3D Comparison

Intel i7 3770K - Ivy Bridge - CPU Review

Add Your Comments

Intel is back today with a new CPU manufacturing process at 22 nm using their new 3D Tri-Gate transistor design. It’s the Sandy Bridge micro-architecture with a new 22 nm twist. Will we see any performance increase or just the standard lower power and better temperatures? Sit back and relax while we cross the Ivy Bridge.

Ivy Bridge Time

Ivy Bridge Time

Bridge Over the River Ivy – Ivy Bridge Architecture

If you are erudite about Intel’s strategy, you know they have a tick-tock strategy that they somehow manage to keep to like, um, clockwork. Ivy Bridge is a tick in this cadence, which is a process shrink to an existing micro-architecture. They also upgraded their iGPU and the platform comes with a new 7-series chipset. Our audience doesn’t exactly use the iGPU for graphics, but the upgrade should come with a decent boost for software that can use it in conjunction with Lucid’s VIRTU software.

Ivy Bridge - TICK!

Ivy Bridge - TICK!

3rd Generation Core Platform

3rd Generation Core Platform

Speaking of the iGPU, they boosted its performance partially with shared cache between the CPU and iGPU. The Ivy Bridge CPU comes in a pretty small 160 mm² package with a hefty 1.4 billion transistors. By comparison, Sandy Bridge had a die size of 216 mm² and 995 million transistors.

22nm Process

22nm Process

Before moving on though, we have the requisite eye candy from some images courtesy Intel.

Ivy Bridge Wafer

Ivy Bridge Wafer

Ivy Bridge Core

Transparent IHS

IHS Off CPU

IHS Off CPU

Ivy Bridge Die - Center

Ivy Bridge Die - Center

Ivy Bridge Die - Right

Ivy Bridge Die - Right

As mentioned, this is Intel’s (and anyone else’s as far as I know) first use of a 3-D transistor. For more information on 3D Tri-Gate transistors, check out this page at Intel. The implications are interesting for overclockers in daily use, which we’ll be sure to check out in detail below. Let’s say the change to this transistor type flips the lid on the whole temperature-and-process-shrink paradigm.

On the right we have more details about the iGPU. We’ll probably explore the iGPU performance later, but this review is all about processing power.

New & Improved

New & Improved

3rd Gen Intel Graphics

3rd Gen Intel Graphics

The 7-series Panther Point chipset is Intel’s newest and it adds some things the 6-series lacked, while staying lackluster in others. The good part is that this is Intel’s first native USB 3.0 chipset. Even the X79 chipset lacked USB 3.0 for some reason or another, necessitating third-party controllers. There are only “up to four” USB 3.0 lanes available, but it’s there finally.

The 7-series chipsets retain Intel’s Smart Response Technology that debuted with the Z68 chipset, which is a boon to those who can’t afford a large SSD. It also gives new support for Intel’s Thunderbolt interface. Frankly, the chances of it usurping USB’s hold on the market is slim to zero, but hey; if anyone wants an alternative to the now-dated FireWire, here you go.

For everything but memory capability (and only at the very extreme end of that spectrum), the 6-series chipset is completely compatible with Ivy Bridge CPUs. You’ll lose some of the features (most importantly, PCIe 3.0 in some cases), but it will work without issue assuming your motherboard manufacturer releases a proper BIOS update. I can confirm that the ASUS Maximus IV Extreme-Z is perfectly ready for Ivy Bridge.

Panther Point Chipset Overview

Panther Point Chipset Overview

Cross-Platform Compatibility

Cross-Platform Compatibility

6-Series & 7-Series Platform Comparison

6-Series & 7-Series Platform Comparison

Here we have the more detailed block diagrams describing how everything is doled out with the three new Platform Controller Hubs (PCH) and the 3rd generation Core CPUs. The PCH on the left (Z77) is where most overclockers will hang their hat, with some budget-minded, more frugal folks going the Z75 route. The latter loses Thunderbolt capability and removes Rapid Storage Technology (RST).

The graphics interface is now PCIe 3.0, with 16x lanes available total. This is more than capable for pretty much anybody that wants to run two GPUs – eight PCIe 3.0 lanes is equivalent to sixteen PCIe 2.0 lanes.

Still – still – Intel only has two SATA 6Gb/s ports available. AMD has had six SATA 6Gb/s ports for two chipset generations now. I can’t fathom why Intel continues to buck the trend and leave us with only two current-generation SATA ports. With full backwards-compatibility, I think it’s time to let SATA II go now, don’t you?

One very important thing to note is the memory speed capability – “Up to 1600 MHz”. Now, I’m not sure if that means DDR3-1600 or DDR3-3200, but what I can tell you is – yowza – do these things have insane memory controllers. I’m not just saying that either. We have a G.Skill DDR3-2666 kit in testing and, whew, are those things fast.

Z77 Chipset Block Diagram

Z77 Chipset Block Diagram

H77 & Z76 Chipset Block Diagrams

H77 & Z76 Chipset Block Diagrams

The first thing many people will do when they get their new Ivy Bridge CPU is (what else?) overclock it. They increased the multiplier lottery this time, with chips capable of up to 63x multipliers. With the new memory capability comes the potential (if partners include it in their BIOS) for more granularity in memory divider options, which is always good to push that last few MHz out of your memory.

Aside from that things haven’t changed much. Ivy Bridge is all multiplier like its predecessor, with rather limited BCLK headroom.

Overclocking Features Comparison

Overclocking Features Comparison

In addition to Smart Response Technology, Intel has added Rapid Start Technology. For those of you using any sort of SSD, this is a nice additional system-suspend mode to allow your PC to wake up faster.

Responsiveness Technologies

Responsiveness Technologies

Rapid Start Technology

Rapid Start Technology

Here’s some more info on the iGPU changes. The most obvious change is DirectX 11 capability – AMD had them on that point for quite a while. It’s nice to see Intel moving forward a bit, but while AMD isn’t exactly competing on the CPU side, Intel still has a lot of catching up to do on the iGPU side.

Graphics Feature Comparison

Graphics Feature Comparison

Architecture & 3D Comparison

Architecture & 3D Comparison

Ahh, here is the moment you’ve all been waiting for – Pricing! Remember. Intel quotes prices based on kilo-unit, so retail could be a little bit more than this.

SKU Comparison & Pricing

SKU Comparison & Pricing

Low Power SKU Comparison & Pricing

Low Power SKU Comparison & Pricing

$313 for an i7 3770K? Yes please. There was a lot of talk on the forums about the potential of a ~$360 price point. It’s nice to see this coming in below the rumors. One important thing to note – the Ivy Bridge processors will not be available immediately. They will go on sale Sunday, April 29th, so you’ve got another week to save.

Speaking of rumors, it has been floated around recently that Intel had raised TDP on these CPUs to 95W. Again, good news – it’s still at 77 W. Not that 95 W is a huge increase, but rest easy, the information reviewers were given most recently still retains the 77 W number.

As far as the retail boxes you can see photographed at the link above, Intel says that “3rd generation Intel quad core standard power processors have a TDP of 77 W. In some cases you may continue to see references to a 95 W TDP. Intel has requested that original equipment manufacturers continue to design platforms based on Intel 7 series express chipsets to a 95 W TDP target to ensure compatibility with 2nd generation Intel processors.”

Meet the Intel i7 3770K

The box is pretty much unchanged save for text editing. No, they didn’t send a retail chip but they do send boxes so we can show you what it will look like.

3rd Gen Core Box

3rd Gen Core Box

Here you have the CPU itself. Not too much to call home about as far as looks; it’s what inside the CPU that counts. Ivy Bridge is still an LGA 1155 package and much smaller than the SNB-E monster we saw in November.

Intel i7 3770K

Intel i7 3770K

Intel i7 3770K

Intel i7 3770K

Intel i7 3770K

Intel i7 3770K

Intel i7 3770K

Intel i7 3770K

Intel i7 3770K

Intel i7 3770K

Of course, Intel was kind enough to send something with which to test their new generation CPU; the DZ77GA-70K.

Intel’s DZ77GA-70K Motherboard

Intel motherboards get the job done. They aren’t often going to compete with more expensive options from other partners, but they are rock solid stable and will run at stock and moderate overclocks quite well. Just don’t go freezing CPUs with one, you’ll fall asleep while it hard-reboots (for the millionth time) every time you make any changes.

On the plus side, it does have a UEFI now and you can actually successfully turn off C-states, so it’s a definite step in the right direction.

Packaging & First Look

Foregoing the windowed look of the DX79SI, Intel’s Z77 motherboard comes in your more standard still-skeleton-themed box.

Intel DZ77GA-70K Box

Intel DZ77GA-70K Box

Intel DZ77GA-70K Box Rear

Intel DZ77GA-70K Box Rear

Box Opened

Box Opened

Intel DZ77GA-70K

Intel DZ77GA-70K

You know me, always one to take lots of photos!

Intel DZ77GA-70K

Intel DZ77GA-70K

While I wish they’d ditch the skull (it’s about as appealing as some of the ridiculous bullet-themed boards MSI and Gigabyte are obsessed with), overall it’s a pretty good looking board.

Accessories

The accessory stack isn’t massive but has all the essentials. It comes with a mouse pad, backplate for the rear I/O and an SLI bracket. New additions are a 3.5″ USB 3.0 adapter and a Bluetooth -slash- WiFi adapter.

Accessories

Accessories

USB 3.0 Adapter

USB 3.0 Adapter

Board Close-Up

First up, you can see the new and the old intermingled. There are two PCIe 3.0 16x slots (which run at dual 8x with two GPUs installed), two PCIe 1.0 slots, one PCIe 4.0 slot and two legacy PCI slots. Those are quickly going the way of the dodo, but legacy connectivity usually never hurts and certainly doesn’t in this case.

Expansion Slots

Expansion Slots

In the upper right corner, Intel has conveniently included power and reset switches. The motherboard speaker is also up here as well as two fan headers. Out of the quad-channel X79 cloud, we’re firmly back on the ground with four memory slots for dual-channel operation. Of course, with the IMC capability of these CPUs, many will probably be asking ‘what quad-channel?’ before long.

Control Corner / DIMMs

Control Corner / DIMMs

In the bottom right, you can see the POST code indicator and some front panel connectivity. Just above the gray SATA ports is the Marvel 6Gb/s SATA controller to add two more SATA 6Gb/s ports to the board.

The bottom left houses the front panel audio connector, another fan header and the LED status indicators. These actually function quite well, akin to ASUS’ Q-LED feature. If you try to overclock something too far, this will tell you quickly where you went wrong.

Bottom Right Corner

Bottom Right Corner

Bottom Left Corner

Bottom Left Corner

Moving around the board, we have another Marvel SATA 6Gb/s controller for the eSATA port. Going clockwise, there is the Intel LAN PHY and second LAN controller as well as the Realtek audio codec.

Marvell Controller

Marvell Controller

Intel Controller

Intel Controller

Intel Controller

Intel Controller

Realtek Audio Codec

Realtek Audio Codec

As alluded to previously, there are eight SATA ports on this board. Two are native SATA 6Gb/s, two are Marvel-controlled SATA 6GB/s, along with four SATA II ports.

SATA Ports

SATA Ports

The rear I/O has plenty of connectivity. There is the welcome return of a PS/2 port (hey, we can’t let go of everything can we?…you can still ditch SATA II Intel). Additionally you have:

  • Four USB 3.0 ports
  • Two “Hi-Current” fast charging USB 2.0 ports
  • Two regular USB 2.0 ports
  • A FireWire port
  • An HDMI connector
  • Analog and digital audio connections
  • The last item down there is the ‘Back to BIOS’ switch in lieu of an actual clear CMOS switch.
Rear I/O

Rear I/O

Under the Hood

Removing the heatsinks, you can see there is good contact throughout. There are thermal pads on the MOSFET heatsinks, which is industry standard. I’m not sure what that gray TIM-like substance on the PCH heatsink is but it’s easier to get off than the ASUS pink goo, for what that’s worth.

Board Heatsinks

Board Heatsinks

Heatsink Contact

Heatsink Contact

There isn’t much you couldn’t see without the heatsinks removed on the bare board, save the power section and the PCH.

Board Rear

Board Rear

Heatsinks Removed

Heatsinks Removed

Speaking of the PCH, here is our new player in the chipset market – the Z77 ‘Panther Point’ Platform Controller Hub.

Z77 Platform Controller Hub

Z77 Platform Controller Hub

Intel has a decent power section. The components aren’t as upgraded and overkill as the upper-tier motherboards on the market, but it serves its purpose well for your average Joe. I wouldn’t freeze a CPU with it, but for 24/7 use it should be sufficient. The voltage controller on this board is brought to us by CHiL.

CPU Socket and Power Section

CPU Socket and Power Section

Power Section

Power Section

CHiL Power Controller

CHiL Power Controller

Overall, it’s not too bad. The UEFI is leaps and bounds better than anything they have had in the past. It’s intuitive and there is plenty of control for most people. The board is stubborn when you really try to push your CPU, so don’t expect to set any records on ambient, much less frozen. The software requires a restart to make changes. POST takes forever and a day. Aside from those issues, it’s an okay board.

Power Consumption

Stock power consumption is always good to check out for those concerned with sipping wattage. Here, the 3770K is far from a disappointment! It sips power in the smallest bit at an astounding 70 W idle. When put under load at stock, power consumption is an amazingly small 134 W – even if temperatures are in the low-to-mid 50 °C range.

Test Setup Idle (Watts) CPU Loaded (Watts)
i7 3770K 70 W 134W
i7 2600K 97 W 158 W
FX-8150 121 W 246 W
i7-3960X 104 W 244 W

As you can see, higher temperatures don’t mean this CPU is taking a lot of power to run. It’s the least of any CPU here. I don’t have numbers on the 2500K, but they are probably similar – and the 3770K trounces it in performance. It’s definitely a power consumption win for Intel.

Of course, when you overclock, all bets are off. Everyone knows wattage rises significantly when you raise voltage and frequency and Ivy Bridge is no different. Loaded consumption at 4.8 GHz is 271 W. Sadly for AMD, that kicks the snot out of anything the FX-8150 is capable of and with only a few watts more than its stock power draw.

To our brothers on the Folding Team, please accept my apologies. Time didn’t allow a Folding@Home run before the review. I’ll be sure to post one up in the comments as soon as I’m able to do so.

Overclocking for Stability

This is where the rubber meets the road. It’s also where a lot of people have been focusing recently after OBR came out with a chart showing how hot these CPUs get when you put a little voltage to them. I’m sorry to say, what you may have seen wasn’t without merit.

Like all our CPU reviews, this chip was tested with a dedicated triple-radiator water loop (Swiftech MCP35x pump, Swiftech MCR-320 radiator and EK Supreme HF Cu water block). It kept up with this CPU, but only just. This i7 3770K ended up dialing in at the same overclock achieved on the 2600K before it –  4.8 GHz. Rather than ~1.33 V though, this was LinX stable at a mere 1.23 V loaded.

4800 MHz Stable

4800 MHz Stable

What’s interesting is that the overclock needed a little boost in voltage to keep up with all of the benches. It’s the first CPU LinX has failed to torture past any reasonable semblance of stability. Actual stability was at ~1.26 V loaded. Temps were right where you see them, which isn’t cold by a long shot.

To put this in perspective, with this same water loop, the 2600K at 1.330 V maxed out at 56 °C. As you can see, with less voltage and the same overclock, the 3770K tops out at a staggering 82 °C! This is what I was implying earlier when talking about the new transistor type. A smaller process in this case does not equal lower temperatures. The hottest core measurements are 46% higher with Ivy Bridge.

Just to see if it could be done, I went for 4.9 GHz to see if the chip would do it (my 2600K couldn’t stabilize at reasonable voltages there). It did for sure, but temperatures were getting too high even on the water loop.

4900 MHz LinX Stable

4900 MHz LinX Stable

90 °C! It’s just insane how hot these things get, it really is. They take the abuse without complaint (long term of course has not been researched yet), but it’s a far cry from cool-running Sandy Bridge. Intel has verified there is a 105 °C Tjunction Max, so the temperatures reported by CoreTemp are, in fact, correct.

When approached about the temperature ‘issue’, they readily admit to hotter temperatures, but are quick to point out that there is more headroom to peak temperature. The reason for these hotter temperatures is that 22 nm has a higher thermal density than 32 nm, so it will -and is expected to- run hotter. As long as the chip stays below the Tjunction Max temperature, it should be just fine. So if it doesn’t matter that you run your CPU at 80 °C+, do you care? Should you?

Regardless, these put out plenty of heat and you’re going to have to dissipate it, so plan on a solid cooling solution if you want to go with Ivy Bridge. One boon to this process shrink is that Intel is now back in a big way with extreme overclockers. There is no cold bug.** Let that sink in for a moment. Intel. Has a chip. With no cold bug. There went AMD’s fun factor!

**There do exist samples that had cold ‘issues’. One report showed issues with high-divider use on the IMC at -120 °C. Another mentions a true cold bug at -170 °C. So there are chips with cold issues, but many of the pre-release showings display zero cold bug.

Test System, Opponents and Methodology

We have a plethora of processors pairing off for you today. So many in fact I had to add a third row to the charts!

CPU Intel i5 2500K Intel i7 2600K Phenom II x6 1100T
Stock / Turbo 3.3 / 3.7 3.4 / 3.8 3.3
Motherboard Gigabyte G1 Sniper2 ASUS P8P67 WS Revolution ASUS Crosshair V Formula
RAM DDR3-1600 9-11-9-24 G.Skill RipjawsX DDR3-2133 9-11-9-24 G.Skill Flare DDR3-2000 7-9-7-24
GPU n/a AMD HD6970 AMD HD6970

 

CPU AMD FX-8150 Intel i7 980X Intel i7-3960X
Stock / Turbo 3.6 / 4.2 3.3 3.3 / 3.9
Motherboard ASUS Crosshair V Formula ASUS Rampage II Extreme Intel DX79SI
RAM G.Skill Flare DDR3-2000 7-9-7-24 Super Talent DDR3-1866 8-10-8-24 G.Skill RipjawsZ DDR3-1600 9-11-9-28
GPU AMD HD6970 n/a AMD HD6970

 

CPU Intel i7 3770K
Stock / Turbo 3.5 / 3.9
Motherboard Intel DZ77GA-70K
RAM G.Skill RipjawsX DDR3-2133 9-11-9-28
GPU AMD HD6970

The BIOS was the latest available from Intel at the time benches were run. There is a newer version now but it has no performance changes, just tweaking to non-essential (for our purposes) items. Performance here is what you should expect in your own system.

All non-graphics benchmarks at stock are run three times with the results averaged. 3D benchmarks, game tests and overclocked benchmarks are run once. The operating system is fully updated Windows 7 x64 to give the most accurate representation of real-world performance.

The results you see below are graphed relative to the Intel i7 3770K’s stock performance. This means that results by the i7 3770K at stock all equal 100% and the other results are graphed as a percentage relative to the its performance. So, for instance, if the i7 3770K scored 200 points on a benchmark and the FX-8150 scored 160 points, on the graph the  i7 3770K would be 100% and the FX-8150 would be 80%, meaning in that benchmark, the FX-8150 is 80% as fast as the i7 3770K. For ease of reading, the stock i7 3770K results are all on the left (editor splat’s idea, and a good one I might add).

Not much more to say about that, other than enjoy the fireworks.

Benchmark Results

First up in our suite is the AIDA64 set of synthetic benchmarks. These are synthetic tests that are great at showing differences between processors. The differences are more pronounced than you may see in more ‘real-world’ benchmarks, but they show the processors’ strengths and weaknesses well. The most pronounced example was Bulldozer’s good integer performance and lackluster floating point performance.

AIDA64 Benchmarks

Starting off with integer calculation tests, we have the five AIDA64 CPU Tests. Remember, all of these graphs are hard to read in the review window; click them to see the non-blurry versions.

AIDA 64 CPU Benchmarks

AIDA 64 CPU Benchmarks

AIDA 64 CPU Benchmarks - 2

AIDA 64 CPU Benchmarks - 2

Right out of the gates, the 3770K does very well for itself, coming in about 8% higher than the 2600K for the most part. In three of the five tests, where the 2600K didn’t quite catch the older hex-core i7 980X, the 3770K comes out ahead. Now, floating point performance.

AIDA 64 FPU Benchmarks

AIDA 64 FPU Benchmarks

AIDA 64 FPU Benchmarks - 2

AIDA 64 FPU Benchmarks - 2

Again consistently higher than its older brother. This is in part due to the 100 MHz advantage the 3770K has over the 2600K, but not all of it because that’s a 2.9% clockspeed boost, where we’re seeing five, six and seven percent (plus) increases across the board.

The Intel board didn’t have any problem running DDR3-2133 this time around (the DP67BG would only run DDR3-1600). It didn’t run high speed memory above that point, but 2133 was okay. That accounts for much of the increase here. Note in the clock-for-clock comparisons below, the memory is run on a better motherboard at DDR3-2133.

AIDA 64 Memory Benchmarks

AIDA 64 Memory Benchmarks

AIDA 64 Memory Benchmarks - 2

AIDA 64 Memory Benchmarks - 2

Where we have expected bandwidth increases, there is a huge (18%) latency drop compared to the 2600K. It bodes well for this memory controller, which is stellar by any standard. We’ll get to that later though.

3D Benchmarking

For 3D benchmarks, we start with the most CPU-bound (3DMark 06) and move toward the more GPU-bound benches down the list. This is also where we start the clock-for-clock comparison between the 2600K and 3770K, both overclocked to 4.8 GHz.

3DMark06

3DMark06

3DMark06 is an obvious win for the 3770K, with graphics and overall scores coming in ahead of even the 3960X. Clock for clock we’re looking at roughly a 2% increase over the 2600K.

3DMark Vantage

3DMark Vantage

Vantage also shows the 3770K puling ahead, almost beating the 3960X when overclocked. It pulls away from the overclocked 2600K a bit here too.

3DMark 11

3DMark 11

Yet again in 3DMark 11 the 3770K continues to shine. Where the 2600K lags behind the stock 3960X overclocked, the 3770K at 4.8 GHz even outshines that one. Mark 3D testing a definite win for the 3770K

Gaming

Since CPU testing goes back a ways, I still use some older game benchmarks for testing. It’s a mixed bag, but overall I think it’s good. You get to see marked differences in CPUs you wouldn’t see with, say, Battlefield 3. We’re going to be updating our GPU test suite after Ivy’s launch but for CPUs I like this batch. Anyway, first up is Stalker: Call of Pripyat.

Stalker: Call of Pripyat

Stalker: Call of Pripyat

Things are heated in battle here. The 2600K seems to come out slightly ahead when clocked the same. At stock, the “W” goes to the higher-clocked 3770K.

Aliens vs. Predator, HAWX2 & Dirt 2

Aliens vs. Predator, HAWX2 & Dirt 2

Regrettably, AvP wasn’t tested with the 2600K overclocked. At lower detail and stock it actually comes in ahead of the stock 3770K. Higher detail is reversed. HAWX 2 tips toward the 3770K and Dirt 2 seems to be inconclusive at best.

Rendering, Video & Compression

Now we get to some real-world comparisons. If you want to know how a CPU will actually perform, this is where you do it. Rendering, video conversion and compression tests are as close as we can get to actual, everyday use. Starting with the former, we have the pair of Cinebench tests.

Cinebench R10

Cinebench R10

3770K for sure here, almost beating the hex-core 3960X when overclocked.

Cinebench R11.5

Cinebench R11.5

The hex-core SNB-E walks away with R11.5, as does the overclocked 3770K from the 2600K

PoV Ray 3.7 RC3

PoV Ray 3.7 RC3

PoV ray also goes decisively to the 3770K, with a clock-for-clock win of over 6%.

x264 Benchmark

x264 Benchmark

x264 benchmark has two phases, the first is a ‘reading’ pass, where the file to be converted is read; then it actually does the video conversion. Why that is important is that when actually converting video, the 3770K separates itself even more from the 2600K than when just reading the file.

7zip

7zip

Compression is another solid win for the 3770K, stock and overclocked.

In the real world, it looks like the 3770K is doing quite well. In these five benchmarks you have an average increase clock-for-clock of 5.5%. If you take the curve-skewing Cinebench R10 out of the mix, that increase is 6.3%. Not too shabby for just a process shrink.

QuickSync Testing

The 2600K and 3770K both have the ability to use their iGPU for specific computing functions, mostly having to do with video manipulation. Using ArcSoft’s Media Converter 7, we’ll take a look at one of the potential gains you can get with the new 4000-series iGPU. To get these times, I used a 1.12 GB AVCHD, 1080p video file and converted it for use on a Motorola Droid, a phone I happen to have. Basically, converting a video from very large to very small.

It doesn’t take very long to perform this task with the CPUs (both clocked at 3.9 GHz), but when you get that iGPU involved, it really flies.

Media Converter 7

Media Converter 7

When using just the CPUs, you get an expected drop in time of about 7%. When you add the iGPUs into the mix is where Ivy Bridge really takes the 2600K and thrashes it around, with an astounding video conversion time decrease of 58%. If you do any video manipulation, this could be a huge boon for your productivity. There are quite a few applications that can take advantage of this technology, so look through the list and see if it’s something you would use.

QuickSync Applications

QuickSync Applications

2D Benchmarking

This area is what benchmarkers have been looking forward to. Getting right to it, we’ll look at single-core performance courtesy SuperPi 1M and 32M.

SuperPi 1M

SuperPi 1M

SuperPi 32M

SuperPi 32M

Ivy Bridge definitely shows greater efficiency, albeit not by leaps and bounds. There is some time knocked off in both of these, but not anything crazy. Now for full multi-threaded testing via WPrime 32M and 1024M.

WPrime 32M

WPrime 32M

WPrime 1024M

WPrime 1024M

Ivy steps away a bit more here, showing stronger improvements when all cores are engaged. It’s still not a huge amount, but it’s enough to make benchmarkers raise their eyebrows. When considering the benchmarking capabilities of these chips (when frozen), these start to look pretty good.

Clock-for-Clock Comparison

Now let’s check out the overall clock-for-clock picture. These graphs have the percentages reversed, with the 2600K always 100% and the 3770K showing how much better it is than the 2600K. Remember, we’re just looking at a process shrink and a different type of transistor. The micro-architecture is the same between the two chips. Slight increases are expected, but nothing grandiose. AIDA 64 isn’t included in this graph because it wasn’t run clock-for-clock.

Overall Clock-for-Clock Analysis: Scored Benchmarks

Overall Clock-for-Clock Analysis: Scored Benchmarks

The good thing is that Ivy Bridge comes out ahead in every benchmark, as well it should. When you average these increases, you come up with an overall 3.8% improvement over its predecessor, which isn’t bad at all really.

Clock-for-Clock Comparison - Timed Benchmarks

Overall Clock-for-Clock Analysis - Timed Benchmarks

Timed benchmarks have a narrower gap. Multi-threaded WPrime is right there where it should be, exactly in line with the 3.8% improvement seen in the scored benchmarks. Single-threaded clock-for-clock wanes a bit, with only about a 2% increase for both benchmarks. Combined, timed benchmarks gain an average of 2.8% over the 2600K.

Pushing the Envelope

Now we get to the fun part. Or, the supposed-to-be-fun part. You saw above that Ivy Bridge isn’t exactly a cool running architecture in its current state. Unfortunately, that limits ambient overclocking quite a bit. While the CPU very well may be capable of plenty higher overclocks, when using ambient cooling – even on water – it will hit a heat wall pretty quickly.

As mentioned before, the Intel motherboard isn’t going to set any records. It wouldn’t even boot into the OS at 5 GHz. As you’ll see the chip is plenty capable of that, but the board was not. To facilitate ‘pushing the envelope’, and since the Z77-based Maximus V Gene hadn’t arrived yet, I popped the 3770K into the Z68-based Maximus IV Extreme-Z to see what we could come up with.

The really fun part about Ivy Bridge is its memory capability. You see, G.Skill has provided us with a very stellar memory kit to review. Rated for DDR3-2666, they ran just fine at their rated speed and even above that. The Z68 board, however, isn’t cut out for this kind of memory frequency. While it will run after a tweak here and there…and if you hold your foot at the right angle while praying to the memory gods, you’ll need a proper Z77 board to really push this kind of memory (and this kind of IMC) to its limit.

Anyway, as you can gather, heat is a problem. The highest WPrime 32M and 1024M would run and not scare me off with temperatures was 5.0 GHz even. Even there – at rather reasonable voltages I might add – the temperature wall loomed ever present. AISuite’s sensor warnings went crazy with both of these, but it was worse with WPrime 1024, hitting a maximum of 94 °C. Times were great for just 5.0 GHz, but don’t plan on going much farther unless you can freeze your chip.

WPrime 32M @ 5000 MHz

WPrime 32M @ 5000 MHz / DDR3-2666

WPrime 1024M @ 5000 MHz

WPrime 1024M @ 5000 MHz / DDR3-2666

SuperPi went a bit farther. Here we can see just how stellar this memory is, running SuperPi 32M with room-temperature RAM at their crazy rated speed of DDR3-2666. Temperatures were also a problem here, even on a single-threaded benchmark, AISuite’s warnings going off showing temperatures of 82 °C. The time is great though and the IMC is even more so.

SuperPi 32M @ 5100 MHz and DDR3-2666

SuperPi 32M @ 5100 MHz / DDR3-2666

With the quick, painless SuperPi 1M bench we were able to go a little farther to show off both the CPU and the IMC. The CPU reached 5175 MHz and the memory reached DDR3-2760!

SuperPi 1M @ 5175 MHz and DDR3-2760

SuperPi 1M @ 5175 MHz / DDR3-2760

No doubt about it, benchmarkers need extreme cooling to bench these chips successfully. There are no ifs, ands, or buts about it – if you plan on benchmarking with Ivy Bridge, you need to freeze these chips. Thankfully that can be done quite successfully. Unlike other Intel chips of late, these have no cold bug (**See caveat above in Overclocking for Stability). Leaks abounding across the internet show successful benchmarking anywhere from ~6.3 GHz for a mediocre chip to ~7 GHz for the best chips.

Ambient overclockers with 5.4 GHz plus Sandy Bridge CPUs may want to hold on to them. Ivy just won’t get there from here, at least in its current iteration. Go cold or go home and be happy with your current Sandy Bridge chip. If you plan on using extreme cooling though, these chips have got ‘it’ where Sandy Bridge never did, with what appears to be consistent 6 GHz + performance.

Final Thoughts & Conclusion

Ivy Bridge is a mixed bag for overclockers. It is a more efficient process than Sandy Bridge, gaining anywhere from two to seven percent across the benchmark tests. If you manipulate videos with programs that can use QuickSync there is no question, Ivy will save you time and, by extension if you do it for a living, money.

Ambient overclocking for 24/7 use is about the same as Sandy Bridge for all intents and purposes. Depending on what you do with your system, maybe 2-7% is good enough to entice you to upgrade. Maybe you want native USB 3.0 without extra controllers. Maybe you want and/or need PCIe 3.0 for some reason. All of these are not great, but decent reasons to upgrade from Sandy Bridge. That’s a decision only you can make.

If you are happy with your Sandy Bridge chip or are a benchmarker that doesn’t use extreme cooling methods, you probably want to stick with what you’ve got and see how things play out with further stepping revisions, if indeed anything can change.

If you happen to use Sandy Bridge to enhance your ego and absolutely must run 5 GHz for 24/7 operation, keep your chip – these won’t be running at that frequency for 24/7 use unless you use a phase change cooling solution. My Sandy Bridge chip never got there for 24/7 use, topping out at 4.8 GHz. That speed on Sandy Bridge and Ivy Bridge is plenty for me as it is for the vast majority of overclockers; most would even call that overkill.

If you are a ‘normal’ overclocker (i.e. don’t bench much like us crazy people on the benching team) and waited out Sandy Bridge to see what Ivy had to bring, it’s as good a time as any to upgrade. The extra features (PCIe 3.0, native USB 3.0, Quick Sync, SRT, solid efficiency, decent 24/7 overclocks) are certainly worth the reasonable ~$313 investment (plus motherboard). Haswell will be a long time coming, so it’s well worth it to upgrade from pretty much anything Westmere architecture and before.

Ivy Bridge is a tick. It gives us back extreme overclocking ability on Intel CPUs. It doesn’t shatter the 2nd Generation Core chip’s clock-for-clock performance, because it doesn’t change the microarchitecture. No one expected that and no one should be surprised. Temperatures were a bit of a shocker, but for 24/7 use, while a little warm, they shouldn’t hurt anything.

Ivy Bridge is Overclockers Approved because it does what we want it to do. You can overclock them to plenty high speeds for 24/7 use – pretty much right where you could overclock Sandy Bridge. You can overclock the begeezus out of them under extreme cooling, an added bonus Sandy Bridge never had. You can do it all with more efficiency than the previous generation. Lastly, you can do it with the added bonus of a ridiculously strong memory controller. It’s not perfect, but it ticks in the right direction.

Jeremy Vaughan (hokiealumnus)

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>

Discussion
  1. bmwbaxter
    Bigger and meaner!!! :attn: Hopefully they will release and 8c/16t for desktop :D


    I could see a couple of these on an EVGA SR-X workstation board running flat out on my desktop :D

    Janus67
    Maybe they'll solder that IHS to the core.

    :chair:


    Yeah they'd better :burn:
    Thanks. :thup:

    My question was mainly about the transistors any components. Not the interconnect pathways. Those pictures makes me appreciate the people who deisgn and implement this stuff.
    hokiealumnus
    Not to my knowledge, but it's not something explicitly stated one way or the other.

    From my own limited knowledge, it would seem very hard to stack transistors on top of one another, requiring additional silicon substrate on top of the first transistor, which would make cooling the first layer excessively difficult - the second layer would act as an insulator to the first. Yea, I'll go with 'most likely not' as an answer.


    I'm also very limited in my knowledge of the actual engineering as well, so I don't know if the transistors themselves are packaged on one plane/layer. However I do know there are many layers (dozens, perhaps many more) within the chip that interconnects all the circuitry. There are interesting pictures and explanations if you google for things like electron microscope and CPUs.

    This is one example I've seen linked around here before:

    http://www.sciencystuff.com/
    I surely hope so, since LGA2011 is supposed to be enthusiasts/workstation grade equipment and not consumer grade stuff. If they don't, there will be no compelling reason to go IB-E instead of SB-E for enthusiasts running ambient cooling.
    Audioaficionado
    I think I'll just skip the IB desktop and wait for the 3rd gen mobile platforms. We'll see what IB-E looks like when it comes out.


    Maybe they'll solder that IHS to the core.

    :chair:
    Audioaficionado
    We'll see what IB-E looks like when it comes out.


    Bigger and meaner!!! :attn: Hopefully they will release and 8c/16t for desktop :D
    I should be impressed, but I am hounestly not impressed. I do like the new tech they have... but it's still kinda... bad. It may be at the top, but the cheaper alternatives, or the out dated models aren't to far behind from the 3770k. So... I really don't believe an upgrade should be necessary to anyone.
    hokiealumnus
    Not to my knowledge, but it's not something explicitly stated one way or the other.

    From my own limited knowledge, it would seem very hard to stack transistors on top of one another, requiring additional silicon substrate on top of the first transistor, which would make cooling the first layer excessively difficult - the second layer would act as an insulator to the first. Yea, I'll go with 'most likely not' as an answer.


    Thanks, I only asked because I was having a discussion with a friend about it and didn't know but I stated the same hypothesis as you. I figured if anyone was privy to that info it would be you since you reviewed it.

    Depending on how far down the rabbit hole I wannna go I might PM Dolk :p
    Not to my knowledge, but it's not something explicitly stated one way or the other.

    From my own limited knowledge, it would seem very hard to stack transistors on top of one another, requiring additional silicon substrate on top of the first transistor, which would make cooling the first layer excessively difficult - the second layer would act as an insulator to the first. Yea, I'll go with 'most likely not' as an answer.
    hokiealumnus
    I'm not sure what you mean. There aren't two dies stacked on top of one another, no.


    Sorry for my vagueness, I mean are the transistors only 1 layer thick accross the whole die? Or is there places where stuff(any component) is atop another one?
    Interesting, removing the IHS apparently did not help with temps at all.

    http://www.overclock.net/t/1249419/pcevaluation-intel-i7-3770k-temperature-measured-without-ihs
    Frakk
    Concidering i removed it some 25 minutes before don't you think it would have been better to PM me instead of quoting what was no longer there?

    I mean do you often go around replacing what respondents removed from there posts by way of your privileges later on in your own quotes?

    Anyway, i think it came from TPU.

    @ xander89.. lol that made me laugh :)
    Nope I caught it live before your edit. I had several tabs open and didn't get around to posting until after the ninja edits :D
    I.M.O.G.
    Picture caption references the source of the picture, which was pt1t, a member who originally shared a delidding pic on hwbot.org. I contacted him to get another picture showing the actual underside of the IHS - we have not delidded our sample, as we are depending on it for other reviews currently in our pipeline. (and we are being wusses about it)


    Hey!

    I represent that remark... :chair:

    EDIT - Shamino says IHS-less helped OC on air, but didn't specify how much. It wasn't expected to, and indeed didn't, help with extreme overclocks. (Reference.)
    Was hoping to see the results, Just delidded my Athlon 630, four pins were broken before hand so it was just out of interest

    Placing a heatsink in the oven at 225C then placing it on top of the CPU to melt the solder and get the ISH off, because you've creative.
    Picture caption references the source of the picture, which was pt1t, a member who originally shared a delidding pic on hwbot.org. I contacted him to get another picture showing the actual underside of the IHS - we have not delidded our sample, as we are depending on it for other reviews currently in our pipeline. (and we are being wusses about it)