Intel Haswell i7-4770K & i5-4670K Review

[SKYMTL]

GPGPU Performance (MuseMage / Photoshop CS6)

GPGPU Performance (Pg.2)

MuseMage

MuseMage is a fully featured image editing program that uses GPU acceleration for its enhanced suite of color effects, filters, adjustment tools and other features. In order to achieve our results, we used MuseMage’s handy Batch script to apply 15 separate and consecutive image modifications to a 50MB JPG file. The results below represent the amount of time it took to complete this task.

RESULTS: MuseMage provides an interesting counterpoint to our previous testing. Since it is optimized for OpenCL, we can see the benefits of AMD's more powerful graphics cores but also how Intel has managed to substantially close the gap through the use of a more compute-focused architecture.

Photoshop CS6

In our previous Photoshop CS6 benchmark, we deliberately disabled the GPU acceleration features so we could get an apples to apples CPU comparison. However, in this test, we are enabling that acceleration to see what affect it has upon the benchmark numbers. Please remember: Photoshop’s GPU features only improve performance on SOME (rather than all) editing tools so performance will not scale in a linear fashion due to the CPU still playing a role.

RESULTS: Since our Photoshop benchmark combines CPU loads alongside several GPU-accelerated features, the i7 4770K and i5 4670K are able to remain well ahead. Ironically, even without GPU acceleration, they manage to be AMD's A10-5800K's accelerated score by a country mile.

 

[SKYMTL]

IGP Gaming Benchmarks (3DMark06 / 3DMark11)

IGP Gaming Benchmarks (3DMark06 / 3DMark11)

Integrated graphics processors have never been thought of as legitimate gaming devices but that stigma has begun to change. In this section, we test 3DMark06 and 3DMark11 to see how modern IGP systems reach to DX9 and DX11 environments.

RESULTS: Since many of the synthetic benchmark tests rely on the CPU and graphics subsystem together, we see Intel's Haswell holding on tight in this race. However, the improvement over last generation is impressive to say the least. If Intel continue on this route, they may be able to match AMD's performance when Broadwell rolls out.

 

[SKYMTL]

IGP Gaming Results (Deus Ex: HR / Dirt 3 / Skyrim)

IGP Gaming Results (Deus Ex: HR / Dirt 3 / Skyrim)

As with previous in-game tests, we have selected a number of games for our IGP-only testing suite. As you may expect, these benchmarks are run without a discrete card installed. All applications are tested at moderate detail settings (remember, these aren’t fully fledged discrete cards so we can’t expect miracles) in both DX11 and DX9 environments at a resolution of 1080P.

RESULTS: Once again we are seeing some rather impressive improvements for Haswell but that hasn't allowed the 4770K and 4670K to overcome AMD's Radeon-fueled lead.

 

[SKYMTL]

IGP Gaming Benchmarks (Super Street Fighter IV / Torchlight)

IGP Gaming Benchmarks (Super Street Fighter IV / Torchlight)

As with previous in-game tests, we have selected a number of games for our IGP-only testing suite. As you may expect, these benchmarks are run without a discrete card installed. All applications are tested at moderate detail settings (remember, these aren’t fully fledged discrete cards so we can’t expect miracles) in both DX11 and DX9 environments at a resolution of 1080P.

RESULTS: When compared against the latest AMD Trinity APUs, Intel doesn't really stand a chance here. While there have been some great performance increases over Ivy Bridge, AMD's superior driver stack and raw graphics throughput wins the day. However, it makes us wonder how the Iris and Iris Pro would fare here.

 

[SKYMTL]

System Power Consumption

System Power Consumption

Our power consumption numbers are broken down into two categories: one which simply stresses all of the CPU cores with WPrime and another which puts a high amount of load on both the CPU cores and the IGP. The latter will only be included if a given processor includes a dedicated internal graphics sub-processor.

For the CPU power consumption test, we use the standard testing system (with an NVIDIA GTX 670 installed) and wait until the system and discrete GPU are at idle speeds in order to log the idle power consumption. After this, WPrime 1024M is looped for 15 minutes while the power consumption is logged with a calibrated power meter to determine the peak watts.

IGP power consumption testing follows very much the same route as above but with some changes. First and foremost, the GTX 670 is removed and the video output is run through the processor’s graphics engine. In order to fully load the graphics cores and the primary processing stages within the CPU, we run the Unit Benchmark (in DX9 mode) from Civilization V for exactly 15 minutes.

Please note that after extensive testing, we have found that simply plugging in a power meter to a wall outlet or UPS will NOT give you accurate power consumption numbers due to slight changes in the input voltage. Thus we use a Tripp-Lite 1800W line conditioner between the 120V outlet and the power meter.

Many of Haswell’s architectural advances target power consumption and performance per watt. Considering the spectacular performance we have seen on previous pages, these results are right in line with expectations but they still set a new high water mark for efficiency. Not only is Haswell in a different dimension than comparable AMD processors (the FX-8350 for example) but it wins by a good margin against similar previous generation Ivy Bridge and Sandy Bridge CPUs. We really couldn’t have asked for more.

One interesting metric we see above is the small difference between the i7 4770K and i5 4670K. While both are 84W TDP processors, the i5 runs at lower speeds and has 2MB of its cache disabled so it naturally consumes a bit less power.

Even in IGP power consumption, the 22nm Intel Haswell processors are able to make quick work of the 32nm Trinity APU line from AMD. All things considered, these are once again some impressive results, especially against the i7 3770K.

 

[SKYMTL]

New Overclocking Features & Results

New Overclocking Features

In many ways, overclocking unlocked Haswell CPUs follows the process of previous generations but Intel has thrown several new metrics into the loop this time around. Some of these help add some much-needed granularity to the process while others may confuse first-time overclockers. With that being said, basic overclocks can still be readily achieved by increasing three primary values: the Turbo Ratio, the voltage and the Ring speed.

One of the more important modifications to the Haswell architecture is the integration of voltage regulation onto the CPU die instead of on the motherboard. This gives overclockers much better control over the various voltage planes which exist within these new processors. Unfortunately, for newcomers to this platform, the large number of voltage modifiers may prove to be a bit daunting at first.

As we mentioned earlier in this review, the core frequency still plays a large part in overclocking scenarios but the Ring speed (the interconnect between the various processing stages) can actually have the last say in whether an overclock is successful or not. In many situations we encountered, the Core speed could be increased by significant amounts but if the Ring frequency remained at default speeds, actual performance benefits were hard to come by.

Since this so-called Ring has now been decoupled, it has to be controlled as a separate entity but still one that’s connected at the hip to the primary cores and LLC. As such, it can run at a speed equal to or slower than the core frequency while maintaining its own voltage parameters. This is important to remember as basic Turbo Ratio overclocks can now be unstable due to the Ring frequency / voltage as well, rather than just Core parameters.

Naturally, BCLK overclocking is present here as well but with some interesting twists. Since the input to the CPU comes directly from the PCH and is limited by the PCI-E and DMI PLL interfaces, judicious use of the included ratios (5:5, 5:4, 5:3) will be necessary. In addition, BCLK overclocking should be done using the SB-PLL clock PLL setting while Turbo Ratio overclocking needs to use LC-PLL. Both of these can be found in the BIOS of all Z87 motherboards.

While there are more intricacies to overclocking Haswell than this section can hold, it’s important that we mention the new voltage options. The default voltage curve will remain relatively constant until Turbo frequencies are hit while the Interpolated setting will set adaptive over-voltage to the overclocked frequency range only.

The offset V/f override applies a constant over-voltage (or under-voltage depending on what’s needed) to the entire curve, providing increased stability throughout the clock range. Finally, Override V/f throws the baby out with the bath water and goes all in by setting a constantly high voltage regardless of clock speeds.

The Results

With all of that taken into account, how did we fare in our first tentative Haswell overclocking steps? Not as well as hoped unfortunately. This is no fault of the components (the MSI Z87-G45 Gaming provided an excellent platform) or the architecture but rather one of timing. We only had about two hours or so to achieve the best results and using a maximum Override voltage of 1.255V along with air cooling won’t grant all that much headroom. And yet, our ES chip easily hit a speed of 4.5GHz without going over the 75 degree level on a Noctua NH-U12S.

Remember, we only consider an overclock to be successful if it can maintain stability through 2D and 3D testing and that’s exactly what was achieved here. For an hour of overclocking followed by an hour or so of stability testing, 4.5GHz isn’t all that bad.

Haswell may not provide an overly intuitive overclocking experience for first-timers; there are a ton of options so its results are infinitely more rewarding than going through simple multiplier changes to achieve results. Personally, I felt the 4770K sample had quite a bit more headroom to play with so in the coming weeks additional overclocking updates will be posted….so stay tuned.

 

[SKYMTL]

Conclusion

Conclusion

Haswell may represent Intel’s newest processor architecture, incorporating many new features but its roots are still firmly grounded in the past. This combination has led to an interesting blend of previous generation highlights alongside new ideologies that focus primarily on performance per watt superiority, graphics leadership and accomplishing more without substantially increasing the processor’s resource pool. For the most part, blending a list of Sandy Bridge’s greatest hits with Haswell’s new caching hierarchy, instruction sets, processor graphics and other elements has led to some stunning successes. But do the i7 4770K and i5 4670K offer enough of an upgrade over the previous generation to fill the needs of today’s demanding enthusiasts? That’s debatable.

Intel have been continually able to achieve 5% to 15% higher performance with each successive generation and Haswell continues the tradition by coming down within that spectrum’s high end. In most processor-centric situations the i7 4770K beats the Ivy Bridge 3770K by an average of 12% without the need for higher clock speeds. The metrics between the i5 4670K and i5 3570K aren't that much different with a 13% gap between them. When taken at face value that may not seem like much but we have to remember these benefits have been achieved through targeted architectural enhancements rather than increased frequencies. That’s particularly impressive but it also reminds us of how conservative Intel has been on the clock speed front for the last few years.

As has been the case for some time now, Intel’s performance against AMD processors is straightforward: Haswell’s enthusiast-level parts don’t really line up with anything in the Trinity or Vishera stables. The i7 4770K thoroughly outperforms the FX 8350 in most games and applications despite AMD’s core and clock speed advantage. Meanwhile, the i5 4670K loses against the FX 8350 in several highly multi-threaded applications but it pulls ahead in games and any program that doesn’t directly benefit from Vishera’s eight cores.

Both Intel processors achieve their results while consuming significantly less power and Haswell’s addition of FMA3 and AVX2 instruction sets will likely pay dividends in the future. More importantly, these new processors remain an excellent performer in programs that use legacy and current generation compilers, unlike the FX 8350 and any other Piledriver-based architecture which tends to choke on programs developed using slightly older technology.

AMD is forging ahead with their GPU-centric APUs so Intel needed to step things up in the graphics area if they wanted to compete in compute-heavy workloads. The new HD4000 processor graphics engines’ additional shaders and updated fixed function stages go a long way towards narrowing a once massive gap but most desktop Haswell processors still lack the horsepower necessary to surpass Trinity, let alone AMD’s upcoming Richland parts. However, Intel’s approach in this area is very much two-pronged and their Iris 5000-series doubles up on performance so it could theoretically outperform most APUs and a fair number of entry level discrete cards. Due to an understandable reluctance to use higher end processor graphics in a desktop market dominated by discrete graphics solutions, Iris will unfortunately only be available within a few select desktop SKUs and mobile processors….for now.

Let’s talk value for a moment since it tends to be one category where Intel’s competition excels. If scene rendering or other professional tasks are at the top of your “to do” list, then AMD’s higher end Vishera products like the FX 8350 can provide a surprising amount of bang for buck. On the other hand, for gaming-focused systems and the programs most enthusiasts use, Haswell’s quad thread i5 4670K is currently an amazing buy since it provides top-shelf performance and there are very few titles which actually take advantage of more than four concurrent threads. OpenCL compute performance has also been substantially increased with Intel slowly closing the performance gap in this area.

The actual “value” proposition behind the i7 4770K a bit more dubious on paper. At $350 it is a whopping 78% more expensive than an FX-8350, not to mention that you could almost buy three A10-5800K APUs for the same price. But none of those solutions would provide someone with anywhere near this CPU’s well-rounded capabilities so its value lies in the eyes of enthusiasts rather than penny pinchers. The i7 4770K is currently the fastest processor around, overclocks reasonably well and in many situations it beats the aforementioned AMD offerings by a country mile. With those things in mind, performance per dollar can be safely thrown out the window.

So who are these new processors really geared towards? Anyone still hanging onto higher end Core 2 Duo or Clarkdale chips may want to give serious consideration to Haswell since it brings tangible platform and computational benefits to the table. Even people with an i7 2600K or i5 2500K and a Z68 motherboard will realize some not-so-insignificant advantages, though any benefits will be tempered by the cost of a comparable Haswell-based system. Enthusiasts using Ivy Bridge on the other hand should steer away and wait since the i7 4770K and i5 4670K will offer minimal real-world benefits over their predecessors.

While notebook and high end tablet users will likely benefit from the lion’s share of Haswell’s improvements, the desktop market certainly hasn’t been left out in the cold. The seemingly minor architectural changes have allowed the i7 4770K and i5 4670K to become the CPUs of choice for gaming and general computing scenarios. While neither offers any tangible performance improvements over comparable Ivy Bridge processors, they will provide a highly efficient heart for tomorrow’s computing needs while also satisfying the demands of overclockers.

i7 4770K

i5 4670K​