The world of integrated GPUs isn’t cut out for our usual suite of tests
FOR MOST OF OUR TESTING, we typically have a set list of benchmarks designed around being run on systems with dedicated GPUs. Most of the time, that involves running X game at Y resolution— usually on the highest graphical settings profile the game comes with as default. That way, it makes it easy for you at home to replicate those same tests with your own machine. It also gives you a better idea of how your rig scores in comparison and, of course, gives us a level playing field to work with when it comes to carrying out any similar testing in the future.
The problem with that is, in the world of integrated graphics, those kinds of benchmarks are grossly unfair to small integrated GPUs, such as the ones featured in these two builds. Yes, integrated graphics have come a long way over the years, and they are far from being stuck operating as media PCs and home office machines today, but throwing Total War:WarhammerII at an iGPU at 1080p Ultra is just cruel. Nor will it give us an accurate representation of how this chip would perform under more reasonable and realistic circumstances.
TotalWar is perhaps the best example of a CPU-intensive title from our testing suite. It’s a game we choose specifically because it stresses the processor during regular benchmarking. In fact, most games (particularly single-player titles) generally leverage most, if not all, of the stress onto the GPU, rather than the CPU, or both. Strategy games, however, with their many thousands of individual units, and some online games too, really do push processors hard.
Ultimately, you can have all the highspeed memory in the world, but it’s just not going to be enough to manage Total War at 1080p Ultra on an iGPU, no matter what. And that’s often reciprocated in our benchmark figures for builds like this. It’s not unusual to see scores of 5 or 6fps in these tests on our usual platform and, obviously, that won’t do.
So, for this feature, we have decided to shake things up and do something a little different with the benchmarks. This time, we’re going to be testing a selection of games at 1080p specifically, but across three separate presets instead, Low, Medium, and High, no Ultra in sight. The games are still challenging, and still comparable to the ones we regularly use in our usual suite, but it should give us a better indication of what you can expect from your iGPU PC. But we’re not stopping there either! Oh no.
GAMING PERFORMANCE
First up on the chopping block are the game benchies. In this case, we’re going to be testing TotalWar:Warhammer II, MiddleEarth:ShadowofWar, and AssassinsCreed:Valhalla (all the colons) at 1080p low, medium, and high presets.
This gives us a nice mix of titles to draw upon, including an older, neutral AAA game from 2017, in the form of Shadowof War, a brand new AMD-optimized AAA title in Valhalla, and an Intel optimized strategy game in the shape of TotalWar. All with included easy-to-run benchmarks.
On top of that, we’re also going to be performing some more “old-school” synthetic tests, namely 3DMark’s Fire Strike and Time Spy tests at 1080p. These will provide us with an index we can refer to, to see how far we’ve come from generation to generation, and also let us test performance both with DX11 (Fire Strike) and DX12 (Time Spy).
COMPUTATIONAL PERFORMANCE
With gaming out of the way, it’s also important we take a quick look at the synthetic computational element of these two builds too. Both offer vastly different architectures and performance figures because of it. To do this, we’ve grabbed
some basic, easy-to-run, synthetic tests that you can use yourself at home.
For processing performance, we’ll be using CineBench R20 (you can download it yourself for free from here: bit.ly/ Cine20MPC) to test out our processor performance, taking advantage of both its single-core and multi-core metrics.
And finally, to test out our SSD performance, we’ll also be taking a quick look at CrystalDiskMark 6 ( bit. ly/CD6MPC). This should give us an accurate representation of how our two SSDs perform under pressure, namely looking at Sequential read/write speeds at a queue depth of 32, 1 thread, and also Random 4K read/write speeds at a queue depth of 1, 1 thread.
Sequential performance is indicative of reading or writing files that are stored close to one another on the drive itself (for instance, your photographs, videos, and media files, things like that), whereas Random 4K testing is more indicative of something like a game reading and writing to a variety of different locations on the drive at any one time. This is almost always more difficult to do than with sequential data, thus the far lower numbers in the tests.
As neither of these systems was really cut out to be video-rendering machines, workstations, or anything of the like, due to a lack of dedicated GPU, which massively accelerates these workloads, we’re going to skip out on any memory testing or power-draw this time around.
So, ultimately, what we’re left with is purely an exercise to see which integrated graphics solution is superior to the other.
OH BOY, did we get some results in this test. It wasn’t necessarily what we had hoped for, but this is a test of the integrated graphics on each side of the battle, AMD and Intel. In an ideal world, we would accompany these CPUs with some hefty GPUs that would improve performance, but it’s a test of iGPUs, not discrete GPUs.
On the AMD side of the head-to-head, the build itself was plain sailing to piece it all together, and yet the opposite once we tried to boot things up for the first time. With nothing but a blank display, we updated the BIOS (which was fairly easy as our ASUS mobo has a BIOS FLBK port on its I/O) and then used a USB Windows 10 installer to get it ready for testing.
Once the machine powered on, we then installed the relevant drivers for the motherboard and the chipset to get the AMD APU up and running properly, or so we thought. After some initial testing, our results weren’t what we expected. Taking a peek at the memory using CPU-Z, it was clear the DDR4 was running at 2,666MT/s, not the 3,600 it’s specced at. Jumping into the BIOS, we realized we’d forgotten to enable the XMP profile, so the speed and timings were out (rookie error!). To do that, we set the Ai Overclock Tuner profile to the DOCP option. This is AMD’s alternative to XMP (eXtreme Memory Profile) usually found on Intel motherboards.
For AMD, with Asus motherboards, it’s located inside of the Ai Tweaker menu. This should enable the XMP settings found on the RAM (increasing timings, frequency, and voltage in the process) and tell the motherboard to use the full speed the kit is capable of running at. RAM frequency matters to AMD chips, and for our PC to work at its best, it needs the full 3,600 MT/s that our Crucial Ballistix RAM should be providing.
Unfortunately, after applying the DOCP profile, the machine kept rebooting and failing to post. So, we headed back to the BIOS and declocked the frequency down to 2,800 MT/s, to see if it would boot at a higher speed than the original stock settings outlined by JEDEC. Thankfully it did, but this was still an issue as we needed the memory speed to be working at its fastest. We kept restarting the PC and booting it up with a higher memory frequency each time to see what would stick but, unfortunately, the best we could get out of these sticks was 3,133 MT/s. It’s not awful, but we didn’t choose 3,600 MT/s sticks to underachieve like that, and these APUs love faster memory.
The only other memory we had spare were some 3,200MT/s G.Skill Trident Zs. Out of curiosity, we looked at whether we could get more performance out of these. Again we could only get a stable boot when configured to 3,000MT/s, so Crucial’s offering still works better here.
Results were lackluster compared with what we’re used to seeing from a discrete GPU, of course, but that’s to be expected. After getting around 30fps on low presets on some games, our hopes for smooth 60fps, 1080p gameplay were dashed. The benchmark results speak for themselves. Ideally, you don’t want to settle for a low preset, but anything less than 30fps isn’t acceptable by today’s PC gaming standards. SSD read and write speeds, however, were a high point of this build and for normal day-to-day usage, so while it works pretty well as a daily driver, it’s not so good if you intend to do any highend gaming.
Lacking a GPU and RAM sticks that run at their full potential hold this machine back from achieving a smooth Full HD gaming experience. If we had sticks that were fully compatible with our motherboard, we may be looking at a different outcome (reports even suggest up to 10-15% additional performance).
Still, the results were much higher than our Intel competition, and in older games or less graphically intense titles, such as Minecraft, Hades, or Roguelikes, it’s an ideal PC to get your foot in the door. It just depends on what you play. Mainstream games, yes. AAA titles, no.