For our 8th Quiet Gaming PC Build Guide, we take on the challenge of two high-end video cards in an SLI configuration featuring a pair of Zotac GTX 970s in the SilverStone Fortress FT05 case.
March 5, 2015 by Lawrence Lee
Since we began posting gaming build guides last fall, we’ve only used single video card configurations, and for good reason. For most gamers one high-end GPU has enough horsepower to render even the most demanding games smoothly at the most common native monitor resolutions of 1920×1080 and 1920×1200. However, there is a vocal subset of enthusiasts who demand more, in particular those with QHD/4K monitors or multiple displays with up to four times as many pixels.
To game at higher than HD resolutions, two or more cards in CrossFire or SLI is practically a necessity. In the past, it was a forgone conclusion that even a dual GPU system would be hopelessly loud without an elaborate cooling solution. While heavy duty aftermarket heatsinks are available and liquid GPU cooling is starting to become more popular, neither with current GPU technology. The GeForce GTX 900 series has substantially lower power consumption than Nvidia’s previous generation of GPUs as well as AMD’s Radeon counterparts, making it much easier to cool with stock cooling solutions.
This article, the 8th in our Quiet Gaming PC Build Guide series, details the component selection, assembly and fine tuning process of creating a very quiet gaming system based around two GTX 970 graphics cards.
COMPONENT SELECTION
GPU: Zotac GeForce GTX 970 AMP! Extreme Core Edition x 2 – US$380 each
Alternatives:
- Asus Strix GTX 980 – US$565
- MSI GTX 980 Gaming – US$560
- Asus Strix GTX 970 – US$330
- MSI GTX 970 4GD5T OC – US$330
Generally there are two types of SLI/CrossFire systems, the über rig that uses the fastest GPUs available, and the more value-oriented machine with slower cards that combine to match/outperform single higher model cards. Our build is the later, utilizing a pair of GeForce GTX 970’s from Zotac. The GTX 980 is the single GPU king but it holds only a 10~15% performance advantage over the 970, while carrying a more than 60% cost premium. Benchmarks conducted by various tech sites place a pair of 970s well ahead of a single 980 and not far behind dual 980s.
However, the GTX 970’s contentious memory allocation issue can’t go unmentioned. Without getting into the nitty-gritty details, the way the new Maxwell architecture works negatively affects the GTX 970’s memory subsystem. While each and every 970 physically has 4GB of onboard memory, the last 0.5GB can only be accessed at a very low speed compared to the first 3.5GB. When 3.5GB of memory usage is exceeded, some games can stutter or even freeze up.
This is not an insignificant problem as high memory usage goes hand-in-hand with gaming at higher resolutions, the resolutions that make an SLI configuration desirable in the first place, and unfortunately VRAM doesn’t stack when using SLI/CrossFire. The gaming community is somewhat divided on the severity of the issue as no one thus far has conducted extensive testing with more than a handful of games to explore the scope of the problem. If this is enough to put you off on 970s, this article still has value as a GTX 980 SLI build guide; the TDP difference between the two cards is only ~20W.
|
|
Zotac has six variants of the GTX 970, the smallest of which was used in a couple of our gaming builds already, but its paltry stock cooling solution forced us to resort to third party solutions. This time Zotac provided us with a pair of their top-of-the-line GTX 970 AMP! Extreme Core Edition cards. This is one of the pricier variants on the market, but by the looks of things, it’s worth the premium. It has a substantial triple fan heatsink and is clocked aggressively at 1228/1380 MHz (base/boost), higher than most competing models, and a significant bump compared to the stock 1050/1250 MHz.
|
|
The cooler bears some resemblance to Arctic’s Accelero Xtreme series as it utilizes three fans, copper heatpipes running along with length of the card, and perpendicularly arrayed aluminum fins. This dual slot card is quite long, measuring 30.2 cm (almost 12 inches) across but the most notable physical aspect about this model is its standard slot width. Many variants of the GTX 970 use larger boards or cooling solutions that extend over the edge of the PCB, making them much wider, causing interference issues with some cases.
|
|
To keep the PCB from bending or bowing from the weight of the heatsink, the card has some armor on the back side in the form of a thick metal plate with ventilation holes.
Case: SilverStone Fortress FT05 – US$185
Alternatives:
- SilverStone Fortress FT02 – US$230
- SilverStone Raven RV03 –
US$140 - SilverStone Raven RV05 –
US$125
SilverStone’s Raven/Fortress series of towers have consistently been superior performers for us, making their latest model, the Fortress FT05, an easy choice for housing this build. The second, third, and fifth generation versions are particularly effective, especially for more demanding systems. The combination of massive intake fans and a rotated motherboard tray has been a proven winner in our lab.
A tower case with a standard layout faces a significant challenge when utilizing an SLI/CrossFire setup. The card positioned higher up in the case is subject to extra heat rising up from the card below. The results from our load tests with tradition tower designs have been consistent— the upper card always requires a much higher fan speed than the lower card, usually in the 400~500 RPM range, to deal with this added thermal load. The FT05 design however, orients the cards vertically, so each GPU doesn’t affect the other nearly as much. The fans can thus can spin slower, and the machine can operate at a lower overall noise level.
|
|
This is the first I’ve worked with the FT05, but it should produce the same results as the Raven RV05 as they are essentially identical cases aside from the exterior. While the Raven line has molded angled plastic on the outside, the upscale Fortress version has a blocky, monolithic design with a wraparound silver aluminum exterior and a highly reflective accent near the bottom. The only plastic visible is the removable top cover where all the heat exhausts out. Other features include a slim, slot loading optical drive bay on the right side in order to maintain a solid front facia, and a large gap at the bottom of the case to feed the two 18 cm intake fans sitting inside on the case floor. Inside the gap is a removable dust filter magnetically attached underneath the fans.
|
|
The power and reset buttons are located at the top, sandwiching a hidden compartment for the front USB 3.0 and audio ports and the three-speed fan control switches for the stock fans. As the motherboard tray is rotated, the back panel and expansion slot ports are all located at the top. The case is taller than it functionally needs to be in order to accommodate and hide all the cabling that runs in through the back.
|
|
One nice thing about this design is the side panels simply drop into place without having to worry about aligning any tabs or slots. Acoustic dampening foam is featured heavily in the FT05, both on the side panels and at various spots inside the chassis.
|
|
Rather than front to back airflow, the Raven/Fortress design mandates bottom to top airflow, working with the natural path of thermal convection. The two 18 cm fans blow up over the entire interior, covering both the CPU and GPU area. Unlike most enthusiasts ATX cases, drive support is limited because the chassis is actually rather compact with a total volume of only 46 Liters. There’s just a small plastic cage at the bottom/rear for a pair of 3.5-inch drives, while two 2.5-inch models can be placed behind the motherboard tray. In reality, the lack of more extensive storage support is hardly significant for a gaming build, especially in this day and age when there are so many good external storage options.
|
|
Both the optical drive tray and hard drive cage are removable, but you may want to leave them in place even if they’re not being used, if only to better hide cable clutter. The motherboard tray is rather small with few available points for tying down or covering cabling.
COMPONENT SELECTION (Con’t)
CPU: Intel Core i5-4690K – US$225
Alternatives:
- Intel Core i7-4790K – US$330
- Intel Core i5-4590
– US$195
|
|
Intel’s superior energy efficiency has us coming back to their Haswell processors time and time again. Quad core LGA1150 chips also deliver excellent all-around performance, both in single and multi-threaded applications/games. An ideal choice is the Core i5-4690K, a quad core chip running at 3.4 GHz (up
to 3.8 GHz with Turbo Boost). Only the most demanding games at very high resolutions can come close to bottlenecking this processor and even if this happens, it has an unlocked multiplier enabling easy overclocking
to relieve any such limitations. A more expensive Core i7 probably isn’t warranted but given the budget involved with this PC, it certainly wouldn’t be out of place.
CPU Cooler: Scythe Kotetsu
– US$40
Alternatives:
- Scythe Mugen MAX – US$50
- Scythe Mugen 4 – US$50
- Be Quiet! Shadow Rock Slim
– US$50
- Coolermaster Hyper 212 Evo – US$30
|
|
Haswell processors don’t run particularly hot but the Intel stock cooler is
woefully insufficient for a silent PC. One of SPCR’s current favorite
value cooling solutions is the Scythe
Kotetsu, which delivers excellent performance at a low price. This modest
tower cooler is equipped with a single 120 mm fan with an overall pleasant/smooth
sound. The FT05’s cables all emanate near the top/rear corner of the motherboard so you may want to avoid larger heatsinks that are difficult to install after the board is secured in the case.
Many gamers regard an AIO water cooling unit as a defacto choice for a high-end
build but I have to disagree as they are neither cost or noise efficient. They
also usually ship with acoustically poor fans, and the buzzing and gurgling
sounds from the pump raises the noise floor, making it difficult to get the
noise level down when the system is idle.
Motherboard: Gigabyte GA-Z87X-UD5 TH – US$230
Alternatives:
- Asus Z97-PRO
– US$190 - Asus Sabertooth Z97 Mark 2 – US$160
- Asus Z97-A – US$140
- Gigabyte GA-Z97X-SLI – US$115
|
|
The Gigabyte Z87X-UD5 TH is far from my first choice, but it is the only suitable model on hand for this build. A premier SKU from the previous generation, it’s no longer being carried by most retailers and not worth buying at this point. There’s nothing wrong with the Z87 chipset per se as all it’s really missing is SATA Express (which some Z97 models dispense with anyway) and support for the upcoming generation of LGA1150 CPUs. However, unless there’s a significant discount involved, you’re better off with one of the many Z97 options available on the market.
|
|
This model has an ample number of controllable fan headers but only three are needed, two for the case fans and one for the CPU fan (the GPU fans have fan control built-in). More importantly, it has two full-length PCI-E 3.0 slots with x8 bandwidth each and there are two slots in between them, creating ample separation between the two video cards. An ATX board is not a necessity either as some microATX models satisfy these requirements.
Power Supply: Be Quiet! Straight Power 10 600W – US$130
Alternatives:
- Seasonic X-650 – US$115
- Corsair RM650 – US$110
- Corsair CS650M – US$90
Today’s power supplies are highly efficient and most models don’t produce much noise. This is especially true when they’re placed on the opposite side of the CPU area and given direct access to cool intake air from the outside like in the FT05. From a thermal standpoint, it’s essentially isolated from the rest of the components and only has to cool itself. While a passively cooled unit would be ideal for many quiet systems, this would produce extra heat compared to an actively cooled model. The GTX 970 isn’t exactly a power hungry beast but two of them will produce a fairly substantial thermal load, so it would be silly to put even more strain on the cooling situation.
|
|
Be Quiet! makes some of the quietest power supplies on the market and
the Straight Power 10 is a fine example, remaining practically silent for much
of its lower range, just like the Dark
Power Pro 550W we reviewed awhile back. Be Quiet! products are
sold by NCIX
in the US and Canada. If you need an alternative, we generally recommend models
manufactured by Seasonic (which includes some SKUs sold by Corsair, Antec,
XFX, and others). We can attest to the quietness of the Seasonic
X-series, and according to techPowerUp!,
the Corsair CS650M and RM650 are fairly quiet at loads between 300~400W.
|
|
This build will require only about 350W DC, so the Straight Power 10 600W offers more than enough capacity. It’s 80 Plus Gold
certified for high efficiency; Compared to a Bronze unit, it will use a bit
less power and run slightly cooler, but this may not be enough to actually lower
the noise it produces in any meaningful way. The claimed baseline noise specification
is 10.7 dBA@1m, which means it won’t register on our sound level meter in our
10~11 dBA ambient chamber. It also has modular cables which means less clutter inside the case.
SSD: Samsung 850 Pro 256GB
– US$170
Alternatives:
- Samsung 840 EVO 1TB
– US$430
- Samsung 850 Pro 512GB
– US$285
- Crucial MX100 512GB
- Crucial MX100 256GB – US$100
|
|
Solid-state storage may be the most significant advance in the last decade
for silent computing. With no moving parts, they generate zero noise, but also
have ridiculously low latency, resulting in fast loading times. As games continue
to grow in complexity, having an SSD becomes increasingly advantageous. A 240~256GB
model should be considered a starting point; with Windows and a few triple-A
gaming titles installed, a smaller drive could be filled close to capacity.
Also, with ever increasing memory density, fewer dies are needed, which can
mean fewer read/write channels being used and slower performance for SSDs of lower capacity.
This time around, the Samsung 850 Pro, the fastest SSD we’ve ever tested, gets put into action. Is it worth extra cost compared to a more mainstream model like the Crucial MX100? Probably not, but upgrading to a premium solid-state solution is a drop in bucket compared to a pair of GTX 970s, at least when we’re talking about sub-300GB capacity models. A larger variant is probably warranted depending on the size of your gaming library and a mechanical hard drive is a pragmatic supplement for storing other mass media.
RAM: Kingston HyperX Genesis 8GB (2x4GB) DDR3-1866 – US$95
Alternatives:
- Crucial Ballistix Tactical 8GB 8GB (2x4GB) DDR3-1866 – US$70
- Kingston HyperX Savage 8GB (2x4GB) DDR3-2133 – US$75
|
|
Precisely what RAM is used as system memory is not critical, although other
web sites have identified DDR3-1600 to DDR3-1833 as the sweet spot, somewhat dependent
on the particular game. Within this clock speed range, small variations in timing
have minuscule effect on overall performance. 8GB is more than sufficient for
any single game and general purpose multitasking. 16GB is a waste of money unless
you have a specific need for it, and RAM is one of the easiest things to upgrade later in a system, if you really need it for some new application. We recommend choosing a brand with
a good lifetime warranty and to avoid models with overly large heatspreaders
as they can interfere with larger CPU coolers. Kingston HyperX RAM has been
solid for us, and it sports lower profile heatspreaders.
ASSEMBLY
Assembling a system inside the FT05 is fairly straightforward, but it’s easier to get the components in place if the case is set on its back or side. This is particularly true for positioning the power supply as you won’t have to fight gravity as much as the screws securing it go in from the top. It’s also a good way to avoid dropping screws into the fans below.
|
|
As there’s only a small gap between edge of the video cards and the intake fans, it actually looks somewhat cramped for an enthusiast case, even with the hard drive cage removed.
|
|
The 18 cm Air Penetrator fans have been cooling SilverStone cases for years. They occupy most of the case floor but they’ve left some room around them to snake some cabling. The fans are powered via 3-pin connectors that plug into the motherboard, but they also have input headers that connect to the built-in fan controller. This allows us to slow the fans two different ways.
|
|
On the other side of the case, it’s a bit messy as there are few points for tying down cables; I ended up using the empty 2.5-inch bay to pin down some of the wiring. A SATA power cable is required to run the power LED which lights up the SilverStone logo at the front.
|
|
SilverStone doesn’t leave much room behind the motherboard tray so try not to bundle anything with the thicker 24-pin and 8-pin power supply cables.
|
|
The side panel doesn’t actually sit flush with the rest of the chassis. The panels simply tuck inside this “lip” making it easier to align than typical panels, especially if the innards are threatening to burst out.
TESTING
System Configuration:
- Intel Core i5-4690K processor – 3.4 GHz (up to 3.8 GHz with Turbo Boost), 22nm, 84W, integrated HD 4600 graphics
- Scythe Kotetsu CPU
cooler - Gigabyte Z87X-UD5 TH
motherboard – Intel Z87 chipset, ATX - Zotac GeForce GTX 970 AMP! Extreme Core Edition graphics card x 2 – 1664 CUDA cores, 1228 MHz clock (up to 1380 MHz with GPU Boost), 7010 MHz memory
- Kingston HyperX Genesis memory – 2x4GB, DDR3-1866, C10
- Samsung 850 Pro solid-state drive – 256GB, 2.5-inch, SATA 6 Gbps
- SilverStone Fortress FT05 case – ATX
- be quiet! Straight Power 10 power supply – ATX, modular, 600W
- Microsoft
Windows 7 Ultimate operating system, 64-bit
|
|
Measurement and Analysis Tools
- Media Player
Classic – Home Cinema to play H.264/VC-1 video. - TMPGEnc Xpress video encoder for stress testing.
- Resident Evil 6 standalone benchmark for stress testing.
- Prime95
processor stress software. - FurMark
openGL GPU stress software. - CPU-Z
to monitor CPU frequency and voltage. - GPU-Z to
monitor GPU temperature, load level, and fan speed. - SpeedFan to monitor system temperatures and monitor/change fan speeds.
- AIDA64
to monitor system temperatures and fan speeds. - Extech
AC Power Analyzer 380803 AC power meter, used to measure system power consumption. - PC-based spectrum analyzer:
SpectraPlus with ACO Pacific mic and M-Audio digitalaudio interfaces. - Anechoic chamber
with ambient level of 11 dBA or lower
Software
Gigabyte’s EasyTune software utility is generally a competent application for controlling fans. In the original review of this board, I had absolutely no problems with controlling 120/140 mm case fans, but it simply won’t work properly with the FT05’s 180 mm Air Penetrators. When the fans are plugged in initially, their speeds correctly register and are controllable with the available preset profiles. However when they go through the calibration process in order to enable customized control, things go a bit haywire.
|
|
The fans go through their entire speed range as per usual, but at the end of the process, an error message states that it can no longer detect the fan speed. At below ~550 RPM, the fans run too slow for the speed to reported, even though they continue to function perfectly fine. This is common for 3-pin fans, but EasyTune’s reaction to this is rather odd. It disables smart fan and changes the fans’ maximum speed to approximately 700 RPM (even though it goes above 1200 RPM at the beginning of calibration procedure), which is about the same as the low speed when using the built-in fan controller. This happens regardless of whether the integrated fan controller is connected.
|
|
Given EasyTune’s bizarre behavior, I turn to our old standby alternative, SpeedFan, which when configured, offers much of the same functionality. Unfortunately, it doesn’t see the board’s CPU temperature or the CPU fan speed sensor, and doesn’t allow the CPU fan to be controlled at all. Thankfully, the CPU fan can be moved to one of the system fan headers and operated under voltage rather than PWM control without issue. EasyTune and SpeedFan fight each others’ fan settings so EasyTune has to be thrown out the door completely. To make up for the missing CPU temperature/fan sensor data, AIDA64 is employed.
|
|
For controlling the GPU fans, ASUS’ GPU Tweak is our GPU utility of choice, as it offers adjustable/dynamic fan control and a customizable charting pane that can keep track of various GPU properties including temperature, fan speed, voltage, clock speeds, etc. Unfortunately, the card lacks the ability to shut the GPU fans off completely; they have a minimum fan speed of 1450 RPM (60%).
Overclocking can also be performed although this particular video card’s power target can’t exceed 106% of stock (unless perhaps the BIOS is modified), limiting how much higher the clock speeds can be boosted. The GPU temp target (the temperature at which fan control is overriden) starts at 79°C an tops out at 91°C.
Baseline Noise
Before any stress testing let’s take a quick look at the noise produced by each part. The system was kept on but idle, and all the fans were stopped/unplugged (except for the power supply) to isolate individual components. Our ultra-quiet hemi-anechoic chamber has a noise floor of 10~11 dBA@1m, and the baseline of this system, with just the PSU on, measures barely above that.
| 2 x GTX 970 + Fortress FT05 System Component Noise Levels (idle, power supply on) | ||
| Noise-producing Components | Avg. Fan Speed | SPL @1m |
| CPU fan | 700 RPM | 14~15 dBA |
| 800 RPM | 16~17 dBA | |
| 1000 RPM | 20 dBA | |
| 1350 RPM (max) | 25 dBA | |
| Case fans | 550 RPM | 19 dBA |
| 600 RPM | 22 dBA | |
| 730 RPM (low) | 27 dBA | |
| 1030 RPM (med) | 36~37 dBA | |
| 1280 RPM (high) | 41~42 dBA | |
| GPU fans | 1450 RPM (min) | 20 dBA |
| 1680 RPM | 24 dBA | |
| 1920 RPM | 27~28 dBA | |
| 2400 RPM (max) | 33 dBA | |
| Measuring mic positioned 1m at diagonal angle left/front of case. Baseline noise level (PSU on only): 11 dBA@1m Ambient noise level: 10~11 dBA@1m. | ||
The GPU fans are almost always the biggest noise generators in any gaming system, and this machine is no different. What is disappointing is that they limit the idle noise level as well. The GPU fans’ minimum speed is 1450 RPM, which effectively produces 20 dBA@1m all on their own.
The case fans are even louder if they are controlled by the built-in controller. The low setting causes the fans to spin at 730 RPM, emitting 27 dBA@1m, well above the noise levels I’m shooting for, even on load, let alone idle. Motherboard fan control can be used to slow them down further, but as mentioned earlier, below 550 RPM, the fan speed can’t be detected, and their physical location makes it difficult to measure manually. 550 RPM is still slightly quieter than the GPU fans, so it’s a good a starting point as any.
|
|
As for the quality of noise, the GPU fans have a benign profile with a mostly smooth sound and little if any tonality. If you put your ear up close to them, a slight rattle is audible but once the side panel is closed, this effect disappears completely.
Similarly, the case fans have a rather clicky nature, something that’s afflicted the Air Penetrator series for their entire history but it’s difficult to detect at distance during normal operations.
Stress Test Results: CPU-centric
Testing begins with CPU-centric applications to see how it performs with non-gaming
tasks. The machine measures 23 dBA@1m, which is quiet, but quite high for a system with zero/little graphical load being placed on it. The video card fans’ minimum speed of 1450 RPM makes it impossible to get the system below 20 dBA@1m, and while the system fans are spinning at a mere 550 RPM, their large size means they emit much more noise than 120/140 mm models at the same speed.
| System Measurements | ||||
| System State | Idle | x264 Playback | Video Encoding | Prime95x4 |
| CPU Temp | 14°C | 19°C | 31°C | 38°C |
| MB Temp | 29°C | 28°C | 29°C | |
| PCH Temp | 34°C | 33°C | 34°C | |
| SSD Temp | 26°C | 27°C | ||
| GPU1 Temp | 26°C | |||
| GPU2 Temp | 29°C | |||
| System Power (AC) | 55W | 73W | 102W | 121W |
| CPU fan at 700 RPM, system fans at 550 RPM, GPU fans at 1450 RPM (minimum/auto). System noise level: 23 dBA@1m. Ambient temperature: 21°C. | ||||
When lightly taxed, the system runs quite cool as one would expect, with CPU and GPU temperatures staying slightly below 20°C and 30°C respectively. Video encoding creates much greater demand on the processor, producing a CPU temperature rise of 12°C while running Prime95 brings it up an additional 7°C. Board and GPU temperatures are fairly stable throughout these tests. The extra thermal output from the CPU doesn’t really affect the rest of the machine.
Stress Test Results: GPU-centric
For our GPU-intensive states, we use the Resident Evil 6 Benchmark Tool. It doesn’t cause a steady load like some synthetic tools but it seems to put the highest peak stress on the GPU. The other test is a more demanding combination of Prime95 and FurMark, an incredibly demanding utility that pins GPUs at their redline. Prime95 is run with only two threads instead of the maximum four, as most games run with less than 50% CPU utilization. The two stress utilities combined still draw more power and creates much more heat than any PC game title. The GPUs’ automated fan control is switched off with fan speeds adjusted manually in order to achieve/maintain a target GPU temperature of 85°C or lower.
| System Measurements | ||||
| System State | Resident Evil 6, Peak | Resident Evil 6, Peak (106% power) | Prime95x2 + FurMark | Prime95x2 + FurMark (106% power) |
| GPU1 Fan Speed* | 1450 RPM | |||
| GPU2 Fan Speed* | 1450 RPM | 1560 RPM | 1920 RPM | |
| GPU1 Clock | 1404 MHz | 926 MHz | 1001 MHz | |
| GPU2 Clock | 1404 MHz | 1215 MHz | ||
| CPU Temp | 35°C | 36°C | 49°C | 47°C |
| MB Temp | 46°C | 46°C | 49°C | 49°C |
| PCH Temp | 52°C | 53°C | 56°C | 57°C |
| SSD Temp | 38°C | 39°C | 42°C | 42°C |
| GPU1 Temp | 62°C | 63°C | 64°C | 67°C |
| GPU2 Temp | 78°C | 79°C | 85°C | 85°C |
| System Power (AC) | 404W | 422W | 429W | 443W |
| SPL@1m | 23 dBA | 23 dBA | 23~24 dBA | 26 dBA |
| *set as low as possible to maintain a GPU temperature of ~85°C on load. CPU fan at 700 RPM, system fans at 550 RPM. Ambient temperature: 22°C. | ||||
The Resident Evil 6 test causes the system to draw more than 400W AC, more than triple the power consumption elicited by a full bout of Prime95. Despite this immense load, the video cards are relatively comfortable, even with their fans set at minimum speed. The card situated closer to the power supply barely breaks a sweat, stabilizing at just 62°C while the warmer card settles at a reasonable 78°C. The motherboard and PCH temperatures increase substantially as one would expect and even the SSD, which is positioned on the opposite of side of the motherboard tray, warms up a great deal. In this state, the GPU uses synchronous clock speeds of 1404 MHz, 34 MHz higher than the designated boost frequency (normal behavior for Nvidia cards). Surprisingly, increasing the cards’ power target to maximum (106%) doesn’t alter the clock speeds of either card but still causes a small bump in power consumption, though the added load isn’t enough to warrant higher fan speeds.
Oddly, our Prime95 + FurMark causes the GPUs to run at asymmetric clock speeds despite being set to sync up in GPU Tweak. This test pushes both GPUs further but as the hotter card runs faster, it’s requires a 110 RPM increase in fan speed in order to hold the temperature steady at the desired 85°C, causing the noise level to go up ever so slightly. Weird behavior is encountered once again when the target power is increased. The cooler card experiences a bump in clock speed while the hotter card remains the same, but the later actually heats up further still, requiring drastically higher fan speed to compensate. I’m not sure exactly what’s going on here but the end result is a 2~3 dB spike in noise level.
This most demanding state draws about 40W more than than the Resident Evil 6 test at stock settings. This is close to what you can expect with a pair of GTX 980s with similar cooling solutions running Resident Evil 6, though it would probably be quieter as the clock speeds would presumably be in sync.
 |
Our GTX 970 samples do suffer from coil whine, squealing in a waxing and waning manner, particularly during the Resident Evil 6 test, but the effect is mild. With the case sealed up, it isn’t noticeable over the sound produced by the other components. Enabling V-Sync dissipates it greatly but the GPUs runs this game with very high framerates and V-Sync caps it at our 60 fps to match our display’s 60 Hz refresh rate, reducing the overall stress experienced by the video cards. Overall, the noise generated by the system is fairly inconspicuous, with a nice broadband profile and a lack of distinct tones. The noise level on load is also much lower than any dual GPU system we’ve ever assembled.
| System Measurements (Prime95x2 + FurMark) | ||||
| CPU Fan Speed | 700 RPM | 800 RPM | 700 RPM | 700 RPM |
| System Fan Speeds | 550 RPM | 550 RPM | 600 RPM | 550 RPM |
| GPU1 Fan | 1450 RPM | |||
| GPU2 Fan | 1560 RPM | 1560 RPM | 1460 RPM | 2070 RPM |
| CPU Temp | 49°C | 44°C | 43°C | 47°C |
| MB Temp | 49°C | 48°C | 48°C | 47°C |
| PCH Temp | 56°C | 56°C | 55°C | 55°C |
| SSD Temp | 42°C | 41°C | 41°C | 40°C |
| GPU1 Temp | 64°C | 65°C | 64°C | 63°C |
| GPU2 Temp | 85°C | 85°C | 85°C | 80°C |
| System Power (AC) | 429W | 428W | 428W | 432W |
| SPL@1m | 23~24 dBA | 24 dBA | 24 dBA | 27 dBA |
| Ambient temperature: 22°C. | ||||
Playing with the other fan speeds has some interesting effects. Speeding up the CPU fan by 100 RPM makes the system a tad noisier and cools down the processor by 5°C. Bumping up the intake fans by just 50 RPM is even better for the CPU and also improves temperatures system-wide. This move allows the hotter GPU to return to almost minimum fan speed while maintaining the same temperature. If you prefer an even cooler GPU, in our system, the hotter video card requires a fan speed of more than 2000 RPM to drop the GPU core down to 80°C. The stock cooler is certainly effective but as its fan speed tops out at 2400 RPM, there’s not that much upper headroom if you prefer a cooler running card.
AUDIO RECORDINGS
These recordings were made with a high resolution, lab quality, digital recording
system inside SPCR’s own 11 dBA ambient anechoic chamber, then converted to
LAME 128kbps encoded MP3s. We’ve listened long and hard to ensure there is no
audible degradation from the original WAV files to these MP3s. They represent
a quick snapshot of what we heard during the review.
Each recording starts with ambient noise, then 5~10 second segments of product
at various states. For the most realistic results,
set the volume so that the starting ambient level is just barely audible, then
don’t change the volume setting again while comparing all the sound files.
2 x GTX 970 + SilverStone Fortress FT05 System, CPU fan at 700 RPM, system fans at 550 RPM
— Resident Evil 6 Demo Benchmark, GPU fans at 1450 RPM (23 dBA@1m)
— Prime95x2 + FurMark, 106% target power, GPU fans at 1920 RPM (26 dBA@1m)
FINAL THOUGHTS
| SPCR’s Quiet SLI Gaming PC Component List | ||
| Build Components | Street Price | Alternatives |
| Intel Core i5-4690K | $225 | Intel Core i7-4790K – $330 Intel Core i5-4590 – $195 |
| Scythe Kotetsu | $40 | Scythe Mugen MAX – $50 Scythe Mugen 4 – $50 Be Quiet! Shadow Rock Slim – $50 Coolermaster Hyper 212 Evo – $30 |
| Gigabyte GA-Z87X-UD5 TH | $120 | Asus Z97-PRO – $190 Asus Sabertooth Z97 Mark 2 – $160 Asus Z97-A – $140 Gigabyte GA-Z97X-SLI – $115 |
| Kingston HyperX Genesis 2x4GB DDR3-1866 | $95 | Kingston HyperX Savage 8GB (2x4GB) DDR3-2133 – $75 Crucial Ballistix Tactical 8GB 8GB (2x4GB) DDR3-1866 – $70 |
| Zotac GTX 970 AMP! Extreme Core Edition | 2 x $380 | Asus Strix GTX 980 – $565 MSI GTX 980 Gaming – $560 Asus Strix GTX 970 – $330 MSI GTX 970 4GD5T OC – $330 |
| Samsung 850 Pro 256GB | $170 | Samsung 840 EVO 1TB – $430 Samsung 850 Pro 512GB – $285 Crucial MX100 512GB – $190 Crucial MX100 256GB – $100 |
| SilverStone Fortress FT05 | $185 | SilverStone Fortress FT02 – $230 SilverStone Raven RV03 – $140 SilverStone Raven RV05 – $125 |
| Be Quiet! Straight Power 10 600W | $130 | Seasonic X-650 – $115 Corsair RM650 – $110 Corsair CS650M – $90 |
| TOTAL | $1715 | $1265 (min. alternative config.) |
| Retail prices are subject to fluctuations. Please use the shopping links to check on current pricing; Don’t rely on the prices cited. | ||
The total for this project comes out to US$1715, making it an expensive build, but certainly not outrageous for a high performance SLI configuration. With some lower cost alternatives, the final tally can shaved down to US$1265.
At stock settings, the system has a pleasant sound and produces a noise level of 23 dBA@1m at idle, a bit high by our standards. The culprits are the GPUs, whose fans are unable to spin slower than 1450 RPM. The Zotac GTX 970 AMP! Extreme Core cards are highly clocked and well cooled but the unnecessarily high minimum fan speed is disappointing. For a heavy duty gaming system, it’s something you could live with but it makes me more greatly appreciate the Asus Strix or MSI Gaming GTX 900 series cards that can stop the fans completely. Surely, with the size of the heatsink employed in the GTX 970 AMP! Extreme Core, Zotac could afford to do the same! [Editor’s Note: Perhaps a firmware update could provide a less aggressive cooling option.]
On the bright side, the minimum fan speed is more than enough to adequately cool both cards when the rig is placed under any real work load, including gaming, so the system sounds the same regardless of what it’s being used for. Synthetic stress tests generate conditions that require slightly higher fan speeds, but the acoustic difference is negligible. This is a remarkable result for a rig with no aftermarket cooling.
The Fortress FT05’s layout is a huge advantage, as the video cards’ thermal loads do not interfere with each other much compared to a standard tower. They’re much happier sitting side-by-side rather than one on top of the other. My only real complaint is the built-in fan controller — its low setting is inadequate for a quiet system. Thankfully, the fans can be slowed down further by the motherboard and there should be plenty of headers available as it uses two 18 cm fans instead of more numerous 12/14 cm models.
For the past few years testing cases in our lab involved using an old CrossFire configuration with a pair of old HD 4870s (150W TDP), which have never yielded a result lower than 28 dBA@1m on load. The fact this more modern dual GPU build manages just 23 dBA@1m during gaming load is astonishing and makes this a successful venture in my view. A quiet high-end SLI gaming system is well within the realm of possibility for anyone with enough funds to pay for it.
Many thanks to Zotac, SilverStone, Be Quiet!, Intel,
Kingston, Gigabyte, Samsung, and Scythe
for sponsoring the components in this build guide.
* * *
Articles of Related Interest
Quiet Liquid Cooled Gaming PC Build Guide
Quiet Mini-ITX Gaming Build Guide #3: BitFenix Prodigy Edition
Quiet Mini-ITX Gaming Build Guide #2: NCASE M1 Edition
Journey to a Silent MicroATX Gamer
Quiet Mini-ITX Gamer Build Guide
Case
Basics & Recommendations
* * *
