Steam Machine Simulation Reveals Single- Vs Dual-Channel RAM Gaming Performance

Whatever you think about the Steam Machine, the fact is it's launching into a brutal market for anyone trying to sell a gaming product. So brutal, in fact, that Valve has been unable to source the 8GB SODIMMs that it originally planned to use in pairs in the system. As a fallback, Valve has had to ship at least some portion of Steam Machines with a single 16GB SODIMM instead of two 8GB modules.

This, obviously, won't activate the back half of the Hawk Point 2 processor's memory controller, leaving the system with just a 64-bit memory interface like an OG Pentium from 1993. Much has been made of this decision, with some in the media and gaming communities even insisting that Valve must walk back performance claims it previously made due to the system's new single-channel memory configuration. However, there's nuance here that bears more investigation.


And that's exactly what we set out to examine in a highly controlled, simulated test configuration. We don't have a real Steam Machine on hand—they're frightfully hard to come by, as you may have heard. However, we do have the next best thing: our Simulated Steam Machine from our article last year, faithfully rebuilt using the original components, with a fresh install of Nobara Linux and manually de-tuned performance to be roughly in line with an actual Steam Machine.

As before, there are a few things to be aware of. Our desktop platform, despite having the exact same silicon, is still not going to behave exactly the same as the mobile chip in the Steam Machine, even configured to the minimum 35W TDP. Likewise, while we've matched power limits and clock rates, the four extra compute units on our desktop Radeon RX 7600 are definitely helping out in GPU-limited workloads, so keep that in mind.

Even still, this is about as close as you're going to get without buying one of those sketchy Chinese graphics cards with a Radeon RX 7600M mounted to it. If you're keen to try building one of these yourself, here's the parts we picked for our specific build:
Now, without further ado, let's check out some benchmarks.

Simulated Steam Machine Single- Vs Dual-Channel RAM Benchmarks

It's true that in some application and production workloads, dropping from dual to single-channel can have an outsized effect on performance. However, DDR5 memory is fast, and games typically don't need that much memory bandwidth on the CPU side of things. The Steam Machine is explicitly a gaming system with a discrete GPU, so in theory, dropping to single-channel memory shouldn't have that much effect on performance... right?

We've tested seven games for you today, from a range of game engines, graphics APIs, and genres. We specifically tested the games we picked because of the wildly varying workloads they place on the system, but we'll talk about that in each game's commentary section. First up: Armored Core 6 Fires of Rubicon (because we're going in more-or-less alphabetical order).

Simulated Steam Machine Armored Core 6 Benchmarks

In Armored Core 6, swapping out our two DIMMs for a single 24GB module made almost no difference. While we ran this benchmark the customary three times to make sure we had lots of data and to help smooth out statistical variations (since this is a live gameplay test), there is still a lot of variability in this test, so we wouldn't put too much stock in the average FPS difference. It's there and it's consistent enough to show up in over 30 minutes of gameplay data, but it's not really conclusive. The 1% low FPS being the same is the tell; usually that's where you'd see the difference in single-channel memory.

Simulated Steam Machine Baldur's Gate 3 Benchmarks

We expected to see more difference in Baldur's Gate 3, considering the extremely CPU-limited nature of this benchmark. Indeed, our Radeon GPU was never more than 70% loaded in this test. However, while we do see a slight decline in performance, it's really not much. BG3 runs really well on Radeon GPUs in Vulkan mode. I was actually worried that BG3 might run better with a single DIMM than with two as our single-channel configuration has 24GB of RAM instead of 16, but that was thankfully not the case.

Simulated Steam Machine Cyberpunk 2077 Benchmarks

In Cyberpunk 2077, we tested with some pretty stout settings that include a lot of ray-traced effects. Note that while we started with the 'RT Medium' preset, we set all of the non-RT options in the options menu to 'High' manually, which actually drops a few options from 'Ultra'. With AMD's new FSR4 for RDNA 3, this game actually looks and runs incredible on this machine, which was surprising. Oh, but we're so GPU-bottlenecked that pulling a stick of RAM barely did anything.

Simulated Steam Machine Counter-Strike 2 Benchmarks

In Counter-Strike 2, we're actually bottlenecked on memory, so you'd think pulling a stick would matter a lot, right? Nope. That's because we're bottlenecked on memory latency, not memory bandwidth. Pulling out a stick of RAM had basically no effect; this result is essentially statistical noise.

Simulated Steam Machine Final Fantasy XIV Benchmarks

In Final Fantasy XIV Online's Dawntrail benchmark we do see a little bit more influence from slashing the memory bandwidth. This game uses a weird old engine, and it has strange performance characteristics, so nothing it does really surprises us that much. Of course, even with single-channel memory, you're still looking at 116 FPS average; not exactly a devastating drop with just one DIMM.

Simulated Steam Machine Hogwarts Legacy Benchmarks

Hogwarts Legacy is the game where we saw by far the biggest performance difference. This was exactly as expected. The reason is that, even on 'High' settings (instead of 'Ultra') we're still running right on the ragged edge of our GPU's 8GB of video RAM. The driver can compensate by feeding textures from system memory, but of course, that's highly dependent on the speed of said memory. By slashing our memory bandwidth in half, we take a huge 26% cut to our 1% lows resulting in a 1% low frame rate below 30 FPS, which doesn't feel great. With both sticks in this is a perfectly playable experience. It's still playable in a single-channel configuration too, but you will definitely feel the 1% lows.

Simulated Steam Machine Quake II RTX Benchmarks

Finally, in Quake II RTX we do see a tiny dip in 1% low frame rate. As we noted earlier, this is where you will usually see the effects of single-channel memory. The reason is that, while memory bandwidth is not normally a bottleneck, it can become one when the system needs to move data into or out of system RAM. Halving your memory bandwidth exacerbates that issue and causes more hitches and stutters in your games. That won't affect the average FPS much, but it does make the experience of playing worse, and that's reflected in the 1% lows. Still, this is a pretty small difference thanks to the heavy GPU load.

How Much Does Single-Channel Memory Affect Steam Machine Performance? 

With all that in mind, how bad does performance suffer overall? Well, the picture isn't all that bad. It's not great; obviously dual-channel memory is preferable, but honestly in the majority of the games we tested the variances basically down to statistical noise. And there is an upside to consider here as well. When Valve equips a Steam Machine with a single 16GB SODIMM, that makes it very easy to upgrade that Machine to 32GB of RAM. Is that really necessary or helpful for a system that's permanently limited to a Radeon RX 7600M graphics card? Well, maybe not for games but it certainly could come in handy if you intend to use the Steam Machine as a PC.

In any case, we find this tempest in a teacup over the Steam Machine being stuck with a single SODIMM to be much ado about almost nothing. A few games will be impacted somewhat, while most everything else runs just fine. Frankly, if you're picking up a Steam Machine at the final pricing, performance is probably pretty low on your list of priorities. Given that, we're confident in saying this single channel RAM situation really isn't anything to worry about, and it certainly doesn't seem worth the backlash, given the scarcity of 8GB SODIMMs and the reason Valve made the call in the first place. You can have a Steam Machine with single-channel memory or none at all: your pick. However, the numbers say you probably shouldn't concern yourself too much with the whole situation.