System Builder Marathon, Q1 2014: The Articles
Here are links to each of the four articles in this quarter’s System Builder Marathon (we’ll update them as each story is published). And remember, these systems are all being given away at the end of the marathon.
To enter the giveaway, please fill out this SurveyGizmo form, and be sure to read the complete rules before entering!
Day 1: The $2400 Reader's Choice PC
Day 2: Our New Enthusiast PC
Day 3: The $750 Gaming PC
Day 4: Performance And Value, Dissected
Introduction
Although Paul, Don, and I typically get vastly disparate budgets, we typically try to push as much performance as possible from the money we're given in the System Builder Marathon. That methodology changed a little this quarter when budget machine builder Paul Henningsen went rogue by starting with a $50 off-the-top cut. Our typical spread of 1x, 2x, and 3x budgets would have resulted in an $800 budget. But he decided that there simply wasn’t much he could do to noticeably improve performance with that last $50.
At the opposite end of the scale, I wasn’t ready to turn his $50 reduction into a $150 haircut, which would have been necessary for us to honor the usual price spread. After last quarter's Marathon, readers requested that I make two major changes to my previous effort, and those alterations wouldn't fit neatly into an otherwise-similar $2250 machine. Don didn't have a problem with that; by sticking with my original $2400 budget, he could keep his $1600. And as we saw a couple of days ago, he needed all of that plus some.

That’s where the real fun begins. Nvidia trimmed the prices on its GeForce GTX 780 and 780 Ti in response to AMD’s Radeon R9 290 and 290X. Shortly after that, a shortage of cards caused a surge in Radeon pricing. Don suddenly found himself able to switch from a pair of GeForce GTX 770s to a single GTX 780 Ti, and still had money left over for an upgrade from Intel's Core i5 to an i7, a higher-end case, and a Blu-ray disc burner.
Price drops on the GeForce GTX 780 weren't as significant, so the requests I received to use two left me without the money to keep the Blu-ray drive or Ivy Bridge-E-based CPU on a machine that, considering price alone, should still be able to do everything better.
| Q1 2014 System Builder Marathon Components | |||
|---|---|---|---|
| $750 Gaming PC | $1600 Enthusiast PC | $2400 Reader's PC | |
| Processor | Intel Core i3-4130: 3.4 GHz, Dual-Core, 3 MB Shared L3 Cache | Intel Core i7-4770K: 3.5 - 3.9 GHz, Quad-Core, 8 MB Shared L3 Cache | Intel Core i7-4770K: 3.5 - 3.9 GHz, Quad-Core, 8 MB Shared L3 Cache |
| Graphics | Zotac ZT-70301-10P GeForce GTX 770 2 GB | Galaxy 78NNH5DV8GGX GeForce GTX 780 Ti 3 GB | 2 x EVGA 03G-P4-2781-KR GeForce GTX 780 3 GB (SLI) |
| Motherboard | Asus H81M-K: LGA 1150, Intel H81 Express | ASRock Z87 Pro3: LGA 1150, Intel Z87 Express | ASRock Z87 Extreme4: LGA 1150, Intel Z87 Express |
| Memory | Adata XPG-2 AX3U1600W4G9-DGV: DDR3-1600 C9, 8 GB (2 x 4 GB) | Corsair Vengeance LP CML8GX3M2A1866C9B: DDR3-1866 C9, 8 GB (2 x 4 GB) | G.Skill Ripjaws X F3-1866C9D-16GXM: DDR3-1600 C9, 16 GB (2 x 8 GB) |
| System Drive | Western Digital WD10EZEX: 1 TB, 7200 RPM, 64 MB Cache | Samsung 840 EVO MZ-7TE120BW: 120 GB, SATA 6Gb/s SSD | SanDisk Ultra Plus SDSSDHP-256G-G25: 256 GB, SATA 6Gb/s SSD |
| Storage Drive | (Uses System Drive) | Western Digital Black WD5003AZEX: 500 GB, 7200 RPM, 64 MB Cache | Seagate Barracuda ST2000DM001: 2 TB, 7200 RPM, 64 MB Cache |
| Optical | Asus DRW-24B1ST/BLK/B/AS: 24x DVD±R, 48x CD-R | LG WH14NS40: 14x BD-R, 2x BD-R, 16x DVD+R | Lite-On iHAS124-04: 24x DVD±R, 48x CD-R |
| Case | Rosewill Line-M MicroATX | NZXT Phantom 410 CA-PH410-B3 | NZXT Phantom 410 CA-PH410-G1 |
| Power | Rosewill Capstone-450-M: 450 W Semi-Modular, ATX12V v2.31, 80 PLUS Gold | Corsair TX650: 650 W ATX12V v2.3, 80 PLUS Bronze | Corsair HX750: 750 W Semi-Modular, ATX12V v2.3, 80 PLUS Gold |
| CPU Cooler | Intel Core i3 Boxed Fan (included) | Cooler Master Hyper 212 EVO | Thermaltake CLW0217 Water 2.0 Extreme |
| PWM Fan | (Uses CPU Fan) | (Uses CPU Fan) | Antec Spot Cool Blue LED Fan |
| Total Cost | $755 | $1615 | $2310 |
Time and again, you've seen us demonstrate that the sweet spot for getting the most performance from your dollar requires a budget in the $700 to $1100 range, or thereabouts. That's a moving target, but it still leaves us expecting Paul's $750 gaming PC to win our quest to maximize value.
It'll be more interesting to see whether my $2400 PC can continue to dominate Don's $1600 effort across most benchmarks. Since our two top machines now use the same CPU, the only hope for my build is that Don somehow screwed up his $1600 configuration.

By using a non-K-series processor and a motherboard without much overclocking flexibility, Paul misses most of the thrill that would have come from tweaking his $750 system. Graphics is all that remains for him to optimize.
In the first day of this quarter's System Builder Marathon, I mentioned that my huge liquid cooler could be seen by some as a waste of money, since the Core i7-4770K's die seems to suffer from thermal density issues. I guessed I'd shed no more than 3 °C and gain no more than 100 MHz compared to my results with larger air coolers. We see that 100 MHz difference here, even though the $1600 machine’s heat sink is slightly smaller those I had in mind.
| Test Hardware Configurations | |||
|---|---|---|---|
| $750 Gaming PC | $1600 Enthusiast PC | $2400 Reader's PC | |
| Processor (Overclock) | Intel Core i3-4130: 3.4 GHz, Two Physical Cores No O/C | Intel Core i7-4770K: 3.5 - 3.9 GHz, Four Physical Cores O/C to 4.4 GHz, 1.20 V | Intel Core i7-4770K: 3.5 - 3.9 GHz, Four Physical Cores O/C to 4.5 GHz, 1.25 V |
| Graphics (Overclock) | Zotac GeForce GTX 770: 1098 MHz GPU, GDDR5-7010 O/C to 1283 MHz, GDDR5-7610 | Galaxy GeForce GTX 780: 928 MHz GPU, GDDR5-7000 O/C to 1128 MHz, GDDR5-7300 | 2 x EVGA GeForce GTX 780: 902 MHz GPU, GDDR5-6008 O/C to 1059 MHz, GDDR5-6720 |
| Memory (Overclock) | 8 GB Adata DDR3-1600 CAS 9-9-9-24, No O/C | 8 GB Corsair DDR3-1866 CAS 9-10-9-27, No O/C | 16 GB G.Skill DDR3-1866 CAS 9-10-9-28, O/C to DDR3-2133 CL 10-11-10-28, 1.585 V |
| Motherboard (Overclock) | Asus H81M-K: LGA 1150, Intel H81 Express Stock 100 MHz BCLK | ASRock Z87 Pro3: LGA 1150, Intel Z87 Express Stock 100 MHz BCLK | ASRock Z87 Extreme4: LGA 1150, Intel Z87 Express Stock 100 MHz BCLK |
| Case | Rosewill Line-M | NZXT Phantom 410 | NZXT Phantom 410 |
| CPU Cooler | Intel Core i3 boxed cooler | Cooler Master Hyper 212 EVO | Thermaltake CLW0217 Water 2.0 Extreme |
| Hard Drive | Western Digital WD10EZEX: 1 TB, 7200 RPM, 64 MB Cache | Samsung MZ-7TE120BW 120 GB, SATA 6Gb/s SSD | SanDisk SDSSDHP-256G-G2 256 GB SATA 6Gb/s SSD |
| Power | Rosewill Capstone-450-M: 450 W Semi-Modular, 80 PLUS Gold | Corsair TX650: 650 W, 80 PLUS Bronze | Corsair HX750: 750 W Semi-Modular, 80 PLUS Gold |
| Software | |||
| OS | Microsoft Windows 8 Pro x64 | ||
| Graphics | Nvidia ForceWare 332.21 WHQL | Nvidia GeForce 334.89 | Nvidia GeForce 335.23 |
| Chipset | Intel INF 9.4.0.1017 | Intel INF 9.4.0.1026 | Intel INF 9.4.0.1026 |
| Benchmark Configuration | |
|---|---|
| 3D Games | |
| Battlefield 4 | Version 1.0.0.1, DirectX 11, 100-sec. Fraps "Tashgar" Test Set 1: Medium Quality Preset, No AA, 4X AF, SSAO Test Set 2: Ultra Quality Preset, 4X MSAA, 16X AF, HBAO |
| Grid 2 | Version 1.0.85.8679, Direct X 11, Built-in Benchmark Test Set 1: High Quality, No AA Test Set 2: Ultra Quality, 8x MSAA |
| Arma 3 | Version 1.08.113494, 30-Sec. Fraps "Infantry Showcase" Test Set 1: Standard Preset, No AA, Standard AF Test Set 2: Ultra Preset, 8x FSAA, Ultra AF |
| Far Cry 3 | V. 1.04, DirectX 11, 50-sec. Fraps "Amanaki Outpost" Test Set 1: High Quality, No AA, Standard ATC, SSAO Test Set 2: Ultra Quality, 4x MSAA, Enhanced ATC, HDAO |
| Adobe Creative Suite | |
| Adobe After Effects CC | Version 12.0.0.404: Create Video which includes 3 Streams, 210 Frames, Render Multiple Frames Simultaneously |
| Adobe Photoshop CC | Version 14.0 x64: Filter 15.7 MB TIF Image: Radial Blur, Shape Blur, Median, Polar Coordinates |
| Adobe Premeire Pro CC | Version 7.0.0 (342), 6.61 GB MXF Project to H.264 to H.264 Blu-ray, Output 1920x1080, Maximum Quality |
| Audio/Video Encoding | |
| iTunes | Version 11.0.4.4 x64: Audio CD (Terminator II SE), 53 minutes, default AAC format |
| LAME MP3 | Version 3.98.3: Audio CD "Terminator II SE", 53 min, convert WAV to MP3 audio format, Command: -b 160 --nores (160 Kb/s) |
| HandBrake CLI | Version: 0.99: Video from Canon Eos 7D (1920x1080, 25 FPS) 1 Minutes 22 Seconds Audio: PCM-S16, 48,000 Hz, Two-channel, to Video: AVC1 Audio: AAC (High Profile) |
| TotalCode Studio 2.5 | Version: 2.5.0.10677: MPEG-2 to H.264, MainConcept H.264/AVC Codec, 28 sec HDTV 1920x1080 (MPEG-2), Audio: MPEG-2 (44.1 kHz, 2 Channel, 16-Bit, 224 Kb/s), Codec: H.264 Pro, Mode: PAL 50i (25 FPS), Profile: H.264 BD HDMV |
| Productivity | |
| ABBYY FineReader | Version 10.0.102.95: Read PDF save to Doc, Source: Political Economy (J. Broadhurst 1842) 111 Pages |
| Adobe Acrobat 11 | Version 11.0.0.379: Print PDF from 115 Page PowerPoint, 128-bit RC4 Encryption |
| Autodesk 3ds Max 2013 | Version 15.0 x64: Space Flyby Mentalray, 248 Frames, 1440x1080 |
| Blender | Version: 2.68A, Cycles Engine, Syntax blender -b thg.blend -f 1, 1920x1080, 8x Anti-Aliasing, Render THG.blend frame 1 |
| Visual Studio 2010 | Version 10.0, Compile Google Chrome, Scripted |
| File Compression | |
| WinZip | Version 18.0 Pro: THG-Workload (1.3 GB) to ZIP, command line switches "-a -ez -p -r" |
| WinRAR | Version 5.0: THG-Workload (1.3 GB) to RAR, command line switches "winrar a -r -m3" |
| 7-Zip | Version 9.30 alpha (64-bit): THG-Workload (1.3 GB) to .7z, command line switches "a -t7z -r -m0=LZMA2 -mx=5" |
| Synthetic Benchmarks and Settings | |
| 3DMark 11 | Version: 1.0.5.0, Benchmark Only |
| 3DMark Professional | Version: 1.2.250.0 (64-bit), Fire Strike Benchmark |
| PCMark 8 | Version: 1.0.0 x64, Full Test |
| SiSoftware Sandra | Version 2014.02.20.10, CPU Test = CPU Arithmetic / Multimedia / Cryptography, Memory Bandwidth Benchmarks |
Several application benchmarks have been updated for today’s presentation, so some of the numbers might appear inconsistent with previous articles, including individual build articles from this series.
Futuremark’s graphics benchmark heavily favors the $2400 machine’s two GeForce GTX 780 cards, producing a chart of nearly perfect performance scaling between five of the six tested configurations. The $750 machine doesn’t live up to its scaling expectations when Paul overclocks it, but that's probably because it's constrained by a multiplier-locked CPU.

That same $750 machine outperforms Don’s $1600 PC in PCMark’s “Work” test, at least at baseline frequencies. At first I was unable to explain Don's loss. But after reading his write-up, I figured out that he was running his baseline numbers with his memory underclocked to DDR3-1333.

PCMark’s storage suite uses traces recorded from a number of productivity, content creation, and entertainment apps to emphasize real-world differences between drives. That's why a lot of SSDs will appear incredibly similar in this particular metric.
This is the only synthetic benchmark used in our value scoring, which is going to give the two SSD-equipped boxes an advantage over the $750 machine.
Sandra’s Arithmetic benchmark slaps my $2400 PC around with a dose of reality. Of course, I should have expected it. Because the top-end and mid-level systems use the same Core i7-4770K CPU, they fare comparably. As consolation for what appears to be a waste of cash, the expensive machine’s higher-end cooling subsystem facilitates a slightly higher overclock. I can also be glad that pure synthetics aren't included in our final value calculation.

The $750 PC only has two physical cores, and even then it fails to produce half the Dhrystone performance of its rivals. Its lack of overclocking capability is also significant, though Intel recently offered hope for future low-cost machines.

The bandwidth-sensitive Cryptography module punishes Don's $1600 PC, since he effectively ran the stock configuration with underclocked memory. In contrast, it rewards the $2400 PC’s memory overclock, even though both systems employ the exact same CPU.

That same difference in memory bandwidth is vividly exposed in the Memory Bandwidth module, where even the $750 machine outperforms Don's baseline configuration.
We all knew that the $2400 PC’s two GeForce GTX 780 cards would outperform the $1600 machine's single GeForce GTX 780 Ti, and that both configurations would be limited by Battlefield's 200 FPS cap at our lowest-detail test settings.
The most significant findings gleaned from running tests at Medium quality appear at 4800x900, where Paul's $750 build reaches nearly 73% of the $1600 machine's frame rate.

The $750 setup appears at least somewhat playable at Battlefield 4's Ultra preset using a 4800x900 resolution. Even if that sounds marginal, the performance Paul enabled far exceeds his goal of smoothness at 1920x1080.

Arma 3 appears either CPU- or memory-bound at our standard quality settings, using resolutions up to 1920x1080. Yet, there are problems with both hypotheses. First, the $1600 and $2400 systems have the same CPU, but achieve vastly different frame rates. Second, the $750 PC pushes more memory bandwidth than the $1600 baseline, and both generate similar frame rates.

My $2400 effort barely outpaces Don's $1600 competition at medium resolutions and Arma 3’s Ultra quality preset. But it establishes a more significant lead at 4800x900 and 5760x1080. The $750 setup easily surpasses its 1920x1080 target, but struggles to offer playability at higher resolutions using the same settings.

Grid 2 tends to be limited by memory bandwidth at its High quality settings, though enthusiast-class graphics cards are needed to reach those limits. That could explain why, at DDR3-1866 and 5760x1080, both the overclocked $1600 PC and baseline $2400 PC reach 146.3 FPS.
Moreover, the DDR3-1600 and GeForce GTX 770-equipped $750 gaming box nearly catches Don's baseline configuration at 4800x900.

The Ultra quality preset taxes most graphics configurations intensively enough to pull frame rates below where they're constrained by system memory bandwidth, though the dual-card $2400 PC looks like it needs even more throughput to support its second card at lower resolutions.
None of this talk of bottlenecks will matter to most players however, since even Paul's $750 effort is powerful enough to push the game smoothly through our highest detail settings.

We're accustomed to Far Cry 3 leaning hard on graphics resources; the scaling of our chart suggests that overhead associated with SLI might be slowing my dual-GPU machine down, though. That ceiling disappears at 5760x1080, as GPU performance finally becomes the more prominent limiting factor.
Paul's $750 build also appears bottlenecked, and again that’s probably due to a CPU limitation. Armed with a weaker single GPU, its bottleneck disappears at 4800x900.

The Ultra preset puts even more stress on the GPUs, making the $2400 machine’s bottleneck less apparent at 1600x900. Paul's box achieves its 1920x1080 goal. In fact, Paul even says that a drop to 2x AA makes enough of a difference to play at 4800x900.

Apple iTunes and Lame MP3 are single-threaded. All systems in this competition use the same CPU architecture, and results almost reflect clock speed. The $2400 machines baseline encoding times don’t fall in line with that theory, however.
The $2400 and $1600 machines both come armed with Core i7-4770K CPUs, so what could explain the difference in encoding times at stock clock rates?
Well, I enabled my kit's XMP memory profile right out of the gate, and my best guess is that the board's firmware also enabled enhanced multipliers at the same time (this is when a motherboard vendor sneakily pushes all cores to maximum Turbo Boost settings). This is why I don't/can't use XMP in my motherboard round-ups.

HandBrake and TotalCode Studio are optimized for threading, so the $750 box gets left far behind. Small differences between my overclocked $2400 machine and Don's $1600 setup can be attributed to the 100 MHz difference. On the other hands a performance delta at stock settings again appear to be caused by CPU multiplier manipulation by my system's firmware with XMP enabled.


Lest we forget this is a competition, I previously mentioned that my only hope to win was that Don might have configured his $1600 platform incorrectly, since both of our boxes lean on the same CPU.
Although it appears that Adobe After Effects is pointing to a problem with his box, I think the issue is actually something outside of Don's control. Our After Effects workload is sensitive to available memory per logical thread. Because we're using the same CPU and I have twice as much RAM, I'm seeing a tremendous performance advance.



Don has another potential issue in Acrobat 11; his Core i7-4770K-equipped platform should be spinning up to 3.9 GHz in a single-threaded workload. But he loses to Paul's 3.4 GHz part at stock settings. Could it be this is another memory bandwidth-sensitive metric?

I’ve never known 3ds Max to be limited by available memory bandwidth, but I have often wondered why an application that’s fully threaded doesn’t always show consistent performance scaling.
After the fairly severe discrepancy between my machine and Don's in 3ds Max, the $1600 PC lands in a more predictable position when we run our Blender benchmark. Apparently, poorly-optimized memory don't affect the fully-utilized Core i7-4770K in that particular metric.



Unlike Blender, Visual Studio compile times are affected by everything, from CPU cache size to SATA driver version. A mediocre showing for the $1600 PC is not unusual, and may indeed be due to its memory's low data rate.

My $2400 PC features more aggressive DRAM settings and a significantly faster SSD compared to the $1600 PC, and either of those traits could explain its better showing in 7-Zip. Combine those traits and, as the builder of the $2400 machine, I’m a happy camper.
The $750 PC has a slower mechanical hard drive, a lower CPU frequency, and half as many CPU cores.


WinRAR appears primarily CPU-restricted, but at least one of our WinZip compression methods is boosted by graphics power. The baseline $1600 machine actually catches its expensive combatant in WinZip’s CPU-optimized workload, though that’s something we expected its identical CPU to do in nearly every benchmark.

They might use the same processor and only slight variations of an identical GPU, but the $2400 PC has twice as many graphics cards as the $1600 machine. That’s quickly illustrated in the power consumption numbers, where the $2400 machine needs 802 W from the socket to output a little less than 700 W from its power supply.

Consuming only 40 W at idle and less than 300 W fully-loaded, the $750 machine is a true miser. Budget-conscious enthusiasts with just enough money for one of the bigger machines (but don’t really need the extra punch) might want to also consider the effect on their power bills.

The $2400 PC runs far hotter than its $1600 rival, even with the same CPU (and even though it has at least four times the cooling capacity). It is clocked higher though, and has enough voltage to keep it stable all the way up to its thermal threshold, where it'd throttle back.
Intel promises to fix Haswell’s heat issues eventually, though the fix might be limited only to new, higher-model cores.
Christopher Ryan, our storage expert, might be able to show you the performance difference between my machine's 256 GB SSD and the $1600 machine’s 120 GB device, PCMark’s productivity, content creation, and gaming traces do not. Then again, load times are also affected by the number of CPU cycles it takes to load these apps into RAM. Both machines score 85% higher than the hard drive-equipped $750 PC.

Gamers will be more interested in the ~18% difference between a single GeForce GTX 780 Ti and dual GeForce GTX 780s. But those are averages. Several benchmarks are bottlenecked at our lower test settings, so we need to address higher resolutions separately.
Nothing short of a perpetual motion machine could be more than 100% efficient, so our efficiency chart is zeroed out by subtracting one (100%) from its performance/efficiency calculations.

Remembering how miserly the $750 PC appeared in our power charts, its loss in average efficiency is a surprise. The $1600 PC had some minor configuration issues, so its efficiency win is also surprising. In the end, Don's build gets there by performing up to 85% better than the $750 machine, while using less than 74% more power.
A quick look at price-to-performance using purchase prices might not help you much today (since our parts were bought more than a month ago), but it can at least validate or negate each of our decisions. We do, after all, see a lot of Monday morning quarterbacking after new prices emerge.

The $750 machine yields the most bang for the buck, as expected, and the $1600 machine serves up $0.72 worth of extra performance for every dollar spent. The $2400 machine provides 58 cents of performance for every dollar spent beyond the $750 baseline, but over 80 cents in performance for every dollar spent over the $1600 PC (58%/72%, above). In other words, waste is a smaller percent of my $2400 machine's budget when I use the $1600 PC as a starting point.

Value improves slightly for the $1600 and $2400 machines after a couple months of price changes, but neither configuration will ever catch the $750 platform's high value. It would be better for enthusiasts who want a high-end machine to instead discuss their minimum performance standards.
Speaking of raising the bar on standards, gamers are the most vocal critics of our builds, and the $750 machine just isn’t quite capable of playing through our most taxing detail settings at 5760x1080. We know that because it failed a couple of tests at 4800x900.

Starting at 99.6% of the $1600 machine’s high-end gaming value, the $2400 PC appears nearly on-par with its rival...until I started overclocking it. More than $80 extra was spent to make it a better overclocker, and that expense is returned with a victory comparing my overclocking successes to Don's.

Recent price changes are even more forgiving to my $2400 effort, as it starts life with a 1.8% gaming-value lead that climbs to 7.7% (125.2%/116.3%, above) when both machines are overclocked.
Still, none of us can agree on which of these to buy. Aside from the $750 machine’s general value victory, the rest of our charts are just numbers, at the end of the day. That’s because each of us uses different games, applications, and monitor configurations in our own systems. That’s why I recommend looking at each individual benchmark result before picking a solution that best matches your needs.
Congratulations to Paul for his continuous string of System Builder Marathon victories, and thanks to all readers for the feedback that helped shape this quarter’s competition.