
System Builder Marathon, Q2 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 Budget Gaming PC
Day 2: Our Mainstream Enthusiast System
Day 3: The Balanced High-End Build
Day 4: Performance And Value, Dissected
Redefining The SBM Brackets
As you've seen over the past two days, Paul, Don, and I pored over feedback and made some changes to this quarter's System Builder Marathon, which we hope reflects more of what you'd like to see.
Perhaps the most frequently-requested update, readers asked that we narrow the entire range to make this a closer competition. So, rather than multiplying Paul's baseline ~$600 machine by 2x and 4x, which used to result in Don's $1200 system and my $2400 config, we dropped back to 2x and 3x ($1200 and $1800). That was complicated by another move: we added operating systems to our cost considerations.
And so the arguments between builders began. If the $600 PC only had $500 worth of hardware, shouldn't the price range for parts be $500, $1000, and $1500? Then, the original debate that $600, $1200, and $1800 budgets should include Windows turned into a compromise. We'd hold the hardware down at a lower target, and make a distinction between components that affect performance and everything else (like the chassis and optical drive).
This would supposedly give me the flexibility to splurge on a fancy case and Blu-ray drive without getting slaughtered in the value competition. But wait, doesn't a case directly affect the cooling system? Isn’t overclocking a vital part of our overall analysis?

As a case reviewer, I know that anyone leaning on a large radiator for liquid cooling has to shop for a chassis able to accommodate the extra equipment. I also know that cases with multiple fans typically cost more than similar enclosures without. So, I went back to the original formula: $1500 should buy everything except for Windows.
| Q2 2014 $1600 Performance PC Components | ||
|---|---|---|
| Processor | Intel Core i7-4770K: 3.5 GHz base (3.9 GHz max. Turbo Boost), Quad-core, 8 MB shared L3 cache | $320 |
| Graphics | PowerColor PCS+ AXR9 290X 4GBD5-PPDHE Radeon R9 290X | $529 |
| Motherboard | Asus Z97-A: LGA 1150, Intel Z97 Express | $150 |
| Memory | G.Skill Ripjaws X F3-14900CL8D-8GBXM: DDR3-1866 C8, 8 GB (2 x 4 GB) | $85 |
| System Drive | Samsung 840 EVO MZ-7TE250BW: 250 GB, SATA 6Gb/s SSD | $150 |
| Power | Rosewill HIVE-750: 750 W Semi-Modular, ATX12V v2.3, 80 PLUS Bronze | $90 |
| CPU Cooler | Thermaltake NiC L32 (CL-P002-AL14RE-A) 140 mm | $38 |
| Platform Cost | $1362 | |
| Storage Drive | WD Blue WD10EZEX: 1 TB, SATA 6Gb/s Hard Drive | $60 |
| Optical | Lite-On iHAS124-04: 24x DVD±R, 48x CD-R | $20 |
| Case | CM Storm Scout 2 Advanced | $90 |
| Total Hardware Cost | $1532 | |
| OS | Windows 8.1 X64 OEM | $100 |
| Complete System Price | $1632 | |
After compiling my list of parts, I was at $1472 without a hard drive or operating system, and ready to place my order. After all, my SSD was large enough that I didn’t need to split games and applications between two drives. I likewise would have preferred a better CPU cooler or Blu-ray drive, but any step up would have put me over $1500.
Eventually, I let the guys talk me into adding a 1 TB drive using the logic that only the parts affecting performance would go into the value calculation, and my mechanical disk drive wouldn't be included. In fact, I wouldn't even need to format my extra hard drive to complete our benchmark suite.
And so I stand alone, with a system that was supposed to cost $1472 of my $1500 hardware budget, and a nod to whoever wins this PC; it has an extra storage drive so you won't have to rely on a 250 GB SSD for all of your movies, music, and documents.
Rather than continuing to argue, I'm listing the complete platform price that they want to use in the value analysis, the total hardware cost that we previously used in our value analysis, and a complete system price (including the operating system). The best part is that I'm the one who gets to divvy-up prices for the various value charts for tomorrow's value shoot-out.
Conversely, as the “big-budget builder”, I’m also forced to make this quarter’s $1532 hardware compete with the previous quarter’s $2400 configuration. I do have one small ace in the hole, though. You may have noticed yesterday that Don's graphics card of choice went up in price, but comes with a free 250 GB Samsung 840 EVO. It just so happens that's the exact SSD I bought for this system back in May. If you don't mind paying $40 more for the PowerColor card in my build, for the next few days it also includes a bundled 250 GB 840 EVO. That $40 expense saves $150. You'd need to hurry though; PowerColor reps say the sale is only on through the end of June.

CPU: Intel Core i7-4770K
Based primarily on reader requests not to use another six-core CPU, last quarter’s high-end build still featured Intel’s revered Core i7-4770K overclocked to 4.50 GHz.
It was the fastest unlocked CPU in Intel’s LGA 1150 arsenal back when we placed our orders for this quarter, and nothing less than the best would approach that machine's compelling performance. Of course, in the time between then and now, Intel introduced the Core i7-4790K. But it's only supposed to hit availability today.

Read Customer Reviews of Intel's Core i7-4770K CPU
For now, we make do with the -4770K. Hyper-Threading will help in a few of our tests, and top-end processors tend to enjoy the best binning. But the conspiracy theorist in me is suspicious that maybe Intel was setting aside its best quad-core Haswell dies for Devil's Canyon way back when this Core i7 was manufactured (Ed.: Are you trying to foreshadow something, Thomas?).
Graphics Card: PowerColor PCS+ AXR9 290X 4GBD5-PPDHE
Last quarter, I struggled and failed to fit two GeForce GTX 780 Tis into my budget, settling instead for two vanilla 780s.
This quarter’s budget is a lot tighter. I don't even have the money for one 780 Ti, let alone two slightly cheaper 780s.

Read Customer Reviews of PowerColor PCS+ AXR9 290X 4GBD5-PPDHE Graphics Card
Given a choice between a single GeForce GTX 780 or a single Radeon R9 290X, most of us are going to go the AMD route. Selling for about $30 more than the noisy reference design on the day we placed our order, PowerColor’s PCS+ comes with a big quiet cooler and a small factory overclock. That kind of value is good enough to earn it an award.
Since we made our purchase, the cheapest Radeon R9 290X cards dropped by $20 while the PCS+ went up by $90 and back down to $570 with the 250 GB Samsung 840 EVO. According to PowerColor, this sale will end in days. So, we're using the original and eventual $530 price point for our calculations.
Memory: 8 GB G.Skill DDR3-1866 CAS 8
At least one of our benchmarks speeds up dramatically when we add more than 8 GB of RAM, but the price of a 16 GB kit would have a detrimental impact on value. Limited to a pair of 4 GB modules, we at least needed a good set.

Read Customer Reviews of G.Skill's 8 GB DDR3-1866 CAS 8 RAM
I’ve reviewed enough DRAM to notice that G.Skill uses the same ICs at various frequencies and latencies under a variety of part numbers. Maybe the company bins these differently? While searching for a set that I know would contain the “good stuff” (DDR3-1600 C8, DDR3-1866 C9, DDR3-2133 C10), I found a great deal on a kit that might have been binned a little higher: G.Skill's Ripjaws X F3-14900CL8D-8GBXM DDR3-1866 CAS 8.
Motherboard: Asus Z97-A
Asus is known for stability, for overclockability, for its good tuning software, and sometimes, for premium pricing. Its Z97-A was unable to out-overclock its competition in Intel Z97 Express: Five Enthusiast Motherboards, $120 To $160, and consequently lost the value war over a $10 price difference.
Read Customer Reviews of Asus' Z97-A Motherboard
That $10 price premium is gone now though, and we’re happy to pay a lower price for a top-tier overclocking platform.
CPU Cooler: Thermaltake NiC-L32
Cooling is an extremely important part of overclocking, and I’d need an equally extreme overclock to approach the performance levels of last quarter's $2400 machine. Unfortunately, I don't have the cash for a $100 dual-fan liquid cooler or even a $80 super-sized tower this time around. The good news is that I know a little about design and the limits of Intel’s Haswell-based CPUs.

Read Customer Reviews of Thermaltake's NiC-L32 CPU Cooler
Lacking the improved thermal interface material implemented on Intel's Devil's Canyon SKUs, our Core i7-4770K realizes relatively minor temperature drops even from big changes in cooling. Compounding the issue, small voltage increases can result in large temperature spikes. Many overclockers report a 1.30 V core limit is the key to CPU longevity. My experience shows that right-sizing the cooler for this voltage level means finding something as effective as the old MUX-120 I use in motherboard comparisons.
Thermaltake’s NiC-L32 hadn’t been reviewed when I placed my order, yet none of the heat sinks I knew would work fit in my budget. Lacking any other reference point, I relied on visual analysis to find a sub-$60 cooler that looked like it might serve my purpose.
PSU: Rosewill HIVE-750
I don’t often get time to read reviews elsewhere. But the moderators of Tom's Hardware's community do a great job of staying up to date. How else would I have learned that Newegg’s house brand, Rosewill, is a distributor of high-end Sirtec-made 750 W power supplies?

Read Customer Reviews of Rosewill Hive-750 PSU
This might not be one of the most talked-about PSUs out there, but anything close to the top is tempting when you're on a constrained budget. And that’s what this 80 PLUS Bronze-rated model is supposed to be, saving us $50 compared to the 80 PLUS Gold-rated unit we used last time.
Case: Cooler Master Storm Scout 2 Advanced
Cooler Master’s Storm Scout 2 Advanced received an honorable mention in our 11-way shoot-out as a top gaming case, coming up shy of the competition because it didn't have the eight expansion cut-outs needed for high-end graphics in a motherboard's bottom PCIe slot. Also, it wasn't selling at the lowest price given its performance level.

Read Customer Reviews of Cooler Master's Storm Scout 2 Advanced Case
The bottom PCIe slot of Asus' Z97-A isn't ideal for high-end graphics cards, so that didn't turn me off of Cooler Master's Storm Scout 2 Advanced. I instead noticed that the case was slightly less expensive than the winner of my round-up. Also, I remembered this enclosure being fairly sturdy. And I considered the practicality of a top-mounted carrying handle.
SSD: Samsung 840 EVO 250 GB
Samsung’s EVO drives earned our value award a year ago, and the 250 GB model continues to pop up in our Best SSDs For The Money column. Who am I to argue?

Read Customer Reviews of Samsung's 840 EVO 250 GB SSD
Besides being fast and cheap, the 250 GB 840 EVO is also the perfect capacity SSD for our test suite, which includes tens of gigabytes of work files and several large game installations.
Hard Drive: Western Digital Blue WD10EZEX
This is where we get into the parts that we don't need in the lab, but you probably will once you get this machine home. Western Digital’s 1 TB Blue drive offers a bunch of capacity to store your photos and media without busting my budget.

Read Customer Reviews of Western Digital's Blue WD10EZEX
The WD10EZEX has a 7200 RPM spindle to improve response time and a SATA 6Gb/s interface for speedy accesses to 64 MB of data cache. But really, I only bought this disk to store information that doesn't need to live on the precious solid-state storage.
OS: Windows 8.1
One of the first questions our contest winners ask is whether we give them software to go along with their new PCs. Up until now, we haven't taken operating system cost into account. But at your request, this expense is factored in for the first time.

Read Customer Reviews of Microsoft's Windows 8.1 Operating System
New systems qualify for the OEM version of Windows 8.1, which costs $100. And we can now put to rest those complaints from readers who were trying to compare pre-built systems, which typically come bundled with operating systems, to our builds.
Optical Drive: Lite-On iHAS124
Many readers claim that they no longer want or need an optical drive, but our operating system shipped to us on optical media. Besides, I'm willing to bet that you have something precious stashed away on a DVD or CD that you’d like to read someday.

Read Customer Reviews of Lite-On's iHAS124 ODD
The iHAS124 also burns DVD media at 24x, just in case you have a few blank discs laying around. Plus, it's great for loading an unrecognized GbE controller driver...since you can't exactly download that until the network connection is up and running.
Thermaltake’s NiC-L32 CPU cooler uses two large mounting rings in a sandwich design. It's easiest to install with the motherboard removed from the case.

Foam is supposed to prevent short-circuits on the base plate, but the legs of some capacitors appear to poke all the way through. Just to prove that they could be shortened without special tools, I trimmed the longer leads with a pair of fingernail clippers.
The screws seen in the above photo engage top-side standoffs in the photo below. Each standoff has a foam ring on one side to protect the surface of the motherboard. Screws then attach the top-side mounting ring to the standoffs.

A large dab of paste in the middle makes a large circle about the size of the CPU heat speader. I like to add a smear next to each corner to maximize contact area.
Also notice the orientation of the top mounting ring. The two threaded nubs act as spacers to prevent over-tightening of the next piece in the installation puzzle.

The NiC-L32’s heat sink is secured with a cross-brace, which uses two shoulder screws to connect to the previously-described top ring. A diagram is provided in the cooler’s installation instructions.
Cooler Master's Storm Scout 2 Advanced uses push-in drive rails to secure 3.5” hard drives, and includes two 2.5” adapter trays that employ the same rails.

Clips hold the fan to the CPU cooler, and a set of nine screws and standoffs hold the motherboard inside the case. The graphics card slides into its slot after removing a couple of slot covers. The drives slide into their brackets. And all of the power supply’s wires are attached to finish the installation. When in doubt, refer to the case and motherboard manual for further instructions.

Do you see anything missing in the shot above? The case originally supported up to seven hard drives, but PowerColor’s PCS+ AXR9 290X graphics card wouldn’t fit with the upper cage assembled. Attached with five screws, the rear wall of the upper cage is removable.
The PCIe power cables look a little messy because they're too short to run behind the motherboard tray. Also, a second CPU power cable (4+4-pin divisible) is permanently attached to our semi-modular supply, which doesn't make sense. Most motherboards only have one auxiliary CPU power header, and semi-modular power supplies are supposed to permanently attach only the universally-required cables. I stashed the spare cable between the motherboard tray and right side panel.

Given what I know now, I probably wouldn’t have picked this case or power supply. Then again, both are priced aggressively, so I might have simply second-guessed myself before buying the same components. At least the finished product is solid and looks great.
Asus offers two modes for overclocking. “XMP” mode starts off with an Intel XMP memory profile. It’s a great way to get your memory running at its rated settings, while simultaneously enabling a wide range of additional overclocking options.

Here's where things take a turn for the worse. My retail Core i7-4770K wouldn't run stably at 4.3 GHz, no matter how much voltage I applied to it. Over time, I determined that the first core was crashing, so I couldn't even set a graduated overclock above 4.2 GHz unless I also wanted to set core affinity for every program. Yes, I know that professional overclockers might be tempted to put in the time, but real-world computing is more practical. I stuck with 4.2 GHz and licked my wounds.
G.Skill’s DDR3-1866 CAS 8 kit overclocked to DDR3-2133 CAS 9, as predicted. I had to manually configure 9-10-10-27 timings, and was faced with a boot failure when I pushed further to 9-10-9-27.

A core clock rate of 4.2 GHz was accessible with a 1.28 V core setting. But 4.3 GHz remained out of reach, even at 1.3 V (I tried higher settings, too, before giving up). Offset voltage mode allows the board to idle down to a lower voltage, and I achieved a maximum 1.288 V core using the 0.025 V offset.

MSI's Afterburner software provides a handy fan profile page under the Settings menu, in addition to power limit and frequency modification. Our Hawaii GPU didn't overclock well, but the Radeon R9 290X's memory scaled all the way to GDDR5-6200.
The following tables include the stock and overclocked settings for this quarter's enthusiast-class gaming build, followed by the configuration we put together earlier this year, which serves as our comparison point.
At the very bottom, you’ll find the programs and games used for benchmarking.
| Test Hardware Configurations | |||
|---|---|---|---|
| Q2 $1600 Performance PC | Q1 $2400 Performance PC | Q1 $1600 Enthusiast PC | |
| Processor (Overclock) | Intel Core i7-4770K: 3.5 to 3.90 GHz, Four Physical Cores O/C to 4.20 GHz, 1.29 V | Intel Core i7-4770K: 3.5 to 3.90 GHz, Four Physical Cores O/C to 4.50 GHz, 1.25 V | Intel Core i7-4770K: 3.5 to 3.90 GHz, Four Physical Cores O/C to 4.40 GHz, 1.20 V |
| Graphics (Overclock) | Radeon R9 290X: 1050 MHz GPU, GDDR5-5400 O/C to 1100 MHz, GDDR5-6200 | 2x GeForce GTX 780: 902 MHz GPU, GDDR5-6008 O/C to 1059 MHz, GDDR5-6720 | GeForce GTX 780: 928 MHz GPU, GDDR5-7000 O/C to 1128 MHz, GDDR5-7300 |
| Memory (Overclock) | 8 GB G.Skill DDR3-1866 CAS 8-9-9-24, O/C to DDR3-2133 CL 9-10-10-27, 1.60 V | 16 GB G.Skill DDR3-1866 CAS 9-10-9-28, O/C to DDR3-2133 CL 9-10-11-10, 1.585 V | 8 GB Corsair DDR3-1866 CAS 9-10-9-27, No O/C |
| Motherboard (Overclock) | Asus Z97-A: LGA 1150, Intel Z97 Express, Stock 100 MHz BCLK | ASRock Z87 Extreme4: LGA 1150, Intel Z87 Express, Stock 100 MHz BCLK | ASRock Z87 Pro3: LGA 1150, Intel Z87 Express, Stock 100 MHz BCLK |
| Case | CM Storm Scout 2 Advanced | NZXT Phantom 410 | NZXT Phantom 410 |
| CPU Cooler | Thermaltake NiC L32 | Thermaltake CLW0217 Water 2.0 Extreme | Cooler Master Hyper 212 EVO |
| Hard Drive | Samsung MZ-7TE250BW 250 GB SATA 6Gb/s SSD | SanDisk SDSSDHP-256G-G2 256 GB SATA 6Gb/s SSD | Samsung MZ-7TE120BW 120 GB, SATA 6Gb/s SSD |
| Power | Rosewill HIVE-750: 750 W Semi-Modular, 80 PLUS Bronze | Corsair HX750: 750 W Semi-Modular, 80 PLUS Gold | Corsair TX650: 650 W, 80 PLUS Bronze |
| Software | |||
| OS | Microsoft Windows 8 Pro x64 | Microsoft Windows 8 Pro x64 | Microsoft Windows 8 Pro x64 |
| Graphics | AMD Catalyst 14.4 | Nvidia GeForce 335.23 | Nvidia GeForce 334.89 |
| Chipset | Intel INF 9.4.0.1026 | Intel INF 9.4.0.1026 | Intel INF 9.4.0.1026 |
Hope for any level of competitive performance between today’s $1600 build and the $2400 machine it replaces is dashed by a CPU overclock that doesn’t even approach mediocrity. Even the previous $1600 machine outclasses it in that arena, though the new build at least has some memory overclocking to push it through a few benchmarks.
| Benchmark Configuration | |
|---|---|
| 3D Games | |
| Arma 3 | V. 1.20 Current PC, V.1.08 Q1 PC 30-sec. Fraps "Infantry Showcase" Test Set 1: Standard Preset, No AA, Standard AF Test Set 2: Ultra Preset, 8x FSAA, Ultra AF |
| 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 |
| Far Cry 3 | V. 1.05, 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 AMD/ HBAO NVidia |
| 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 |
| Audio/Video Encoding | |
| HandBrake CLI | Version: 0.99, Video: Video from Canon EOS 7D (1920x1080, 25 frames) 1 Minutes 22 Seconds, Audio: PCM-S16, 48,000 Hz, Two-channel, to Video: AVC1 Audio: AAC (High Profile) |
| 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) |
| TotalCode Studio 2.5 | Version: 2.5.0.10677, MPEG2 to H.264, MainConcept H.264/AVC Codec, 28 sec HDTV 1920x1080 (MPEG-2), Audio: MPEG-2 (44.1 kHz, Two-channel, 16-bit, 224 Kb/s) Codec: H.264 Pro, Mode: PAL 50i (25 FPS), Profile: H.264 BD HDMV |
| Adobe Creative Suite | |
| Adobe After Effects CC | Version 12.0.0.404: Create Video, 3 Streams, 210 Frames, Render Multiple Frames Simultaneously |
| Adobe Photoshop CC | Version 14 x64: Filter 15.7 MB TIF Image: Radial Blur, Shape Blur, Median, Polar Coordinates |
| Adobe Premiere Pro CS6 | Version 6.0.0.0, 6.61 GB MXF Project to H.264 to H.264 Blu-ray, Output 1920x1080, Maximum Quality |
| Adobe Acrobat XI | Version 11.0.0: Print PDF from 115 Page PowerPoint, 128-bit RC4 Encyption |
| Productivity | |
| ABBYY FineReader | Version 10.0.102.95: Read PDF save to Doc, Source: Political Economy (J. Broadhurst 1842) 111 Pages |
| 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 | Version 10.0, Compile Google Chrome, Scripted |
| Compression | |
| 7-Zip | Version 9.30 alpha, LZMA2, Syntax "a -t7z -r -m0=LZMA2 -mx=5" Benchmark: THG-Workload-2012 (1.3 GB) |
| WinRAR | Version 5.0, RAR, Syntax "winrar a -r -m3" Benchmark: THG-Workload-2012 (1.3 GB) |
| WinZip | Version 18.0 Pro, Syntax "-a -ez -p -r" Benchmark: THG-Workload-2012 (1.3 GB) |
| Synthetic Benchmarks and Settings | |
| 3DMark Professional | Version: 1.2.250.0, Fire Strike Standard and Extreme |
| PCMark 8 | Version: 1.0.0 x64 Full Test |
| SiSoftware Sandra 2014 | Version: 2014.02.20.10, Processor Arithmetic, Cryptography, Memory Bandwidth Benchmarks |
The new build faces sound defeat in 3DMark Professional, though the margin between it and the previous $1600 build is small enough to leave a little hope for real-world gaming analysis. This quarter's configuration looks a little better in PCMark 8's Work suite, though its storage scores are merely par. Then again, that's significantly better than what Don achieved yesterday using a single mechanical hard drive.

It's hard to guess why the Creative and Home scores from Don’s previous $1600 machine are so high, but I'm not worrying about them, since they don't count towards our final performance comparison. I just wish he’d tell me his secrets.

Sandra Arithmetic reminds me that this quarter's high-end build overclocks dismally; it's solidly beaten by the previous $1600 PC, even. And, in comparison, my older $2400 machine achieves a monstrous overclock.
The only silver lining is that my $1600 build fares better in Sandra's Memory Bandwidth module than the last PC we tested at the same price point. In turn, my Cryptography results are stronger. Don didn't get very aggressive with his memory overclock, and his system's performance reflects that.


Again, Sandra's Memory Bandwidth benchmark shows how Don's $1600 PC falls behind. He used the kit's SPD timings to represent stock performance, and simply dialed in XMP settings, calling that overclocked. Meanwhile, my system benefits from XMP in its stock configuration (this is as simple as flipping a BIOS switch, after all) and a true memory tune-up when the rest of the system is tweaked.

Battlefield 4 employs the same 200-FPS limit as its predecessor. That performance ceiling diminishes the value of ultra-high-end graphics configurations, since they slam into it and cease benefiting from powerful GPUs. Then again, DICE correctly suggests frame rates this high do little for your actual experience. Instead, it's better to focus our analysis on the most relevant resolutions.
My newest build slightly outpaces Don's $1600 effort from last quarter. The older $2400 PC, which I armed with two GeForce GTX 780s, requires more taxing quality settings to show off the worth of its hardware.

The $2400 PC I built last quarter starts to shine as I switch on the Ultra quality preset in Battlefield 4. Both $1600 PCs are roughly matched, even though they employ dissimilar graphics architectures.

The fun part of benchmarking Grid 2 is trying to figure out where graphics bottlenecks stop and the memory bandwidth limitations kick in. The built-in metric can even be CPU-constrained if all of your other subsystems are ample. By that point, though, you're looking at an academic exercise; performance is typically plenty-smooth. It’s because this game is so forgiving that I’m personally so shocked when I test a system that doesn’t excel.

Grid 2 is the second title (out of two) that requires an Ultra quality preset to demonstrate the value of two very high-end graphics cards working cooperatively. As with the previous title, Grid 2 also creates approximate performance parity between the two $1600 builds.

Arma 3 is the first benchmark in today's comparison that shows a strong preference for AMD's graphics architecture. Paul's extensive Day 1 analysis suggested that this sequence is strongly affected by CPU performance. However, I'm using a much faster chip than he was.
And so we see Arma 3 give my new $1600 build, with its Radeon R9 290X, a lead over Don's comparably-priced setup from a quarter ago. Both boxes employ Core i7-4770Ks, so host processing isn't a variable.

Even using the Ultra preset, Arma 3 appears platform-bound at resolutions up to 1920x1080. Triple-monitor resolutions make it easier for the $2400 machine’s SLI configuration to break performance barriers encountered by both $1600 PCs, and the new machine’s AMD-weighted advantage goes away.

I'm going to go out on a limb and call Don's Far Cry 3 results suspect; a single GeForce GTX 780 shouldn't beat two, and yet his $1600 configuration outperformed my $2400 collection of hardware last quarter.
That oddity also punishes my $1600 build this time around. It takes triple-monitor resolutions before the frame rates hit a point where they're expectedly similar.

Setting aside the unbelievable single-monitor results posted by Don's $1600 build last quarter, his GeForce GTX 780 also edges out my top-end system's Radeon R9 290X in the more credible 4800x900 and 5760x1080 tests.

The Apple iTunes and LAME MP3 encoding workloads depend almost exclusively on clock rate. Yes, IPC throughput matters when we compare dissimilar architectures. But all three of these machines employ the same Haswell design.
As a result, Don's slightly misconfigured $1600 PC falls slightly behind at its stock frequency, and outpaces my effort this quarter when both boxes are overclocked, since my Core i7 doesn't scale well.


Copy what I just said about audio transcoding to video, for the most part, since these threaded applications are running on three identical CPUs.
There is a minor anomaly where my weak overclock this quarter outperforms the higher clock rate Don achieved last quarter, but only by one second. More than likely, the outcome is affected by his less-thoroughly-tuned memory subsystem.


Adobe After Effects is the one application we test that benefits noticeably from platforms with more than 8 GB of RAM. Access to a $2400 budget last quarter allowed me to splurge on 16 GB, and the results look almost artificially high. We've seen this behavior before though, and know it accurately represents our custom workload.


Using our CPU-based filters, Adobe Photoshop shows all three Core i7-4770K delivering identical performance. Overclocked, the platforms scale proportionally to our success tuning each machine. Unfortunately, that means my meager overclock this quarter lands in last place.

Acrobat is single-threaded, and it responds to clock rate in a comparison of platforms based on the same architecture. Don's $1600 configuration from last quarter is the exception at its stock settings. That might be attributable to poorly-configured memory.

This quarter's high-end $1600 PC establishes a tiny lead in 3ds Max among the three stock configurations, but loses that lead to my previous $2400 effort as a result of a weak overclock.
Don's $1600 configuration from last quarter demonstrates dramatically worse behavior. We'd need to chalk that up to an unoptimized memory subsystem, since we're looking at the same CPU across a trio of platforms.


Blender yields the exact performance scaling we’d expect from three systems equipped with the same CPU, but different clock rates.

FineReader and Visual Studio may be showing signs of a memory bandwidth bottleneck based on the poor performance of Don's $1600 setup from last quarter. It sported a better overclock, and nothing else could explain the large deficit incurred by its stock configuration.

When we compare this quarter's $1600 machine to my old $2400 PC using 7-Zip, it appears that the utility shows slight favor to the pricier box's higher memory capacity. Then, when we compare my high-end build to Don's previous effort (both of which are $1600 systems), the same piece of software looks like it's punishing lower memory bandwidth from Don's older platform.


WinZip’s various approaches give us a variety of data points to digest. For example, the “EZ” switch, which triggers maximum compression, prefers the faster memory configurations of the new $1600 PC and the old $2400 machine, while its “CPU” switch gives us performance that scales with frequency differences.

I configure my builds to maximize low-load energy savings. The same doesn't look like it applies to Don's approach. Or, maybe his motherboard didn’t idle down properly even with those settings enabled. Either way, the new $1600 build appears to draw far more energy under load, and even the 80 PLUS Gold-rated power supply in last quarter's $2400 machine couldn’t promote power-savings with GeForce GTX 780s rendering away.

And now for the shocker: my new $1600 configuration doesn't run cool enough. But should we blame its cooler?

The above chart doesn’t tell the whole heat story, since my measurements were taken with both the GPU and CPU under duress. First, a CPU-only load would have dropped CPU temperature by up to eight degrees. Second, voltage levels approaching 1.30 V are extremely difficult to cool Haswell-based CPUs predating Devil's Canyon. Indeed, Thermaltake's cooler produced similar temperatures as the old MUX 120 that I use in motherboard round-ups to push the same Core i7-4770K up to 4.7 GHz. This truly was a bum processor for overclocking.
I could have spent twice as much money on CPU cooling to save a few degrees, and might have passed on a pass/fail system. But if I really wanted to be sure the platform would stay under its thermal threshold, I would have also needed more case cooling or an externally-venting graphics card cooler (you know, the noisy blower-style reference cooler from AMD that everyone hates). To that end, perhaps I could have used a GeForce GTX 780 rather than a Radeon R9 290X, since it appears that only Nvidia knows how to make a well-behaved blower-style cooler.

This quarter's $1600 build performs 3.4% better than Don's $1600 machine from last quarter at its stock settings, but falls a little more than 1% behind when both of us overclock. Again, it's a bummer that my CPU wouldn't stabilize above 4.2 GHz.
It’s also less efficient, and I can probably point to the graphics card’s high power consumption for most of that. The new CPU did need a bunch of extra voltage to run at its meager overclock though, which further hurt the new machine’s overclocked efficiency.
The highest-end machines I'm tasked with building in our System Builder Marathon always get pushed well past the point of diminishing returns, where a relatively large increase in component price produces a fairly small gain in performance. That means I expected today’s $1600 machine to beat my previous $2400 build on the value front. But how does it fare compared to our previous $1600 PC?

The previous page showed a combined performance score, where this quarter's System Builder Marathon configuration was 3.4% faster at its stock clock rates, but 1% slower when overclocked, compared to Don's $1600 build from last quarter. But it’s also technically cheaper, since this $1600 machine was built on a $1500 budget; the extra $100 went towards its OS.

And yes, that means I win the value fight. Slightly lower hardware costs give my PC a 9% lead over last quarter's $1600 effort. And rather than losing once both builders overclock their hardware, my configuration simply drops that value victory to a 5% margin.

A lower budget this month compelled me into a more game-centric focus on the highest-priced build, which also explains the LAN party-style case and my original choice to build it at $1472 without a hard drive. But even at $1532 (including an unbenchmarked storage disk), it proves itself the better gamer. My lower-cost alternative produced 10%-better performance than the $1600 PC that preceded it. And it still has a 2% performance margin when overclocked, despite the minimal clock gain achieved.
This is no apples-to-apples comparison though. The retail CPU I ended up with really is sub-par. If I had the chance to borrow last quarter's more average Core i7-4770K, I'd really get the chance to show off how well the rest of my machine could work. Hopefully, if you follow in my footsteps, you'll have better luck. Or, simply replace my -4770K with a -4790K, since they're in stock at Newegg now.
There's one other option for adding value this month: at least for the next few days, PowerColor's PCS+ AXR9 290X's sells at a slightly higher price, but includes the 250 GB Samsung 840 EVO I bought for $150. If that deal sounds good to you, jump on it fast. Beyond June, PowerColor reps say the card will drop to $530-ish without the SSD.


