Best $800 Streaming PC [Part 3]

Thanks again for tuning in with us.  We know the journey is arduous but with your interest and feedback, the sky’s the limit!  If you’re new to the party, you can find previous posts here.

 

custom-thumb

For this test, we felt a video would be more appropriate due to the large amount of benchmarking footage. If you cannot watch the video or would prefer not to, we’ve included the important data below.

Benchmark 2: CPU Selection

Now that we know what sort of video card we need, we will check to see what sort of CPU and platform we can use that won’t hold the video card back. Through some preliminary tests, we can easily eliminate anything below and i3 and AMD CPUs with less than 6 cores.

We selected 4 test platforms:

 

streaming platforms Before we start, we need to make clear that we actually have two performance goals here. One is that the PC be able to stream in a reasonable quality at a reasonable bitrate for twitch tv, and the other that the PC can record a high quality 1080p 60 fps video for those who want to create content and edit it.  

For each game we ran several benchmarks. We started by running an off-stream benchmark a few times in a row to see how repeatable the test was. This is very important because it hugely affects the accuracy of your final results.

csgo data*green text indicates the “winner” of the category, red numbers indicate the game was noticeably more difficult to play than off stream

In the first example, CSGO, we can use the game’s built-in replay to create an almost perfectly repeatable test. We saw a variation of less than 1% across the 3 test runs, so we know that our results are going to be directly comparable. Next we ran an off-stream test with each CPU to see how the game runs without any overhead. Here you can see Intel’s clear advantage, producing noticeably higher framerates than AMD.
After that, a test with 1080p 60 fps recording at 10000 kbps, first with CPU encoding, and then NVENC. With a very CPU bound game like CSGO, you can see the drastic difference. It is very important to note that while recording with OBS, the default settings will always try to give your game priority, so this awful choppy video you see is not the same as what you experience in game, indicated by the still rather decent 150-250 fps benchmark results. However, the effect is definitely noticeable to the player.

Last, we ran a similar test, but dropped our OBS settings down to a 720p 30 fps stream at 2000 kbps. This is representative of what a lot of up and coming twitch streamers will be using as their settings. Again, we compare CPU encoding to NVENC. Everything looks fine here, and there’s no significant bottlenecking.

bf4 data

Our next title is Battlefield 4. This is a non-repeatable test since we have to actually play in multiplayer to do this. We saw a 12% variation in the fps data between runs. Basically this means that one CPU could produce up to 12% lower or higher framerates, but we still have to consider them potentially equal because that is within our uncertainty bounds.

d3 data

dai data

gw2 data

t2 data

wow data

dota2 data

We normalized the data around the 4690k’s scores so you can see how everything stacks up in the end.

normalized data

After reviewing the benchmark data, we can draw some conclusions. The first is that the i5 CPUs tended to be better or equal to the AMD CPUs in off-stream testing. This is consistent with what we’ve seen time and time again in both internal and external reviews of these CPUs. In most games its neck and neck, with no one CPU showing too considerable a lead. However, the two source engine games, CSGO and DOTA2 definitely favor intel, with AMD’s performance actually trailing behind in a noticeable way, especially in DOTA 2.

Once we look at streaming and recording performance, we are not just comparing framerates, but the drop in framerate relative to the off-stream test. For the 720p test, the i5s and FX8350 did fairly well, but the FX6300 just could not keep up, showing double digit framerate drops in multiple titles.

In the 1080 tests, the 8-core 8350 did actually put up a good fight against the other CPUs for x264 encoding, but still could not handle CSGO or BF4. In fact, none of our CPUs could. We re-ran these tests with i7 CPUs, and it took until we got to the 6-core i7s that we saw good results. Even the 4790k at 4.5 Ghz could not handle BF4 with CPU encoding. The 6-core parts require X79 or X99 platforms, which cost hundreds more than what we are working with here. For 1080p 60fps recording, you just have to use some kind of GPU encoding method. NVENC produced perfectly acceptable videos, and with very little effect on the gameplay itself.

Benchmark 2 Conclusion

In the end, we settled on the i5 4460 as our CPU of choice, paired with this little H81 motherboard. It has everything you need and nothing you dont. Even though its small, you still get an x1 pcie slot under the video card, so you can add an expansion card if you need. The RAM slots will be fully populated, but we’ve yet to encounter a usage scenario where you’d need more than the included 8 GB of RAM. Another conclusion we’ve made is that this machine absolutley has to have an nvidia graphics card so that we can use nvenc for encoding. AMD’s graphics hardware supports the same feature but its not yet built into OBS, and we can only recommend based on what is available and works now.

Preliminary Streaming PC Spec

Now that we’ve nailed down our CPU and GPU, the rest of the components can fall into place. A 1 TB hard drive and 8 GB of DDR3 are standard equipment, ensuring the the system is never for want of storage or memory capacity. We observed no higher than 220W of power consumption during our testing so we are setting the PSU requirement at 400 Watts or higher.

CPU: Intel i5 4460
Motherboard: Asrock H81M-HDS
RAM: 2 x 4GB DDR3 1600
HDD: 1 TB 7200 RPM
GPU: Geforce GTX 750ti
ODD: DVDRW
PSU: 400W
OS: Windows 8.1

For those interested, here is a link to our configurator for this PC. We’ve set the default case to the Phantom 240 by popular vote amongst the IBP staff, though you can feel free to change it or any other component to your liking.

 

 

Comments

comments

by tyronewang, on January 6, 2015