sacah.net

ASRock A-Tuning - This utility is not for this platform - Windows 10

Sun, 30 Aug 2015 20:55:27 +1000

While most things went smoothly for my upgrade to Windows 10, A-Tuning from ASRock would not start. There is no Windows 10 version for download, and the Windows 8.1 version wouldn’t start. It would produce this error:

Error 0xFFFE “This utility is not for this platform”

thisisafold

After some searching I couldn’t find any answers, so I downloaded the Windows 7 version of A-Tuning and installed it in Compatability mode (Win 7), plus ran it in Win 7 compatability mode and it all runs fine now, controlling the fans is also working, so not sure why they can’t release a Win 10 version.

Intermittent NVidia 970 SLI Crashes Post-FX Full System Reset

Sat, 31 Jan 2015 21:55:27 +1100

After upgrading to 2 Galax GTX 970 OC EX gpus and setting up SLI, I found that occasionally while gaming in Surround, I’d get a full system crash, black screens, hard system reset, the works.

This first heppened in BF4 and GW2, though it was hard to figure out exactly what the issue was, I tried various driver reinstalls, under-clocking, closing all background programs, even going back 3 versions of NVidia drivers, to no avail.

thisisafold

One way I found to consistently crash it was Geeks3D FurMark, setting Resolution to 5760x1080, Anti-aliasing to 8x MSAA and in the Settings, Dynamic background, Burn-in, Xtreme burn-in and Post-FX. It’s the Post-FX that is what would cause it to crash.

I monitored voltages & temperatures, all seemed normal, so as a last resort I tried swapping the GPUs around. While doing this I noticed some writting on my PSU, as I’d installed it upside down I couldn’t read it, so I took a photo and realised it said each column was a different 12v line, so I was supplying half the power to the second card from the first 12v line, and half from the second 12v line. The CPU and first GPU were on the first 12v line.

I swapped the power cable so the second card was powered fully from the second 12v line and was able to run FurMark with Post-FX no problem.

Update

So it started crashing again, I was able to narrow it down to PSU. If I started the Pc after it had been off for 20min, I could run FurMark+PostFX no problem. Let the Pc warm up a bit and run FurMark+PostFX again after 15min and it would turn off.

Ordered a new PSU, installed, no more Pc turning off with FurMark+PostFX.

Some interesting stats on power draw, taken from the plug.

Power Draw
Watts	FurMark + PostFX + Prime95615 FurMark + PostFX500 FurMark + Prime95500 Prime95230 Idle100

Hopefully this is helpful to someone else (-:

Galax GTX 970 EX OC SLI Review

Fri, 23 Jan 2015 21:55:27 +1100

Recently upgraded from a GTX 770 to 2 GTX 970s. One concern I had was keeping the heat down to get the best performance, and most threads you read about SLI tell you to get a blower style cooler, although these give worse cooling results by themselves. I looked at my motherboard and case and figured I had enough room and airflow to take the risk, so after reading a ton of reviews I opted for 2 Galax GTX 970 EX OCs.

thisisafold

These are quite amazing cards, and an equally amazing price. I’d recommend you include these in your short list of which 970 to buy, check out all the in-depth reviews there are out there.

With both cards installed in my ASRock X99 motherboard, you can see there is a 2.5cm gap between the top and bottom card.

At the front of my case I have 2 120mm fans that pull air in and over the HDDs, this is to the right of the above picture.

At the rear, above the GPUs there is 1 120mm fan pulling air out the back of the case.

On the top of the case are 2 80mm fans pushing air in over a radiator that cools the CPU.

With the direction of the fins on the GPUs some air also makes it out the back of the graphics cards.

I also have my wifi PCIe card installed in the last panel, which blocks 25% of the left fan on the bottom card. These cards turn off the right fan at idle, so it’s the left fan that is running at idle, so maybe once I can remove this wifi card the bottom card will get slightly better temps.

With this not great airflow I manage the following temps:

Galax 970 EX OC Dual SLI Temperatures
Idle	Ambient28°C Top card35°C Bottom card33°C
Load	Ambient28°C Top card78°C Bottom card65°C

Idle (Windows 8.1 Desktop)

Load - (Unigine Heaven Benchmark - Extreme in Surround)

The side panel of my Corsair 400R case has room for 2 140mm fans that point right at the 2 GPUs, I’m awaiting some fan extension cables, then I’ll try adding fans to the side and optimising the airflow of my case a bit better.

In a quick test I set one fan on the side to push air in, over the GPU and it lowered the temp by about 5°C, so once I set this up properly, I think these cards should do very well.

Overclocking

I was able to push both cards core +180mhz and mem +400mhz with a power boost of 120%, nice and stable. Pushing the core to +185mhz would cause issues. These figures seem pretty standard for what other Galax 970 EX OC cards are getting.

Google Analytics - How to log duration of a single page visit

Wed, 07 Jan 2015 21:55:27 +1100

So no doubt you are here because of Google Analytics showing you Avg Session Duration being 00:00:00 and Bounce Rates of 100%. You are thinking, surely people aren’t that turned off my site that they are leaving instantly.

Google Analytics calculates a users time on your site (session duration) by comparing the time from the first page view to the second. This is fine for sites where people browser content, most of the views will be accurate. But for sites with a variety of information in single pages, like blogs, you may be like me and find people enter 1 page, find their answer and leave. But without knowing how long someone spent reading your page you don’t really know if they took 1 look and figured it wasn’t right for them, or if they spent 15min carefully reading everything.

thisisafold

So looking into how others have solved this, here are the solutions I found others using:

beforeunload or unload events

The listener function usually fires anothe page view or a custom event, either the old ga.js code of

_gaq.push(['_trackPageview', 'track']);

Or the newer analytics.js code of

ga('send', 'pageview');

The issue I have with this is it seems to create another pageview record, and there is no guarantee any requests will complete before the browser destroys your page, so depending on the users machine and network, the event may never reach Google to be recorded.

In the future when more browsers support beacons this will be a great way to do it.

custom events from user interactions

They would fire a custom event on some user interaction, such as scrolling, mouse move or click.

The issue I have with this is it’s not hugely accurate, they might not click and on a large monitor they might not scroll, but still have spent a good amount of time reading. And for mouse moves, you could be firing many events.

my solution

Google Analytics will count any event as an interaction, so this will increase the Session Duration to the time of the last fired event.

I fire an event every 30 seconds, which means for single page visits, I’ll be accurate to the nearest multiple of 30 seconds. Depending on the range of your content, you may find you want to decrease this number. 30 seconds seemed to be a good number of not spamming a users network, while still getting reasonable statistics of visitor duration.

Below is the code I used

 setInterval(function () { ga('send', 'event', 'beacon', 'page', 'open'); }, 30000);

Here is where I put it in relation to the Google Analytics code

 (function(i,s,o,g,r,a,m){i['GoogleAnalyticsObject']=r;i[r]=i[r]||function(){ (i[r].q=i[r].q||[]).push(arguments)},i[r].l=1*new Date();a=s.createElement(o), m=s.getElementsByTagName(o)[0];a.async=1;a.src=g;m.parentNode.insertBefore(a,m) })(window,document,'script','//www.google-analytics.com/analytics.js','ga');ga('create', 'UA-XXXXXXXX-X', 'auto'); ga('send', 'pageview'); setInterval(function () { ga('send', 'event', 'beacon', 'page', 'open'); }, 30000);

Here are the results after a few days

SLI scaling analysis with GTX 780

Mon, 10 Nov 2014 20:03:18 +1100

In this SLI scaling analysis, I have selected a number of different variables to test, which all come into play with SLI scaling in games.

The cards used to test are Nvidia GeForce GTX 780’s (yes they are not the most powerful cards, but still up to the task, and this is about scalability, not brute power). These are tested in single, dual and triple card configurations, in a variety of different games.

Settings were chosen which work well with a dual card configuration, designed to maximise all available graphics settings whilst still being playable. This allows the cards to be pushed extremely hard, which also allows the different variables to show their effects on scaling, while producing a more accurate result.

My definition of playable is not an arbitrary number of frames, but the subjective user experience of the game feeling smooth enough to be an enjoyable experience.

The following variables have been chosen to assess their impact on fps, perceived smoothness i.e. with microstuttering etc and to discover any bottlenecks which may impact on 2-way and 3-way SLI, for instance: is a quad core hyperthreaded processor capable of keeping up with 3 demanding cards, I’ve read many reviews that have often made assumptions about where the bottleneck (if any) is occuring, the data presented here will hopefully shed some more light on these factors.

What we’re testing
Frame rates; minimum, average and maximum, of course the minimum and average fps is more important to playability.
Each graphics cards maximum temperatures during play.
Each GPU’s load.
Each GPU’s core clock rate.
The amount of gpu memory used.
The CPU’s load and frequency (frequency is shown as a % of the CPU clock rate, which is at 4.4Ghz)

thisisafold

Test rig
OS	Windows 8.1
CPU	Core i7 4820k OC’d to 4.4Ghz
Mobo	Asrock x79 Extreme 6
RAM	16GB of G.Skill Sniper 1866Mhz in quad channel configuration
HDD	Samsung 840pro 250GB HDD (System & BF4) W.D. Black 7200rpm 3TB
Case	Thermaltake Overseer (Closed to represent a normal gaming setup)
Monitors	3 x 1920 x 1080 res in surround for res of 5760 x 1080
GPUs	Gigabyte Windforce GTX 780 OC (Boost clock of 1058Mhz)
	Gigabyte Windforce GTX 780 OC (Boost clock of 1085Mhz)
	Palit GeForce GTX 780 with reference cooler (Boost clock of 1006Mhz)

In the 3 card config, a fan is placed on the side of the case to help control temps, however temps are still pushed to achieve throttling where possible, GPU3’s boost clock was also lowered from 1085 to 1032 to assist with heat. The graphics cards were also moved around to also facilitate better heat management.

With the three card config, I had to override the default nvidia SLI rendering setting to SLI_Rendering_Mode_Force_AFR_OF_SFR_Fallback_3AFR

Software used to gather testing results and edit the the profile settings:

Nvidia Inspector
MSI Afterburner
Windows Resource Monitor
Battlefield 4
Batman: Arkam City
Farcry 3
Metro Last Light Redux
Tomb Raider
Watch Dogs

I am not responsible for any damage caused by using these programs

Battlefield 4:

Res 5760 x 1080
Res scale 100%
Motion blur off
Everything on ultra
MSAA x 4
AA post off
Ambient occlusion HBAO
Multiplayer map - Siege of Shanghai, ~60 players in map.

Battlefield 4 FPS Results
Max	Triple110fps Dual80fps Single45fps
Avg	Triple75fps Dual65fps Single40fps
Min	Triple40fps Dual55fps Single31fps

Battlefield 4 SLI Scaling Results
Max	Triple244% Dual177% Single100%
Avg	Triple187% Dual162% Single100%
Min	Triple129% Dual177% Single100%

Single configuration

GPU1 - Temp: 65 degrees. Load: 98-99%. Core clock: 1058Mhz. Mem: 2800Mb
CPU - Load: 50-80%. Frequency: 100%
FPS: min - 31, average - 40, max 45.
Complete maps played: 1

In the single run, the GPU load was very good and stable, only varying between 98 and 99%. The CPU was worked fairly hard, but never reached levels to bottleneck and frequency never dropped below 100%. The memory use got quite high, but didn’t quite max out the 3GB on board the 780. The GPU temperature was quite cool throughout the game, settling on 65 degrees celcius and the core clock remained stable at 1058Mhz with GPU boost, never showing signs of throttling.

Game play with the single card was manageable but certainly not optimal, the fps dropped to noticeable ranges fairly consistently, and the change in frame rates was quite noticeable. Lowering settings such as MSAA would make it run smoothly.

Dual GPU configuration

Note second GPU’s clock normally 1006 at boost this can produce a variance of up to 2.5% less compared to both GPU’s clocked at 1058Mhz.

GPU1 - temp: 79 degrees. Load: 90-95. Core clock: 1058Mhz. mem: 2800Mb
GPU2 - temp: 79 degrees. Load: 95-99. Core clock: 1006Mhz. mem: 2800Mb
CPU - Load: 60-70%. Frequency: 100%
FPS: min - 55/65/80
Complete maps played: 2

In the dual test, the GPU loads remained very stable with neither dropping below 90% load and spending the majority of the time 95% load and above for both. The temperatures did get warm, but stopped just short of reaching the throttling point of 80 degrees. GPU core clocks remained at their respective optimals of 1058 and 1006Mhz, memory remained just short of max. The CPU load of this run stayed between 60-70% (keep in mind all BF4 multiplayer games are different and will have a difference in CPU load depending on what is happening). The FPS remained respectable with min-55, average of 65 and max of 80. Gameplay felt smooth for the most part with only a very occasional notice of slowdown when it dipped to 55 which, while noticeable, did not effect gameplay.

Triple GPU configuration

Note With adjusted clocks, 3GPU config can produce a variance of up to 2.5% less than with all GPU’s at 1058Mhz.

GPU1 - temp: 67 degrees. Load: 70-90%. Core clock: 1006Mhz. mem: 2800Mb
GPU2 - temp: 77 degrees. Load: 70-90%. Core clock: 1058Mhz. mem 2800Mb
GPU3 - temp: 80 degrees. Load: 80-95%. Core clock: 875-1032Mhz. mem 2800Mb
CPU - Load: 80-90%. Frequency: 100%.
FPS: min - 40. Average: 75. max - 110
Complete maps played: 3

With 3 graphics cards, things get a bit more unpredictable. I had to play this map from start to finish 3 times to get the proper data, which includes two of those being done after a restart. All 3 times, the tower went down, each game lasted from 20-30 minutes. As evidenced by the graph, the frame rate would drop to as low as 40, this wasn’t very often, but still happened enough that it had to be declared the minimum frame rate. For the most part, the minimum frames didn’t drop below 60, and gameplay was fairly smooth for much of the time, however there were periods microstuttering which happened fairly often, while it was certainly still playable, the microstuttering did make it feel less smooth than playing with two cards.

The CPU load went from 80% up to a whopping 90% with 100% frequency, however at no point did any of the threads reach 100%, indicating that the slow downs are a scaling issue rather than a CPU bottleneck issue. The GPU temps were good, with two of the GPU’s staying below throttling temp, the third however did reach 80 degrees and did throttle to as low as 875Mhz, however the temp didn’t go any higher. While having 3 cards was playable, two cards is optimal for BF4 at these settings.

Batman: Arkam City

Res - 5760 x 1080
AA = CSAA x16
Dx11 = MVSS & HBAO
Tesellation / Detail = Maximum
Physx = High
Everything else maxed out
Built in benchmark tool, run several times to warm the cards up

Batman: Arkam City FPS Results
Max	Triple150fps Dual87fps Single44fps
Avg	Triple121fps Dual79fps Single40fps
Min	Triple75fps Dual54fps Single39fps

Batman: Arkam City SLI Scaling Results
Max	Triple200% Dual161% Single100%
Avg	Triple161% Dual146% Single100%
Min	Triple112% Dual102% Single100%

Single GPU configuration

GPU1 - Temp: 62 degrees. Load: 70-98%. Core clock: 1058Mhz. Mem: 2000Mb
CPU - Load: 40-60%. Frequency: 100%
FPS: min - 39, average - 54, max 75
Runs: 3

In the benchmark run. The Gpu’s temperature stayed cool, even after a few runs. The load on the GPU varied from 70-98% with most of the time it being in the 90’s. The time that it did drop in load and to it’s minimum frames was during a section that involved a lot of physx, the section also had the feel of being less than fully optimised, considering the same section produced almost identical frames in the dual test. The core clock never bottlenecked, remaining at 1058Mhz and the memory stopped at 2000Mb, leaving plenty of headroom. The CPU load was fairly average, staying between 40-60% with the frequency never below 100%. The scene was fairly smooth for the most part, however one scene with lots of particle effects and physx was noticeably chunkier.

Dual GPU configuration

Second GPU boost clock: 1006Mhz) This can create a variance of up to 2.5% decrease in performance compared to both cards at GPU1 clocks.

GPU1 - temp: 77 degrees. Load 85-95%. Core clock 1058. Mem: 2000mb
GPU2- temp: 75 degrees. Load 90-99%. Core clock 1006. mem: 2000mb
CPU - load: 50-70%. Frequency: 100%.
FPS: min - 40, average - 79, max - 121
Runs: 3

In the dual card benchmark. The GPU’s temperature stayed just short of throttling temps, even after several runs. The load of both the GPU’s varied, but not nearly as much as with the single GPU. The core clock showed no signs of throttling and the CPU load stayed in the 50 - 70% range with the frequency at 100%. The scaling from 1 to 2 GPU’s was quite poor, the minimum being almost the same as with a single card, suggesting some poor optimisation. The average increase was quite poor at a meager 46.2% over 1 card. The scene was quite smooth, with the only noticeable slowdown being the same scene with lots of particle and physx effects.

Triple GPU configuration

Note With adjusted clocks, 3GPU config can produce a variance of up to 2.5% less than with all GPU’s at 1058Mhz.

GPU1: temp - 57 degrees. Load: 50-70%. Core clock: 1006Mhz. mem: 2000Mb
GPU2: temp - 65 degrees. Load: 50-70%. Core clock: 1058Mhz. mem: 2000Mb
GPU3: temp - 73 degrees. Load: 50-70%. Core clock: 1032Mhz. mem: 2000Mb
CPU: Load: 50-70%. Frequency: 100%.
FPS: min - 44, average - 87, max - 150
Runs: 3

When I ran with 3 GPU’s, the load on each GPU dropped to 50-70%, resulting in relatively poor scaling, particularly for the minimum fps. The temps stayed all under control with the help of the side fan, and the cpu ranged between 50-70% usage with 100% frequency, indicating no CPU bottleneck, however 1 thread did reach 95% for a moment, however that was only for a second and didn’t result in any decrease of frames that lined up with the benchmark. The whole benchmark was smooth with only a little stuttering noticed when it was transitioning at a couple of points. The particle and physx effects scene from the earlier benchmarks where I noticed the slowdown, was far smoother with 3 cards.

Farcry 3

Res: 5760 x 1080
MSAA x 2
Alpha to coverage = Enhanced
SSAO = HBAO
Every other setting = maxed out
Random driving and explosions

Farcry 3 FPS Results
Max	Triple86fps Dual60fps Single40fps
Avg	Triple73fps Dual55fps Single32fps
Min	Triple60fps Dual49fps Single28fps

Farcry 3 SLI Scaling Results
Max	Triple215% Dual150% Single100%
Avg	Triple228% Dual171% Single100%
Min	Triple214% Dual175% Single100%

Single GPU configuration

GPU: Temp - 68 degrees. Load: 99%. Core clock - 1058Mhz. GPU mem - 2000Mb
CPU: Load - 40%. Frequency - 100%
FPS: min - 28, average - 32, max - 40
Time played: 15mins

In this benchmark run, gameplay was rarely smooth due to the high settings for a single card. The GPU load and core clock showed no throttling and that the load was stable, maintaining 99% the entire time. The CPU load stayed fairly low and consistent at 40% and the frequency never dipped below 100%. While the gameplay was hardly ideal, there wasn’t too much variation between min and max frames which shows good optimisation and stability.

Dual GPU configuration

(Second GPU clock rate normally at 1085Mhz) This can create a variance of up to 1.2% increase in the performance compared to both cards being at GPU1 clocks.

GPU1: Temp - 85 degrees. Load - 99%. Core clock - 1058 - 1045Mhz, mem - 2000Mb
GPU2: Temp - 77 degrees. Load - 98-99%. Core clock - 1085Mhz, mem - 2000Mb
CPU: Load - 50%. Frequency - 100%
FPS: min - 49, average - 55, max - 60
Time played: 30mins

In the dual GPU benchmark. Gameplay maintained being smooth for much of the time, occasionally a slight slowdown was noticed, but did little to effect gameplay, as listed, the min and max framerates were fairly close so it maintained good stability. Of note is GPU1 which became very hot, this game does push the GPU’s very hard and this is the problem with having GPU’s with aftermarket coolers in a closed case, the top card did throttle, however only 1.2%, it was able to stay well above stock clocks and still deliver a very stable experience.

Triple GPU configuration

Note With adjusted clocks, 3GPU config can produce a variance of up to 2.5% less than with all GPU’s at 1058Mhz.

GPU1: temp - 73 degrees. Load - 85-98%. Core clock - 1006Mhz, mem - 2000Mb
GPU2: temp - 79 degrees. Load - 85-98%. Core clock - 1058Mhz. Mem - 2000Mb
GPU3: temp - 86 degrees. Load - 90-98%. Core clock - 1032Mhz. Mem - 2000Mb
CPU: Load: 50%. Frequency - 100%
FPS: min - 60, average - 73, max - 86
Time played: 30mins

With 3 graphics cards. Gameplay was very smooth for the most part, there were very rare micro-stutters, and the frame rates never dropped below 60, if you have the standard 60hz monitors, then this would be an ideal game and settings to cap the fps to 60, it would also save on some of the heat, which did get rather warm, with one of the cards reaching 86 degrees, that’s still pretty good for a tri sli setup. I did install a 200mm case fan onto the side to pull warm air away from the cards which seems to have helped, otherwise it would likely have gone higher. The other two cards stopped at 73 and 79 degrees. Without capping the frames, 3 cards felt smoother than 2, only just though, capping the frames at a suitable level will make both better.

Metro Last Light Redux

Res - 5760 x 1080
SSAA = off
Advanced physx = on
TF = 16x
Tesselation = very high
Motion blur = normal
Quality = Very high.
Heavy squad mission

Metro Last Light Redux FPS Results
Max	Triple64fps Dual45fps Single23fps
Avg	Triple55fps Dual42fps Single22fps
Min	Triple48fps Dual40fps Single19fps

Metro Last Light Redux SLI Scaling Results
Max	Triple265% Dual195% Single100%
Avg	Triple250% Dual190% Single100%
Min	Triple252% Dual210% Single100%

Single GPU configuration

GPU: Temp - 65 degrees. Load - 98%. Core clock - 1058Mhz. Mem - 1600Mb
CPU: Load - 30%. Frequency - 100%
FPS: min - 19, average - 22, max - 23
Runs: 3

In the single GPU benchmark, the game was noticeably slow and would need to have the settings turned down considerably to make it playable. The GPU temp stayed a cool 65 degrees. The load remained steady at 98% and the core clock stayed at 1058Mhz. The CPU was idling away at 30% load and 100% frequency.

Dual GPU configuration

(second GPU’s normal clock rate is 1085Mhz, this can create a variance of up to 1.2% increase over both GPU’s being clocked at 1058Mhz)

GPU1: Temp - 79 degrees. Load - 80-90%. Core clock - 1058Mhz. Mem - 1600Mb
GPU2: Temp - 69 degrees. Load - 98%. Core clock - 1085Mhz. Mem - 1600Mb
CPU: Load - 50%. Frequency 100%
FPS: min - 40, average - 42, max - 45
Runs: 3

In the Dual GPU benchmark, we notice something odd, aside from the fact that the average and maximum scaling is extremely good, the minimum scaling as increased to 210% of the single card. Some of the improvement can be attributed to the second card having a slightly higher clock rate, amounting to a possible increase in performance of 1.2%, there is also the possibility of a small difference in performance which can happen as well. However I have seen this replicated in benchmarks that others have made, possibly having the additional card was able to deal with the regular textures and calculations as well as the Advanced physx in a more efficient manner than simply 1 card attempting to manage all of those elements at once (Note that the GTX 970 and 980 handle physx considerably better and the benchmarks that I have seen with them, the minimum scaling isn’t quite as high).

Regardless, the results are exceptional and the gaming experience is vastly improved by having a second card, while it does feel a little “chunky” due to the fps being a little lower than optimal, it doesn’t effect gameplay too badly and is certainly vastly better than a single card.

The GPU’s temp gets very close to throttling but thankfully doesn’t. The load fluctuates a little, but as you can see from the scaling and the closeness of the min and max frames, it’s keeps everything stable. Notice the CPU use has increased substantially from a single card, still no evidence of any bottleneck.

Triple card configuration

Note With adjusted clocks, 3GPU config can produce a variance of up to 2.5% less than with all GPU’s at 1058Mhz.

GPU1: temp - 62 degrees. Load - 70-85%. Core clock 1006Mhz
GPU2: temp - 71 degrees. Load 70-85%. Core clock 1058Mhz
GPU3: temp - 80 degrees. Load 85-95%. Core clock 967-1032Mhz
CPU: Load - 50%. Frequency - 100%
FPS - min - 47, average - 52, max - 60
Runs: 3

Playing the same mission with 3 cards, the GPU usage was a little lower per card, but still delivered more power, while the scaling wasn’t nearly as impressive from 2-3 as it was from 1-2, it’s still a very impressive improvement with solid increases across the min, average and max fps. The framerates were all improved but also had more variation from the min to the max, which was noticeable only on occasion, mainly at the very start of the mission where all 3 screens were full of explosions.

Tomb Raider

Res - 5760 x 1080
Motion blur = on
Screen effects = on
Texture Quality = Ultra
Anistropic filtering - 16x
AA = SSAA 2x
Everything else maxed out
Built in benchmark tool

Tomb Raider FPS Results
Max	Triple76fps Dual54fps Single29fps
Avg	Triple60fps Dual44fps Single23fps
Min	Triple46fps Dual35fps Single18fps

Tomb Raider SLI Scaling Results
Max	Triple262% Dual186% Single100%
Avg	Triple260% Dual191% Single100%
Min	Triple255% Dual194% Single100%

Single card configuration

GPU: Temp - 65 degrees. Load 99%. Core clock - 1058Mhz. Mem - 2000Mb
CPU: Load - 20%. Frequency 80-100%.
FPS: min - 18, average - 23, max - 29
Runs: 3

In the single GPU benchmark. The scene was noticeably choppy, understanding from the strain put on the graphics card. The temp stayed cool at 65 degrees, core clock stayed stable at 1058Mhz. The CPU idled at 20% and in fact had so little to do that the frequency dropped to 80% at times. The GPU’s load stayed steady at 99% indicating that it was handling almost everything.

Dual card configuration

(second GPU’s normal clock rate is 1085Mhz, this can create a variance of up to 1.2% increase over both GPU’s being clocked at 1058Mhz)

GPU1: Temp - 79 degrees. Load 99%. Core clock - 1058Mhz. Mem - 2000Mb
GPU2: Temp - 70 degrees. Load 99%. Core clock - 1085Mhz. Mem - 2000Mb
CPU: Load - 30%. Frequency - 100%.
FPS: min - 35, average - 44, max - 54
Runs: 3

This is a game that scaled extremely well. The scene was smooth for the most part, there was the occasional point where the slow down was noticeable, the settings are probably a little high, however AA goes from FX AA (blurry edges) straight to SSAA x2 (nice crisp textures, but huge impact on performance). GPU1 stayed just short of throttling. Both GPU’s stayed on a constant load of 99% and maintained their respective core clocks. The CPU had to wake up a bit with dual cards, increasing to 30% load and 100% frequency.

Triple card configuration

Note With adjusted clocks, 3GPU config can produce a variance of up to 2.5% less than with all GPU’s at 1058Mhz.

GPU1: Temp - 65 degrees. Load 90-99%. Core clock 1006Mhz
GPU2: Temp - 75 degrees. Load 90-97%. Core clock 1058Mhz
GPU3: Temp - 80 degrees. Load 90-98% Core clock 928-1032Mhz
CPU: Load 30%. Frequency - 100%.
FPS: min - 46, average - 60, max - 76
Runs: 3

Tomb Raider scales excellently to 3 cards. There was only a very small amount of micro stuttering, that I had to look for. Overall the benchmark felt smooth and certainly was smoother with 3 cards than with 2, and significantly better than 1, with most frames sitting around the 60 mark. The GPU’s load all stayed 90% and above, the 3rd card did throttle a little when it hit 80 degrees after running the benchmark 3 times, but still managed to remain smooth. These extremely high settings are definately something three cards feels suited to manage.

Watch Dogs

Res - 5760 x 1080
Textures = high
AA = FXAA
GPU max buffered frames = 1
HBAO+ low
Everything else on and maxed out
Driving around the city, using same route for each test

Watch Dogs FPS Results
Max	Triple80fps Dual55fps Single45fps
Avg	Triple35fps Dual45fps Single35fps
Min	Triple23fps Dual35fps Single28fps

Watch Dogs SLI Scaling Results
Max	Triple177% Dual122% Single100%
Avg	Triple100% Dual128% Single100%
Min	Triple82% Dual125% Single100%

Single GPU configuration

GPU: Temp - 60 degrees. Load 98%. Core clock - 1058Mhz. Mem - 3000Mb
CPU: Load - 50-60%. Frequency - 100%.
FPS: min - 28, average - 35, max - 45
Time spent in game: 20mins

The single GPU run, while having overall fairly low frames, did feel a little smoother than the dual GPU run, as there were less wild fluctuations in fps. The GPU load stayed steady at 98% and the core clock didn’t move off of 1058Mhz. The memory maxed out at 3000Mb. The CPU maintained a steady load of 50-60% and 100% frequency. While it was noticeably slow, it was still playable.

Dual GPU configuration

(Note second card boost clock is 1085Mhz, this can create a variance of up to 1.2% increase performance compared to both at 1058Mhz)

GPU1 - 79 degrees. Load 50-85%. GPU core clock - 1058-797Mhz. Mem 3000Mb
GPU2 - 69 degrees. Load 50-85%. Core clock 1085Mhz. Mem - 3000Mb
CPU - Load - 50-60%. Frequency 100%
FPS: min - 35, average - 45, max - 55. Drop to 17-30fps in one area
Time spent in game: 20mins

The dual GPU run wasn’t overly ideal. The GPU load dropped significantly and fluctuated wildly depending on which area of the city you were in and what you were doing. Loading the Farcry 3 SLI profile helped in some areas of the city, and hindered in others (same thing happened when using the Assassins Creed Black Flag profile), I found it was best just to run with the default profile.

There were certain areas of the city which showed poor optimisation, with the primary GPU dropping it’s clock rate to 797Mhz before returning to 1058 shortly afterwards, with no identifiable reason why. GPU load for both GPU’s would go from being relatively stable around 85-95% to fluctuating wildly between 50-80%, and the frames along with it. The CPU load stayed fairly stable at 50-60% and the frequency at 100%. temps stayed just short of throttling for the period of testing. Stuttering happened occasionally.

The game was playable but the stuttering and slow downs would get rather frustrating at times, particularly when you could drive through a very busy section with lots of textures, to another area of the city which didn’t appear to have anything more going on and in some cases, had less going on, and yet would stutter and slow down. Playing with the single card actually felt smoother for the most part, despite having lower frames. There was a section in which the frames dropped as low as 17/sec, coinciding with the drop in GPU1’s slowed down clock rate, this didn’t happen with the single card. I left this off the chart because it was the only area I found which did it and putting it in would skew the results, as normally it was fine.

Triple GPU configuration

Note With adjusted clocks, 3GPU config can produce a variance of up to 2.5% less than with all GPU’s at 1058Mhz.

GPU1: Temp - 60 degrees. Load - 30-70-80%. Core clock - 863Mhz
GPU2: Temp - 68 degrees. Load - 30-70-80%. Core clock - 954Mhz
GPU3: Temp - 80 degrees. Load - 30-70-80%. Core clock - 954Mhz
CPU: Load - 50-70%. Frequency 100%
FPS: min - 23, average - 35, max 80
Time played: 2 hours

The scaling for 3 cards is absolutely abysmal, I checked the individual CPU thread loads to look for any signs of a bottleneck and they all stayed 70% or below. The GPU loads were pathetic, as well as the game slowing all cards to their base clocks. Most areas were very stuttery and slow, felt worse than both 1 and 2 cards were. Occasionally the fps would pick up briefly, then drop again, rain or shine, the results were the same. Basically this is a game to steer clear of if you intend to go more than 2 cards (Or disable SLI).

Conclusion

Some interesting results were found throughout the tests performed, with the different games all scaling slightly differently, at no point did the CPU ever show itself as being a bottleneck, even in BF4 where the load got to 90%, none of the threads reached 100%, and in all of the games tested, every thread had something to do, even if in some games, they were considerably lower load than others. The way the GPU’s behaved in different games showed the difference between a well optimised game and a poorly optimised game.

Tomb Raider showed itself to be an excellently optimised game with fantastic scaling, while Watch Dogs proved to be the exact opposite, with wildly fluctuating GPU loads and clocks, as well as some clear evidence of poor texture optimisation, while true that the GPU memory did max out in Watch Dogs, the way the GPU’s behaved in 2-way and 3 way, particularly with their load and clock rates, and in certain areas of the city, shows that memory is only 1 piece of the puzzle.

For people thinking about going 2-way or 3-way SLI, I found that a few tweaks such as the SLI rendering tweak mentioned earlier, and capping the fps at an acceptable level, proved a much more enjoyable experience, for instance, playing Seige of Shanghai in BF4 with 3 cards and the fps capped at 60, proved a much smoother result, eliminating almost all of the micro stuttering, the minimum fps was 57, that was when the tower went down, the rest of the time it never moved off of 60, this prevents the cards from wasting precious power and time rendering frames that will never be displayed on the monitors, as well as helping to keep them a little cooler. Finding an acceptable spot between min and average fps works very well. I hope this has shed some light on what really goes on with SLI scaling in games, and show that by analysis far more than just a couple of common factors, we get a much better picture of what is effecting the way multi card systems behave with different games.

Living in 104 Miller St, Pyrmont, NSW

Wed, 22 Oct 2014 19:33:32 +1100

My wife and I have been living in the apartment complex at 104 Miller st, Pyrmont, NSW for over 4 years now. We’ve split our time between a large 1 bedroom, and a 2 bedroom once our daughter came along.

thisisafold

The location is amazing, IGA across the road, post office, countless restraunts all within a few minutes walk. Darling Harbour is an easy 10min stroll and the city is a 15min walk. Busses to the Circular Quay and Central station are pretty much at the front door, as is the Light Rail stop.

The apartment has an indoor pool, spa, suana, gym and games room. Pool is solar heated, spa is heated. Gym has some old, some new equipment, and the weights room has what you’d need.

Internet is serviced by a few providers, you’ll get between 7000-8000down, 900-1000up. Water pressue is decent, though you’ll need to adjust your taps a couple of times during peak times.

The complex is reasonable quite, level 4+ has a nice view of the city and you have bbqs and entertaining areas both in the courtyard and on the roof.

Jekyll - Inline include SASS/SCSS to HTML

Mon, 20 Oct 2014 21:55:27 +1100

Playing around with some mobile first, basic design using Jekyll and Github for my new blog. I wanted to have the critical CSS inlined into the HTML for speedy delivery.

So I created a critical.scss file with my mobile first styles in it, and figured out the code below to include them in the head.html

thisisafold

{% capture critcss %}{% include critical.scss %}{% endcapture %} <style type="text/css"> {{ critcss | scssify }} </style>

The capture tag assigns the contents of critical.scss into the variable critcss.

Then I display critcss passing it through the scssify filter, which turns SASS into CSS.

Greasemonkey script for RealEstate.com.au - Hide previously viewed properties from map

Fri, 10 Oct 2014 22:28:00 +1100

Looking for a new house, and got sick of having to wade through the same listings every few days. So this script stops pins showing up in the map if I’ve already viewed them on the map.

Previously viewed properties are stored in your browsers local storage, so easiest way to clear is pasting the code below into the console while on RealEstate.com.au

localStorage.setItem('viewedProps', '');

Can get script at https://gist.github.com/sacah/49cb7073605513af76a1#file-hide-previously-viewed-properties-realestate-com-au-user-js

SaveMedia.dll causing issues with Playfire and CorsairLink

Tue, 30 Sep 2014 18:27:00 +1000

Just recently had this issue, programs crashing and reporting

Faulting module name: SaveMedia.dll, version: 1.38.0.0

in the event viewer.

Turns out it’s an issue with RivaTuner, I added both Playfire and CorsairLink to the application list in RivaTuner and choose ‘None’ on the ‘Application detection level’.

Seems to have fixed the issue.

New Pc x99

Sun, 14 Sep 2014 18:16:00 +1000

Specs:

Corsair 400r case
Intel i7 5820k
Crucial 8gb DDR4 sticks
ASRock X99 Etreme4 motherboard
Cougar CMX1000 V3 1000W psu
Corsair H100i CPU Cooler

There is a rather large heatsink at the top of the X99 Extreme4, this will stop you fitting 1 of the stock fans inside the case, I mounted them both to the top and don’t fully close the side.

thisisafold

I’ll get some thinner fans soon and setup a push/pull configuration.

Half the SATA ports are also blocked once I install my GPU, so when I add some drives later I’ll probably have to remove that.

The length of the main power harness of the Cougar is also just long enough to fit through the back and around to the motherboard, a few cm of extra length would have been a much nicer fit though.