Intel Core i7-5775C review: The unwanted desktop Broadwell has one neat trick
As someone who watched this story unfold, I know the truth is actually the opposite. You see, Intel's 3.3GHz Core i7-5775C is actually something the Internet demanded.
Remember that in 2012, the Internet said Intel would effectively "kill the desktop" PC because the company had no plans for a socketed CPU past Haswell.
The truth is, Broadwell was always intended as a lower-power part with no practical appeal outside of laptops, NUC-style computers and all-in-one PCs. Instead, Intel told me at the time, desktop sockets will get a workout with the next big thing: Skylake.
Anger and hand-wringing ensued. Soon, bowing to pressure from its die-hard desktop community and PC vendors, Intel rolled over and decided to make a socketed version of Broadwell.
That brings us to the new Core i7-5775C--the CPU Intel never wanted. Now it looks like not even the crowd that demanded the chip will want it.
What Broadwell is
Intel announced 10 Broadwell CPUs in June at Computex. Half will go into laptops, and another three will be soldered on motherboards in all-in-one or Mini PCs, as Intel originally planned. But of those 10, two will fit into the traditional socket that desktop users want. That's what I'm looking at today.
The Core i7-5775C will work in most 9-series motherboards and systems out there with BIOS support. Just be advised, that's not exactly straightforward. To get my Core i7-5775C up and running, I didn't just have to update the BIOS; I had to update it the right way. Check with your motherboard maker if you plan to go with a Broadwell C part.
Broadwell is a 14nm-process CPU that should offer a 5-percent or so performance increase over Intel's 4th-generation Haswell CPUs, if all things were even. My own tests of Haswell vs. Broadwell mobile parts confirmed that.
Everything isn't even, though. The pair of Broadwell C chips Intel has produced feature a massive 128MB of Level 4 (L4) cache using embedded DRAM. This cache is slower than the cache integrated into the CPU itself ,but because it's actually a chip sitting next to the CPU and wired directly to it, it's a magnitude faster than using system RAM.
This isn't a new trick. Intel actually sold three Haswell CPUs with a similar 128MB L4 cache aboard. Those CPUs were available only in NUC-style machines like the Gigabyte Brix Pro and soldered down to the motherboard. Like the Broadwell desktop chip, the eDRAM's existence is mostly to address graphics performance, where memory bandwidth is king.
How we tested
For my test, I used an Asus Z97-Deluxe motherboard with 16GB of DDR3/1600 and 240GB SSD. I updated the motherboard to the latest UEFI available, installed a fresh copy of 64-bit Windows 8 and hunted down the latest drivers as well. Because the graphics capability of the the Broadwell C is a driving feature, I tested it without a discrete GPU.
I realize that decision doesn't track with what most people are doing with these expensive CPUs. I'd bet it's close to 90 percent who pair higher-end CPUs with graphics cards for gaming. Still, the graphics performance of these chips is a very important feature.
The competitor was a natural choice: Intel's Core i7-4790K, aka Devil's Canyon.
The Core i7-4790K has a clock speed of 4GHz and will Turbo Boost up to 4.4GHz on some loads. It has Intel HD4600 graphics and on the street sells for $340. For more details, I've lined both them up at Intel's ARK website. It's pretty much the pinnacle of quad-core Haswell performance.
Besides the Broadwell to Haswell generation difference, the other factor that matters here is the clock speeds between the two chips. The Broadwell Core i7-5775C has a base clock of 3.3GHz with a Turbo Boost of 3.7GHz.
Encoding
The first test we can discuss is our Handbrake workload. We use the free and popular Handbrake encoder to convert a massive 30GB MKV file into a much more compact MP4 file using the Android Tablet preset.
Notes: Our normal system review benchmark uses the 0.9.9 version but for this, I wanted to use the latest version of Handbrake 0.10.2. Because it does effect the encoding time, you can't compare results between the versions. Handbrake allows you to choose different encoders so for the first one, I used x264, which is heavily multi-threaded and keeps the workload to the x86 cores in the CPU.
The result was a big win for the Core i7-4790K and clock speed in general.
But wait, encoding video isn't done just on the x86 side anymore. Today, encoding on the GPU is the way to go for performance, and Intel has dedicated transistors for hardware encoding into its chips, called Quick Sync. What happens when we give Handbrake the same task, but using the graphics cores instead
Boom: The improved Broadwell graphics cores combined with its massive eDRAM flips this battle on its back. Besides whistling at how much of a recovery the Core i7-5775C made here, you should also take note of the encoding time it took using the GPU cores rather than the x86 cores in the CPUs. The Core i7-5775C, for instance, takes a third of the time to get the work done. There are arguments that GPU encoding leads to visual impurities, but when you're crunching a file down to watch on your phone, who cares
3D rendering
Next up, we used Cinebench R15, a test based on Maxon's Cinema 4D animation package. It's a great benchmark that is pure CPU and heavily multi-threaded. The result is no surprise, as 3.7GHz even with the newer Broadwell cores can't beat 4.4GHz.
Work performance
Our next test is PCMark 8, which simulates various computing tasks broken down into Work, Home and Creative categories. Work measures general Office Drone tasks, Creative throws in photo, video and light gaming, while Home factors in more casual gaming.
The first result is from PCMark 8 Work Conventional. Again, the Core i7-5775C's lack of clock speed shows even on these basic tasks that a Celeron could run.
Creative performance
Once PCMark 8 mixes gaming performance into this test, the Broadwell makes a nice comeback. The result is similar in the PCMark 8 Home test, which adds gaming into the mix and puts the Broadwell in a much better light than the Haswell, thanks to its better gaming capability.
Chess performance
The Fritz 12 Chessbenchmark measures a CPU's ability to calculate chess moves. The yard stick is a 1GHz Pentium III. A score of 10 would mean the tested CPU is 10 times faster than a 1GHz Pentium III performing the same tasks. The results again don't look great for the Broadwell desktop part. Even though we know Broadwell's cores are maybe 5 percent more efficient doing the same task, it's not enough to compensate for the higher clock speeds of the Core i7-4790K chip.
Overall meh right
If you're thinking 'stick a fork in Broadwell, and call it a day,' you need to read on to see where the Core i7-5775C is better than Haswell.
Where Broadwell really shines
On the compute side, the Broadwell desktop part can't hang with the higher clock speeds of the Haswell chip but what happens when the GPU is the primary driver of the task It's a different situation.
For reference, I'm going to toss in the brand-new "Godavari" AMD A10-7870K chip. It's technically a 12-core CPU by AMD's standards, but that really means it's a quad-core CPU with eight GPU cores in it. The x86 cores run at 3.9GHz to 4.1GHz, and the integrated Radeon R7 graphics cores buzz along at 866MHz.
While AMD's x86 performance has been weak sauce these last few years, the graphics performance has rained pain on Intel's just as badly. The best part of AMD's Godavari is its price. It's a chip with a $137 list price, but it's actually selling for more, at $149. That's less than half of either Intel CPUs. Some may cry foul at using the Godavari because it's so much cheaper, but I think it's fair to put the APU's graphics performance in context.
Note: I tested the A10-7870K with its RAM set to DDR3/1600 to match the Intel systems and also grabbed the latest Catalyst driver available.
Synthetic gaming performance
First up is the popular 3DMark. This is a great showing for the Core i7-5775C Broadwell chip when compared to the Core i7-4790K Haswell chip and shows just how far Intel has come. In fact, Intel has always said Broadwell was a minor tick in performance on the x86 side, but its graphics performance was a major tock. Helped by the eDRAM in the chip, that seems to be true. Intel has finally caught up with AMD's integrated graphics performance. Of course, there is that price discrepancy between the AMD and Intel chip to consider.
Firestrike is a little over the true capability of these CPUs so I also ran the easier Sky Driver test. Interestingly, the Core i7-5775C starts to pull away from the AMD A10-7870K in the overall score. That's likely due to the more efficient Intel cores, boosting its final scores higher.
Dirt Showdown
Enough with the synthetic tests. To see how the integrated graphics would perform in actual games, I threw Code Master's Dirt Showdown at all three. The result is a huge win for the Broadwell desktop chip. Dirt Showdown actually runs fairly well at 1920x1080 on all three, too. I would never play a first person shooter at 32fps, but in driving games, where you don't whip around like you do in a shooter, it's surprisingly tolerable.
BioShock Infinite
Moving on to Irrational Games' BioShock Infinite, a lot of visual quality settings had to be compromised to get to playable framerates. The result is a big win for the Core i7-5775C again. It leaves the AMD APU in the dust, and its sibling is a very distant third. If I were really attempting to play BioShock Infinite on integrated graphics, I'd notch it down to 1366x768 to get the frame rates higher.
Tomb Raider
The last game I ran was Square ENIX's Tomb Raider. Like the other games, it shines on the Core i7-5775C chip. At 1920x1080 resolution, the Broadwell chip can even run it at the magical 60+ frames per second. You make a lot of compromises in visual quality, though. I'd personally run it at 1366x768 at normal quality, where the game looks much better and the Broadwell can push 72 fps. The AMD A10-7870K interestingly takes a back seat to even the Core i7-4790K here. That may indicate Tomb Raider's dependence on the x86-side, and the AMD's chips weaker cores dragging it down a little.
OpenCL performance
The final graphics test I ran was Luxmark 3.0, a benchmark that measures OpenCL performance. OpenCL is a way to run general computing tasks on a graphics chip. The story is the same: the Core i7-5775C clocks the other two CPUs by a healthy margin.
The actual surprise is the A10-7870K, which loses badly to the Core i7-4790K chip. I didn't expect the Godavari to ace the Broadwell and its big cache, but losing to the Haswell chip was a surprise.
Conclusion
There's no way to talk about the Broadwell socketed chip without talking about what's coming: Skylake. Intel's next desktop chip is rumored to be dropping soon, and Skylake is the rightful successor to Haswell. So why even waste time with Broadwell
After all of my testing, I think that conclusion is probably the right one. If Broadwell in a socket had been introduced six months ago, it might have had a fighting chance. But now No.
If Skylake does appear soon and if it does introduce a new socket with it, no one in their right mind would build or buy a new system using a Broadwell chip.
Even the strongest argument for Broadwell in a socket is very niche. In general use, it's actually slower than the cheaper Haswell chip.
Broadwell's strongest point is actually as an integrated graphics chip. With its embedded DRAM it smokes all other integrated graphics you can buy today by a huge margin. It's actually capable of playing some games at 1920x1080 at the magical 60+ fps frame rate at lowered image quality settings, which is amazing for integrated graphics.
But who spends $366 on a CPU to play games without a graphics card As good as the Core i7-5775C is for an integrated graphics chip, a $150 GPU would run circles around it. Take that $150 GPU and pair it with a $200 Core i5, and you'd get better gaming results.
You may certainly make some arguments that its best use would be in a space-constrained mini system or All-in-One PC but that's the perfect place for a soldered version of the chip, not a socketed version.
That's also just like Intel originally planned.