Do you still need a graphics card in 2026? Did you need one in 2018? Can this cheap old APU handle all your gaming needs, making these newfangled expensive ones redundant? Welcome to Esk Computers, I’m Scott and today on the desk of Esk, we’ll confirm the, frankly, pretty obvious answers to these questions, but it’s the journey, not the destination, right?
When troubleshooting some previous issues I’d wound up picking up a Ryzen 5 2400G, and after recently re-examining the performance of my laptop with a similarly benumbered Ryzen 5 2500U, I thought it might be interesting to see what similar silicon can do with the additional power and cooling capacity of a desktop PC.
I’d also been thinking of taking the recent small form factor ITX builds to another level of miniaturisation, and what better way to do that than by removing the graphics card? Surely an excellent idea for a gaming machine. So let’s smash those ludicrous ideas together and see if we can build a console sized PC on the cheap that’s still capable of playing games well enough in space year 2026, or alternatively find out if I’m just expecting too much from an eight year old budget APU and should instead just go buy a Playstation.
Of course, in an ideal world where money was no object we’d want something like an AM5 board with the Ryzen 7 8700G and it’s much more powerful 780M RNDA3 based graphics, or if AMD ever make it, some theoretical desktop form of the AI MAX Bunch Of Numbers EX Plus Alpha Or Whatever with the amazing looking 8060S. However I’m neither made of money or able to conjure non-existent chips from the ether, so we’re stuck with what’s on hand.
The 2400G is a four core, eight thread part with a TDP of 65W, so it can cope with more than double the power draw of the laptop 2500U. It should also be clocking slightly higher, with a turbo speed of 3.9GHz and base clocks of 3.6Ghz over the 2GHz and 3.6GHz of the laptop part.
We also get a few more of the Radeon RX Vega cores, 11 vs 8 on the laptop, with slightly higher Turbo Frequencies of 1250 MHz versus 1100MHz. Will this be enough to push some of the more modern games in the test suite into the realms of playability, or will they remain a stuttery slideshow?
We will find out soon, but first a quick look at the system we’ve built for this. This is a Metalfish T10 case and I quite like the look of it, even if it’s a not much more than a metal rectangle. It’s a mere 4.8 litre capacity, although why case manufacturers list case sizes in litres I do not know. I’m not going to fill it with Pepsi. In more useful measurements, it’s roughly 24 cm x 21 cm x 9 and a bit cm tall. It’s got front panel USBs 3 in A and C flavours, although both go back to the same Gen 1 header. There’s no front panel audio.
It should support CPU coolers up to 70mm tall, so this 67mm DH670, also from Metalfish should do the job. We round off the Metalfish expo with this 600W Flex power supply unit, which claims to be 80+ Gold rated. I’m adding this 512GB CUSU NVMe drive for storage – the case can fit a 2.5″ drive too, but I’m fresh out of those. There’s the wifi/bluetooth card, a pain to install as always. The RAM is this mystery unmarked kit of 32GB DDR4 that I took a gamble with on eBay. Integrated graphics love fast RAM, so “What speed will it run at?” is a critical question. The suspense is killing me.
The motherboard we’re plugging all this into is this JGINYUE B350i Plus ARGB, which I’ve also used on a few previous builds on the channel. They’re cheap and they work, so why not? I’ll install Windows 11 to keep it the same as the laptop, although Bazzite or similar would give a better lean-back, big-screen, console-like experience.
Turns out the memory has an XMP profile of 3600Mhz, which would be nice, nicer than this CPU can handle I think, and a latency of 18, which could be better. Integrated graphics in particular want fast, low latency memory. However, that didn’t boot, and although a bit of manual tuning later get us 3200MHz with a latency of 16 booting and passing Prime95 tests, unfortunately, the computer is haunted.
Essentially running either the memory at anything above it’s default JEDEC speed of 2666MMz or doing any sort of tweaking to the CPU in Ryzen Master will result in this corrupting the display drivers and other generally phantasmagorical errors that need a Windows reinstall to properly fix. In the absence of an exorcist, I’ve just settled on running this at stock settings and seeing what we get with this combination of parts that clearly don’t get on with each other.
Where relevant, I’m comparing here with the results from my laptop, as detailed in a previous video, and also the recent budget X79 Xeon E5-2450 V2 build with an RX480, which we’ll see later came in at a similar price to this build.
The Cinebench 2024 results are on the lower side of my expectations, with a single core results of 56, and what seems a particularly low 259 multicore result. In some of the faster but unstable configurations this went to it crept up to 273 points, but still lower than I’d expect. Such is life.
3DMarks next, with Timespy getting us 1238 points, so a good 51% faster than the laptop on the same test, although something like a quarter of the results even an ageing RX480 would get you as we saw in the recent X79 Xeon build. It’s a broadly similar story in Fire Strike, getting 2990 points, about 40% more than the 2500U’s 2149 points, and again roughly a quarter of my budget X79 build.
Deja vu sets in with Superposition, getting 1491 points at 1080p High settings, about 43% higher than the laptop’s 971 points, and quarter of the RX480 result.
Some actual games now, with Shadow of the Tomb Raider at our normal 1080p High Settings getting 15 fps, for what it’s worth 87% faster than the laptop’s 8 fps result but obviously a long way from playable, and only 21% of the RX480/Xeon system’s 69 fps. If we drop down to 720p low, we get a borderline playable 33 fps, again about 43% faster than the 23fps the laptop can provide.
The Counter-Strike 2 benchmark appears to have been fixed, and is giving us 25 fps at 1080p Very High Settings, or more realistically 75fps at 1080p Competitive settings, those being Low, but with high Shadows. That’s actually a good bit faster than the laptop, 90-odd% faster at the higher settings result of 13fps and the laptop could only get up to 55fps by dropping the resolution to 720p at Competitive settings. Not a terrible showing in comparison to the RX480 either, hitting about a third of its 76fps Very High result and about 63% of 119fps Competitive settings result.
Surely Cyberpunk 2077 is too much for this? Well, yes, certainly at the 1080p Ultra settings in-built benchmark I run here, where its 10fps is a brave effort, and two and a half time the laptops 4 fps result. I think having access to more RAM really helps out in an actual game settings, although of course not a playable result. When the laptop was dropped to 720p Low settings, it gave 14fps, but the 2400G manages 31 fps on the same test, more than twice as fast an tantalisingly close to being playable, surely?
Well, not exactly, as this short run around Kubuki at 720p Low but with High crowd density shows, particularly while causing a bit of a ruckus where it drops to an unpleasant 23 fps. Lowering the crowd density might help a little, but it’s just asking a bit too much of the eight year old processor.
If Cyberpunk’s too much for it, the Unreal Engine 5 contingent surely will be too. Black Myth Wukong’s benchmark at 1080p High, with no Framegen or upscaling gets 6 fps, technically 50% faster than the laptops 4 fps on the same test. It also improves, technically, at 720p Low settings and FSR applied, getting 18fps, but even if you were to add framegen to the mix, that’s still not playable.
Marvel Rivals tells a similar story. At 1080p Native resolution, Medium settings, it gets 6 fps, three times faster than the laptop but, y’know, single digits here. At 720p Low Settings with Performance FSR3, it struggles it’s way to 26 fps, twice as fast as the laptop’s 13fps but by no means playable, whereas the RX480 can get a more or less playable 55fps result at 1080p Low with FSR applied.
Forza Horizon 5 should provide some better results. While the benchmark run at 1080p Medium only nets 26fps, that’s over twice as fast as the laptop’s 12fps at the same settings. Reducing down to 720p Lowest settings gets a playable 53fps, about 82% faster than the laptop’s 29fps at the same settings. Of course, miles behind the RX480 which can get 63 average fps at 1080p High settings.
I think I hoped for slightly better results from the Assassin’s Creed Mirage benchmark, which gives us 27fps at 1080p High Settings. Dropping to 720p Low gives us a 31 fps result, so maybe on the borderline of playable compared to the laptop’s 22 fps, depending on your standards.
We’re in much the same space with Red Dead Redemption 2, where are 720p favour performance settings we’re getting 31fps, about 72% more than the laptop’s 18 fps. The RX480 system gives playable 75fps results at a higher 1080p resolution favour performance settings.
Horizon Zero Dawn also struggles up to borderline playability at 34fps at 720p Favour Performance Settings, up from 19fps at 1080p. That’s way better than the laptop’s 18fps at 480p, so I expect it was choking due to a lack of memory. Again, the RX480 gives playable 52fps results at 1080p Favour performance settings.
A few results to hammer the CPU next, starting with Total War: Warhammer III at 1080p Ultra settings in the inbuilt Mirrors of Madness map benchmark getting 8 fps. At 720p Low Settings, it gets a more playable, but really ugly 26 fps.
Civilization VI’s Gathering Storm AI Benchmark at 1080p highest settings gives an average turn time of 72 seconds, so maybe a bit painful in the end stages of a game, but it is working.
And rounding things off with off with Baldur’s Gate 3, where a walkaround the Lower City in Act 3 at 1080p Ultra gives us a 7.9 fps average. This does increase to 19 fps at 720p Low settings, which is arguably playable enough given the sort of game it is, depending on your standards.
What does all this mean? Well, to answer my initial question, the 2400G is getting something like an extra 40-50% performance over the laptop’s 2500U APU in situations where there’s enough RAM on both systems, and to somewhat unfairly compare, two to two and a half time the performance where the laptop ran up against the limits of its 8GB of RAM.
I’d struggle to call this a proper solution for modern, or modern-ish gaming, even if you were happy to run it at 720p. For older games, e-sports titles and other well scaling games like Minecraft, Roblox, Fortnite and the like, it might still have a place.
To truly judge if it’s worth it we need to consider the price. In this case we’re coming in at a little over £200. Roughly, the CPU was £25, as was the case, the motherboard was £50, the RAM £20, the SSD £25, the PSU £30, the Cooler £20 and the Wifi card £5.
I do have to add the disclaimer that these components were bought before the advent of the rampocalypse, and I doubt I’d get as good a deal on particularly the RAM and SSD today.
Comparing that to the £175 that the X79 Xeon E5-2450 V2 build with an RX480 came in at, if we’re only comparing raw performance then clearly it’s an absolute blow out in favour of the machine with a dedicated graphics card. What a shocking revelation. Of course, a small form factor machines always have compromises and usually higher costs, but it does result in a very different desktop aesthetic.
This unassuming little box doesn’t scream gamer in RGB capitals, and in previous decades would have looked at home underneath a TV serving media and playing the odd light game, but these days that’s a role mostly usurped by PlayStations, Xboxes, and Amazon Fire Sticks or built into the TV itself. This is the sort of build that fits into an evolutionary niche that’s shrinking to the point that it might actually be extinct. So, while I quite like this little fella, I’d be struggling to defend it on value for money charges.
What can we do to improve performance without breaking the bank? Something like a Ryzen 7 5700G would be faster but not, I think, transformative. Installing Bazzite over Windows might eke out a few more frames in some titles, but probably at the expense of slightly worse performance in other games.
There is maybe just enough space to cram a single slot, half height GPU in. Hmm. I suppose that’s another video idea sorted then.
Well, maybe you’ve not seen the last of this, but if you have any questions or want further details please leave a comment down below, and if you enjoyed this videotronic missive then consider subscribing. Until next time, take care of yourself, and each other.