Just to be clear, this is testing the same handheld on both Steam and Windows and is in line with previous findings on a small set of AAA games.
Best guess, as someone who runs both Linux and Windows on both handhelds and desktop gaming PCs, the issue here is probably memory and driver optimizations around them. Windows is just heavier than SteamOS and, while the 32 GB in the Legion Go should be enough for at least some of these tested games, they are shared between CPU and GPU. I don’t have a Go S, but I’ve seen significant performance improvements on Windows handhelds by manually assignign more VRAM in heavy games like these.
Shame, I’ve been waiting for more thorough testing (more games, desktop hardware references and a deeper look at memory management in Windows, but this is pretty superficial still.
EDIT: For what it’s worth, and I DON’T have the time or the setup to do a full set of benchmarks, but running South of Midnight on both Linux and Windows, same settings, same PC, just dual booting I got almost 2x the fps on Windows. That’s suspicious the other way, I’d expect the difference to be less dramatic, so there may be some resolution stuff going on here. Or perhaps the DLAA I’m running on both runs slower on the Nvidia Linux drivers? I’ll give one more game a try with no DLSS before I call it an experiment.
EDIT 2: Damn, this is why benchmarking modern games sucks. I tried Marvel’s Midnight Suns (just because it was there on both) and… well, the performance is the same on both, but Windows is clearly bugged and stutters for like a second every couple of seconds, consistently. So it’s really nice on Linux but entirely unplayable on Windows (on this machine, at least).
If I’m learning anything from this is that despite modern advances PC gaming is still a tinkerer’s game and that I really wish Linux/Windows drive sharing was less flaky because it’s increasingly obvious that dual booting is a great tool for gaming, given how temperamental modern big games are.
Ive had monster hunter world bsod, crash and lag on windows while it runs perfectly through compatability layer on an os not designed for gaming using a gpu it doesnt actually have first party drivers for.
Windows has stopped being “everything just works out of the box instantly” os for a while now.
That’s what I was saying, it’s all shaky right now. Wilds runs about as well on both, but it’s noticeably less stuttery for me under Linux. Other stuff, particularly when leaning hard into Nvidia features, is either performing poorly or has features disabled on Linux. Plus the compatibility issues.
There is just no one-size-fits-all solution on PCs thede days, even before you start considering the weirdness of running the same games in ridiculous 1000W powerhouses and 15W handhelds at the same time.
PC gaming has become a LOT less plug-and-play this last decade, and I don’t know that it’ll go back to where it was any time soon.
Drivers and “other stuff” have more impact than the OS itself. I would expect if you installed Windows 11 from a USB stick onto this device that it probably puts performance into “balanced” mode for example, fires up antivirus/malware protection, runs a bunch of esoteric services, throws in a WHQL (stable but crappy) GPU driver etc.
I think the article would have been fairer and more useful to install Windows, and optimize the life out of it and then compare performance and other factors (e.g. battery, heat, fan noise etc.)
That depends. In this case, where the Lenovo drivers are clearly outdated and kinda broken, definitely they’re the bottleneck for at least some games. That much they’ve shown, by installing newer drivers and showing a massive performance upgrade.
Although I’d caveat that by saying that their flashier results with big updates across OSs and driver variants are running at outright unplayable settings. They are benchmarking on settings resulting on framerates in the teens. When they say they saw 12% performance increases on the newer drivers they mean going from 14 to 16 fps in some cases.
Benchmarking properly is hard, I guess is my point.
It is in some ways. I can tell you I tried to run Prototype 2 on a handheld today and it didn’t run natively on Windows 11 because it’s old but putting it into a Proton session and keeping it contained did wonders for it and the Deck ran it maxed out at 90fps (you forget it can do that if you insist on playing modern games on it, but man, does it look nice on the OLED).
So hey, it certainly Windows 8s better than Windows 11. There is that.
But it’s not magic, so I’d still like to figure out what we’re seeing in these examples.
Ever found a way around Lutris asking for a CD for games? I was using Lutris and one of the games I tried installing from a mounted ISO installed, yet I can’t find any way to get Lutris to recognize the mounted drive as the CD. Tried adding it to Steam as a non-steam game as well and get the same result. Tried various versions of proton and wine, but I assume I need to direct it to the ISO somewhere… But couldn’t find anyone who had an answer online.
I haven’t tried, sorry. I use Heroic rather than Lutris for my non-Steam digital libraries and I haven’t messed around with older physical releases too much, so I don’t know what Lutris is expecting. Maybe someone else here can help?
Heroic is very straightforward, as long as what you want is access to your GoG, Epic, Amazon and Battle.Net libraries. Lutris is meant as a more general purpose launcher, so they’re aiming at slightly different use cases that overlap.
Heroic won’t solve your Lutris ISO problem, but if you want to play some non-Steam ways it works great, is easy to use and is very Steam-like.
1: SteamOS don’t run unnecessary services in the background. (especially stuff like print services and other random shit)
1b: Even regular Linux, which does run a bunch of extra services, still generally has less overhead because it’s still being optimized for lighter weight systems, and it idles more efficiently too. Meanwhile Windows doesn’t have a good way to tell your printer driver and its corresponding services to shut up when you’re gaming.
2: Antivirus programs
3: Drivers, graphics system. This is both a plus and minus, but for performance mostly plus. More efficient driver model, less overhead again. Sometimes the performance comes from lacking features which doesn’t get executed fully, though. Sometimes it comes from translating to Vulkan, because DirectX has some more overhead (and in these specific cases you can get the same performance boost on Windows by switching to Vulkan).
I mean, this is wrong. The CUPS daemon literally is a print service and does exist in Linux. It’s just socket based so tends not to use resources until asked.
The CUPS daemon does not get killed during games. It wouldn’t be needed anyways
Antivirus programs? When was the last time you tried Windows, the mid-00s?
Anyway, it’s not random print services causing CPU overhead, that’s old timey stuff. In this case it’s being RAM heavy in a RAM-limited scenario and, from their testing, Lenovo being really terrible at keeping their AMD Windows drivers updated. As part of the test they manually switched to an ASUS version of newer AMD drivers and saw significant boosts in some games.
Modern graphics drivers are a mess of per-game features and optimizations. Different manufacturers keeping things at different levels of currency is a nontrivial issue and why some of this benchmarking is hard and throwing five random games at the problem doesn’t fully answer the question.
All modern versions of Windows will have Microsoft’s Defender antivirus/malware protection turned on by default. That means you incur a penalty every time a file is accessed from disk, or a process is launched, or a library loaded, or sockets are used or certain APIs are called.
It’s better than most 3rd party AV software but it’s still a performance overhead that could be turned off.
I mean… you can turn it off. I wouldn’t, but you can.
I just haven’t heard it referred to as “antivirus programs” in ages, it sounds so 20th century to me. Say what you will about MS’s monopolistic tendencies, but at least they killed the parasitic “antivirus” industry with that one.
I think Windows Defender is a fantastic line of defence and it’s definitely better than installing garbage from Trend, McAfee etc. That said, Lenovo, HP, MSI, Dell etc still preinstall crapware on their new machines from Norton or their ilk threatening that my machine is “at risk” if I don’t pay them money.
I wouldn’t turn it off unless I knew that I was only installing games from Steam. But if I did I think performance would improve. A game from Steam could still contain malware so you have to exercise some common sense. Even on SteamOS a game could be malicious but since its containerized the scope for damage is limited but not necessarily impossible to break out.
Nope, the S is 16GB on the base models. It uses either the same Z1 Extreme chip as the non-S or the Z2 Go which is only 4c/8t, and is only 16GB with the latter. I think the former, which just released, has 512/16GB and 1TB/32GB variants.
As for memory sizes, it really depends on the chip. I think the Z1E would likely breathe a bit easier with 24GB but I think 32GB is kind of overkill for it. 16GB is likely plenty for the Z2 Go, and 24GB on the Z2 A in the ROG Xbox Ally is definitely overkill as that’s just AMD finally releasing a SKU for the Steam Deck APU, which is two to three generations older than every other Z2 chip.
The Z2, Z2 Go and Z1E all have only 12 graphics cores and that’s why I feel like 32GB is overkill for them, whereas the Z2 Extreme has 16 so it actually makes sense to dedicate more than 8GB of memory to it.
I think you’ve got to remember that “more memory better” doesn’t always hold true on a device like a handheld as it does include a cost on battery life even at idle.
As of right now, both models of the Go S listed on Lenovo’s website have 32 GB of RAM (screenshotted below, if the weird screenshot functionality here works). So no, you’re wrong here. The version with 16 GB is the Go 1. If there is a 16 gig SKU of the Go S, which there may be, they currently don’t have it listed.
Memory size requirements depend on what you’re trying to run. Easier to run stuff will run on everything, but from hands-on experience I assure you a bunch of newer games struggle with the default allocation of 4 gigs of VRAM and can use the extra RAM. You can still give 8 gigs to the GPU with 16 but then you’re a lot more likely to start struggling with system RAM. If these AMD APUs worked like an Apple chip and could dynamically allocate RAM that wouldn’t be such a pain, but at the moment you need a reboot to change this even on current-gen hardware, so it’s easier to have a larger pool and give the GPU a little too much.
The amount of CUs and the VRAM aren’t necessarily related. Even with larger RAM allocations and weaker GPUs you can find yourself in the wrong setup, which is annoying. And it’s not just amount of RAM, these shared architectures can struggle with bandwidth as well, so speed can matter (although it’s more giving you more or smoother FPS and the less the fall-off-a-cliff unplayable mess you get if the game is entirely out of RAM budget). That’s also why I suspect being lighter on memory and perhaps having a better default setup may be a part of why SteamOS performance is disproportionally better on heavier scenarios compared to what you see on desktop PCs. I can’t be sure, though.
This comes from me messing around with a literal handful of PC handhelds on Windows, SteamOS and Bazzite. I’m not guessing, I’m telling you what happened during hands-on testing.
Either different SKUs per territory or one of the sites is straight up wrong. FWIW, cross checking with Amazon shows the Z2 versions of the Go S having 16 GB and the Z1 Extreme versions with 32 GB for my territory.
All else aside, man, is the world of PC handhelds the wild west right now. Everything is on effectively the same two or three pieces of hardware, but somehow nothing is consistent.
Just to be clear, this is testing the same handheld on both Steam and Windows and is in line with previous findings on a small set of AAA games.
Best guess, as someone who runs both Linux and Windows on both handhelds and desktop gaming PCs, the issue here is probably memory and driver optimizations around them. Windows is just heavier than SteamOS and, while the 32 GB in the Legion Go should be enough for at least some of these tested games, they are shared between CPU and GPU. I don’t have a Go S, but I’ve seen significant performance improvements on Windows handhelds by manually assignign more VRAM in heavy games like these.
Shame, I’ve been waiting for more thorough testing (more games, desktop hardware references and a deeper look at memory management in Windows, but this is pretty superficial still.
EDIT: For what it’s worth, and I DON’T have the time or the setup to do a full set of benchmarks, but running South of Midnight on both Linux and Windows, same settings, same PC, just dual booting I got almost 2x the fps on Windows. That’s suspicious the other way, I’d expect the difference to be less dramatic, so there may be some resolution stuff going on here. Or perhaps the DLAA I’m running on both runs slower on the Nvidia Linux drivers? I’ll give one more game a try with no DLSS before I call it an experiment.
EDIT 2: Damn, this is why benchmarking modern games sucks. I tried Marvel’s Midnight Suns (just because it was there on both) and… well, the performance is the same on both, but Windows is clearly bugged and stutters for like a second every couple of seconds, consistently. So it’s really nice on Linux but entirely unplayable on Windows (on this machine, at least).
If I’m learning anything from this is that despite modern advances PC gaming is still a tinkerer’s game and that I really wish Linux/Windows drive sharing was less flaky because it’s increasingly obvious that dual booting is a great tool for gaming, given how temperamental modern big games are.
Ive had monster hunter world bsod, crash and lag on windows while it runs perfectly through compatability layer on an os not designed for gaming using a gpu it doesnt actually have first party drivers for.
Windows has stopped being “everything just works out of the box instantly” os for a while now.
And nothing has replaced it.
That’s what I was saying, it’s all shaky right now. Wilds runs about as well on both, but it’s noticeably less stuttery for me under Linux. Other stuff, particularly when leaning hard into Nvidia features, is either performing poorly or has features disabled on Linux. Plus the compatibility issues.
There is just no one-size-fits-all solution on PCs thede days, even before you start considering the weirdness of running the same games in ridiculous 1000W powerhouses and 15W handhelds at the same time.
PC gaming has become a LOT less plug-and-play this last decade, and I don’t know that it’ll go back to where it was any time soon.
Drivers and “other stuff” have more impact than the OS itself. I would expect if you installed Windows 11 from a USB stick onto this device that it probably puts performance into “balanced” mode for example, fires up antivirus/malware protection, runs a bunch of esoteric services, throws in a WHQL (stable but crappy) GPU driver etc.
I think the article would have been fairer and more useful to install Windows, and optimize the life out of it and then compare performance and other factors (e.g. battery, heat, fan noise etc.)
That depends. In this case, where the Lenovo drivers are clearly outdated and kinda broken, definitely they’re the bottleneck for at least some games. That much they’ve shown, by installing newer drivers and showing a massive performance upgrade.
Although I’d caveat that by saying that their flashier results with big updates across OSs and driver variants are running at outright unplayable settings. They are benchmarking on settings resulting on framerates in the teens. When they say they saw 12% performance increases on the newer drivers they mean going from 14 to 16 fps in some cases.
Benchmarking properly is hard, I guess is my point.
Maybe wine/proton is just better at Windowsing than Windows is.
Very unlikely that has anything to do with it.
It is in some ways. I can tell you I tried to run Prototype 2 on a handheld today and it didn’t run natively on Windows 11 because it’s old but putting it into a Proton session and keeping it contained did wonders for it and the Deck ran it maxed out at 90fps (you forget it can do that if you insist on playing modern games on it, but man, does it look nice on the OLED).
So hey, it certainly Windows 8s better than Windows 11. There is that.
But it’s not magic, so I’d still like to figure out what we’re seeing in these examples.
Ever found a way around Lutris asking for a CD for games? I was using Lutris and one of the games I tried installing from a mounted ISO installed, yet I can’t find any way to get Lutris to recognize the mounted drive as the CD. Tried adding it to Steam as a non-steam game as well and get the same result. Tried various versions of proton and wine, but I assume I need to direct it to the ISO somewhere… But couldn’t find anyone who had an answer online.
I haven’t tried, sorry. I use Heroic rather than Lutris for my non-Steam digital libraries and I haven’t messed around with older physical releases too much, so I don’t know what Lutris is expecting. Maybe someone else here can help?
Maybe I’ll look into Heroic tomorrow, thanks for the info though, never used it before
Heroic is very straightforward, as long as what you want is access to your GoG, Epic, Amazon and Battle.Net libraries. Lutris is meant as a more general purpose launcher, so they’re aiming at slightly different use cases that overlap.
Heroic won’t solve your Lutris ISO problem, but if you want to play some non-Steam ways it works great, is easy to use and is very Steam-like.
1: SteamOS don’t run unnecessary services in the background. (especially stuff like print services and other random shit)
1b: Even regular Linux, which does run a bunch of extra services, still generally has less overhead because it’s still being optimized for lighter weight systems, and it idles more efficiently too. Meanwhile Windows doesn’t have a good way to tell your printer driver and its corresponding services to shut up when you’re gaming.
2: Antivirus programs
3: Drivers, graphics system. This is both a plus and minus, but for performance mostly plus. More efficient driver model, less overhead again. Sometimes the performance comes from lacking features which doesn’t get executed fully, though. Sometimes it comes from translating to Vulkan, because DirectX has some more overhead (and in these specific cases you can get the same performance boost on Windows by switching to Vulkan).
I mean, this is wrong. The CUPS daemon literally is a print service and does exist in Linux. It’s just socket based so tends not to use resources until asked.
The CUPS daemon does not get killed during games. It wouldn’t be needed anyways
Antivirus programs? When was the last time you tried Windows, the mid-00s?
Anyway, it’s not random print services causing CPU overhead, that’s old timey stuff. In this case it’s being RAM heavy in a RAM-limited scenario and, from their testing, Lenovo being really terrible at keeping their AMD Windows drivers updated. As part of the test they manually switched to an ASUS version of newer AMD drivers and saw significant boosts in some games.
Modern graphics drivers are a mess of per-game features and optimizations. Different manufacturers keeping things at different levels of currency is a nontrivial issue and why some of this benchmarking is hard and throwing five random games at the problem doesn’t fully answer the question.
All modern versions of Windows will have Microsoft’s Defender antivirus/malware protection turned on by default. That means you incur a penalty every time a file is accessed from disk, or a process is launched, or a library loaded, or sockets are used or certain APIs are called.
It’s better than most 3rd party AV software but it’s still a performance overhead that could be turned off.
I mean… you can turn it off. I wouldn’t, but you can.
I just haven’t heard it referred to as “antivirus programs” in ages, it sounds so 20th century to me. Say what you will about MS’s monopolistic tendencies, but at least they killed the parasitic “antivirus” industry with that one.
I think Windows Defender is a fantastic line of defence and it’s definitely better than installing garbage from Trend, McAfee etc. That said, Lenovo, HP, MSI, Dell etc still preinstall crapware on their new machines from Norton or their ilk threatening that my machine is “at risk” if I don’t pay them money.
I wouldn’t turn it off unless I knew that I was only installing games from Steam. But if I did I think performance would improve. A game from Steam could still contain malware so you have to exercise some common sense. Even on SteamOS a game could be malicious but since its containerized the scope for damage is limited but not necessarily impossible to break out.
The Legion Go only has 16GB of RAM natively.
The Legion Go 1 yeah, this was on the Legion Go S, which has 32, apparently.
I’d say if you’re buying a handheld these days you should aim at 32 and look into having at least 8 available for the GPU.
Nope, the S is 16GB on the base models. It uses either the same Z1 Extreme chip as the non-S or the Z2 Go which is only 4c/8t, and is only 16GB with the latter. I think the former, which just released, has 512/16GB and 1TB/32GB variants.
As for memory sizes, it really depends on the chip. I think the Z1E would likely breathe a bit easier with 24GB but I think 32GB is kind of overkill for it. 16GB is likely plenty for the Z2 Go, and 24GB on the Z2 A in the ROG Xbox Ally is definitely overkill as that’s just AMD finally releasing a SKU for the Steam Deck APU, which is two to three generations older than every other Z2 chip.
The Z2, Z2 Go and Z1E all have only 12 graphics cores and that’s why I feel like 32GB is overkill for them, whereas the Z2 Extreme has 16 so it actually makes sense to dedicate more than 8GB of memory to it.
I think you’ve got to remember that “more memory better” doesn’t always hold true on a device like a handheld as it does include a cost on battery life even at idle.
As of right now, both models of the Go S listed on Lenovo’s website have 32 GB of RAM (screenshotted below, if the weird screenshot functionality here works). So no, you’re wrong here. The version with 16 GB is the Go 1. If there is a 16 gig SKU of the Go S, which there may be, they currently don’t have it listed.
Memory size requirements depend on what you’re trying to run. Easier to run stuff will run on everything, but from hands-on experience I assure you a bunch of newer games struggle with the default allocation of 4 gigs of VRAM and can use the extra RAM. You can still give 8 gigs to the GPU with 16 but then you’re a lot more likely to start struggling with system RAM. If these AMD APUs worked like an Apple chip and could dynamically allocate RAM that wouldn’t be such a pain, but at the moment you need a reboot to change this even on current-gen hardware, so it’s easier to have a larger pool and give the GPU a little too much.
The amount of CUs and the VRAM aren’t necessarily related. Even with larger RAM allocations and weaker GPUs you can find yourself in the wrong setup, which is annoying. And it’s not just amount of RAM, these shared architectures can struggle with bandwidth as well, so speed can matter (although it’s more giving you more or smoother FPS and the less the fall-off-a-cliff unplayable mess you get if the game is entirely out of RAM budget). That’s also why I suspect being lighter on memory and perhaps having a better default setup may be a part of why SteamOS performance is disproportionally better on heavier scenarios compared to what you see on desktop PCs. I can’t be sure, though.
This comes from me messing around with a literal handful of PC handhelds on Windows, SteamOS and Bazzite. I’m not guessing, I’m telling you what happened during hands-on testing.
It seems like maybe it’s a regional thing, as this is from Lenovo AU:
Either different SKUs per territory or one of the sites is straight up wrong. FWIW, cross checking with Amazon shows the Z2 versions of the Go S having 16 GB and the Z1 Extreme versions with 32 GB for my territory.
All else aside, man, is the world of PC handhelds the wild west right now. Everything is on effectively the same two or three pieces of hardware, but somehow nothing is consistent.
Idk if I’d call pc gaming still a tinkerers game, it’s 2025, console games have pretty glaring issues too.
Hah. Sure, but you can’t do much about those, so no tinkering.