Athlon X4 845 CPU Benchmarks
Cinebench is a good tool to measure differences across generations, and certainly can show us not just the improvements between those generations but also between the competitors. The Intel G4400 was sourced from a dear friend, and is trusted though not something that I ran myself.
When looking at the results we see a great improvement in both single and multi-threaded workloads. The smaller L2 cache doesn’t seem to hinder its performance in this case. Crucially, though the Intel has a higher single-threaded score, it falls slightly short when we add in more cores. That’s, of course, due to the Pentium only having a maximum of two-threads.
The next benchmark is that of x264 FHD, a test of how fast we can encode a video using the x264 codec. This is indubitably a realistic benchmark because it’s a task that most of us will inevitably do at some point with our PC. This is one task that generally appreciates more cores and a high IPC per thread.
Measured in FPS, Excavator is suitably faster than Steamroller. It’s capable of converting a video in in a few minutes and reaching a good speed. Not quite comparable, though the G4400 seems to be able to render a video through Handbrake at a speed of around 13-15FPS on average. Thus, this the two extra cores are advantageous. Again, the lack of L2 cache doesn’t seem to be an issue for the “lowly” Athlon X4 845 when compared to the 860K.
Here we take a look at how well and efficient the X265 codec is utilized by Excavator. It’s a relatively new codec that isn’t seriously used by too many denizens of technology. In fact, it isn’t wholly supported by all programs even yet, despite being more efficient and overall a better choice.
There’s a decent increase in performance, though it’s not mind-blowing. The real improvements come in being able to perform this work while using less power, which we’ll see shortly.
We all use browsers for an increasing amount of work, so it makes sense to take a look at the most common benchmarks built specifically for browsers; Google Octane and Mozilla Kraken.
There’s an increase in performance, a slight but significant increase that show that the efficiencies inherent in Excavator combined with the new instructions seem to be playing to its advantage.
The next benchmark is a special one that I think represents future workloads that many individuals and especially scientists will appreciate. I’ve used Microsoft’s CNTK deep neural network to test how fast this CPU can sample through and learn a series of convoluted images. This benchmark can be run by anyone, all you have to do is download the framework itself (it’s one of a few that works in Windows) and simply run the convoluted image sample included with the framework. It reports samples per second.
What we see here is that it’s not exactly the strongest when sorting through and recognizing images, though the differences and inclusion of AVX2 actually helps to increase performance appreciably, despite the lack of L2 cache compared to its sibling. It’s not the ideal DNN processor, but it certainly is more than capable of testing out DNN capabilities. If you wanted to break into the AI world, this will work great. It won’t break any speed records for recognition, but it doesn’t have to for the price. This isn’t even meant to compete at a higher level, and for the price it’s very fast.