hunk_alloc failed usually means you ran out of system memory and it couldn't allocate more. 128mb's kinda low for a late 2001 game, so what's your swap/virtual memory situation? The video card seems ideal. CPU's a little low but shouldn't cause this error.

(I do believe RTCW is pretty bloated and could've been better optimized, they've already tried to do compressed model formats in to keep things down)

Hi,

As leileilol says it is an out of memory error.

Technical details:
Return to Castle Wolfenstein uses the Quake III engine and Hunk_alloc uses malloc to allocate a continuous segment of memory.(It also zeros the memory with memset and put it into a list, but those parts shouldn't fail) The many X-es part is the size of the memory it tries to allocate. Usually this means that your system doesn't have a single continuous block of memory of the given size available for the game to allocate. 128MB is the minimal memory requirement for that game, you are right on that, most likely the system cannot find a continuous memory block large enough to allocate.

You can try to:
1. Increase the size of the Virtual Memory(As leileilol asked about), so the OS can swap out other programs to the disk while the game is running enabling more continuous memory space.
2. Shut down other programs running in the background, including those that are running windowless.
3. Lower the graphics detail, most of the graphics data(textures) is stored in VRAM, but some of it is in the system memory, lowering the detail might free up enough memory to be able to allocate.
4. If your BIOS have a memory hole setting, try disabling it, as it might free up a continuous memory block just enough that it might run. (Unlikely to be enough to make a difference, and if you have ISA cards you might need it, however all HW listed in the specs should not be ISA, I think "Genius SoundMaker 128 Value" only came in PCI form, but "Shuttle HOT-591P VIA MVP3" came with 3 ISA slots if the info I got from the net is correct)

I was able to run RTCW on my 98SE K6-2+/550 system with a voodoo3 and 128MB RAM but it was dog slow due to a massive CPU bottleneck. I also had to use an OpenGL shim to get the game running (Which shouldn't be required on a GF4 MX). My opinion is that trying to get the game to work would be purely academic, you'd play it for 1 minute at the terrible framerate then uninstall the game.

While I also don't think that playing this game would be a too good experience even if you can start it on this HW, there are also some more things that you can try if you are desperate to run it:

1. Set "com_soundMegs" cvar to a lower value, the default value is fine-tuned to not really load sounds during gameplay, decreasing this will free up memory, but will make the game load sounds as they are played, causing stuttering.
2. If I remember correctly the Quake3 engine downsamples sounds to the Khz used by the sound system freeing up more memory, if they didn't changed this in RTCW then you can set "s_khz" to something lower, I think the default is 22, try something lower like 11.

I think most of these should already be set if you choose a low detail option but if not, these will decrease detail further:
3. Set r_roundImagesDown to 1 (Most of the textures should be in VRAM, but some might stay in system RAM depending on how NVIDIA drivers handle it)
4. Set r_detailtextures to 0 (Again these mostly should be in VRAM, but NVIDIA drivers sometimes keep copies in system RAM)
5. Set r_subdivisions to a higher value. (The higher the uglier, but lower memory usage)
6. Set r_dynamiclight to 0 (Only helps if the dynamic lightning is calculated on the CPU, in case of an MX it might be the case)
7. Set r_cacheModels to 0 (Highly increased stuttering, only set if really desperate to run it)

You might have to set these in the config file rather than the console to apply properly. However if you set all of these then the game will look more like Quake 1 than how it supposed to look 😁

Thinking more about that Genius SoundMaker 128 Value. Does that card reserve system memory for a midi DLS bank?

Sorry for no updates, had to go to sleep.

First of all, thank you very much for pointing me towards the Virtual Memory! Setting it to 256 MB allowed me to stop running into "hunk_alloc..." at 80% of loading bar. Unfortunately, I'm now blocked at 95% of loading by error regarding the .mdl file, something about exceeding 4000 vertices. I can't recall the whole error message, I'll post it when I'll get back from work.

Hellvis wrote on 2026-06-09, 08:25:

Sorry for no updates, had to go to sleep.

First of all, thank you very much for pointing me towards the Virtual Memory! Setting it to 256 MB allowed me to stop running into "hunk_alloc..." at 80% of loading bar. Unfortunately, I'm now blocked at 95% of loading by error regarding the .mdl file, something about exceeding 4000 vertices. I can't recall the whole error message, I'll post it when I'll get back from work.

I have looked around and now it seems that RTCW is also open source(https://github.com/id-Software/RTCW-SP), checking that, the error message you most likely receiving is "R_Load<Model Type>: <model name> has more than 4000 verts on a surface (<Actual number of verts>)", the 4000 vertices is a #define so it gets hardcoded into the executable, not configurable, so it should work with all models coming with the game, that particular model file might be corrupted.

The game needs a faster CPU. It runs great on PIII 900E with GeForce FX 5600.

NeoG_ wrote on 2026-06-09, 00:37:

I was able to run RTCW on my 98SE K6-2+/550 system with a voodoo3 and 128MB RAM but it was dog slow due to a massive CPU bottleneck. I also had to use an OpenGL shim to get the game running (Which shouldn't be required on a GF4 MX). My opinion is that trying to get the game to work would be purely academic, you'd play it for 1 minute at the terrible framerate then uninstall the game.

Depends, the amd folks bragging about 40fps in RTCW are usually folks with a k6-3 and an overclocked FSB with good sdram and a motherboard with higher than normal memory transfer rates. Oddly win98se is not always the best os to deliver max frame rate either.

I have known folks who claimed it was playable on a standard k6-2 (vintage definition of playable is likely different than what you considered playable)
but again they almost always had a unicorn motherboard that could cache a full 256mb of ram with an overclocked FSB and higher than normal memory transfer rates per the usual utilities. An overclockable pci bus could also help if your board and hard allowed and was stable.

People who ran on underpowered hardware usually needed more than 128mb of ram to be playable.

last I ran RTCW on a k6-2 was during the Wolftest period in mid-2001. It was okay in indoor sections, it's just a lot of overdraw when out (was just the beach map at this point). RTCW does reimplement the broken flares and that's a depth/memory read based effect that could be disabled. But also i've had 192mb at least.

so i'd instinct towards r_subdivisions 128, r_lodbias 2 and r_dynamiclight 0 to keep the cpu from being choked by too many polys and decals being projected on triangles. r_primitives 2 also matters. Keep GL extensions ON - the purpose of the extensions used are for performance.

RTCW does not use detail textures, that's a feature only two licensees used (Alice and EF1) so that shouldn't matter.

Allright, I've installed the RTCW: Platinum Edition release (ver 1.32), which didn't really change much, aside from loading the first briefing cutscene, but as soon as the first gameplay level was loaded, I was greeted with the same "R_Load<Model Type>: <model name> has more than 4000 verts on a surface (<Actual number of verts>)" error message.

So I tried a workaround - using a cheat command "/spdevmap escape2"loaded straight into the next level. Lo and behold, no crash, the game works just fine.
By fine, I mean anywhere between 10-30 FPS, mostly hanging around 17 FPS depending on environment and number of enemies in the firefight. It means I can "enjoy" it the same way i "enjoyed" beating American McGee's Alice on my second rig (Celeron 500MHz, ATI Rage 128 Pro).

Thank you all so much for help and suggestions!

10-30 fps is definitely not playable. Crysis runs like that on my Athlon 64. I definitely wouldn't play it on it. You need to get 25-50 fps to enjoy it. No need for 60+ as these games rarely have too many enemies attacking at once.

FWIW I pulled out my TXP4 system and installed RTCW to see if it would work. Specs: K6-3+ (modified K6-2+) @ 500 MHz (6x83), 256 megs of RAM, an FX 5500 PCI video card, Win98 SE, and a CT2900 sound card. I can install and play RTCW, but even at 640x480 and the details turned down it turns into a slide show when the shooting starts, but it does run. This system scores around 2,800 points in 3D 2000.

Repo Man11 wrote on 2026-06-09, 22:29:

FWIW I pulled out my TXP4 system and installed RTCW to see if it would work. Specs: K6-3+ (modified K6-2+) @ 500 MHz (6x83), 256 megs of RAM, an FX 5500 PCI video card, Win98 SE, and a CT2900 sound card. I can install and play RTCW, but even at 640x480 and the details turned down it turns into a slide show when the shooting starts, but it does run. This system scores around 2,800 points in 3D 2000.

Yeah I wouldn’t expect an 83mhz FSB board to work well. On k6 systems the game is more FSB dependent than clock dependent.

Have you tried disabling L3 cache?
Is your pci bus overclocked?

rmay635703 wrote on Yesterday, 20:40:

Yeah I wouldn’t expect an 83mhz FSB board to work well. On k6 systems the game is more FSB dependent than clock dependent. […]
Show full quote

Repo Man11 wrote on 2026-06-09, 22:29:

FWIW I pulled out my TXP4 system and installed RTCW to see if it would work. Specs: K6-3+ (modified K6-2+) @ 500 MHz (6x83), 256 megs of RAM, an FX 5500 PCI video card, Win98 SE, and a CT2900 sound card. I can install and play RTCW, but even at 640x480 and the details turned down it turns into a slide show when the shooting starts, but it does run. This system scores around 2,800 points in 3D 2000.

Yeah I wouldn’t expect an 83mhz FSB board to work well. On k6 systems the game is more FSB dependent than clock dependent.

Have you tried disabling L3 cache?
Is your pci bus overclocked?

My K6-2+/550 @ 100x5.5 also runs at or near single digit framerates when there's a decent amount of enemies on screen, I just don't think it's feasible on this class of hardware

rmay635703 wrote on Yesterday, 20:40:

Yeah I wouldn’t expect an 83mhz FSB board to work well. On k6 systems the game is more FSB dependent than clock dependent. […]
Show full quote

Repo Man11 wrote on 2026-06-09, 22:29:

FWIW I pulled out my TXP4 system and installed RTCW to see if it would work. Specs: K6-3+ (modified K6-2+) @ 500 MHz (6x83), 256 megs of RAM, an FX 5500 PCI video card, Win98 SE, and a CT2900 sound card. I can install and play RTCW, but even at 640x480 and the details turned down it turns into a slide show when the shooting starts, but it does run. This system scores around 2,800 points in 3D 2000.

Yeah I wouldn’t expect an 83mhz FSB board to work well. On k6 systems the game is more FSB dependent than clock dependent.

Have you tried disabling L3 cache?
Is your pci bus overclocked?

I've done every tweak I can think of to increase performance on this system, and I've even played through Half-Life on it (not great, but not terrible). And the RTCW experience wasn't awful, I was actually enjoying it (and I could see how you would finish the game on such a system if it was all that you had), but I'd rather play it on my Socket A system. It was mainly to see if I would have issues like the OP did (I didn't) so his system ought to be upgradeable to run the game at least as well as mine did.

I ought to throw the VooDoo 2 that I have in it and see how that does.

V2 will be hit with tex cache thrashes pretty quickly.

AlexZ wrote on 2026-06-09, 21:20:

10-30 fps is definitely not playable.

the nice thing about vogons is that experience with older computers isn't an absolute as determined by phil or some other influencer. one does not need max performance, a fucked up patchwork repack of win98 and a pentium 3 (the most gaslit CPU around here) to enjoy a computer game. My recent completion of Deus Ex was a 7-20fps experience and I was just Fine. I'm not debating about 10fps sucking, i'm just here focused on the crash part.

I think that the "How many FPS a game needs to be enjoyable?" question depends on person to person, for example one of my oldest experience with 3D graphics was this game on the C64:
https://www.mobygames.com/game/6712/total-eclipse/
I don't remember the exact FPS it had run with, but I think it was <5 FPS (Maybe saying it was Seconds per Frame is more appropriate for it 😁), I still enjoyed playing it. I do agree that more FPS is better as long as the timing of the game mechanics isn't tied to it.

Another thing that I want to mention is in regards of this game and the AMD K6-II and K6-III CPUs, now that the game is open source, if someone is willing to put the time into it, it is possible to make a version of it that is optimized for it. John Carmack is a genius programmer, so beating his code in speed in general is very hard to do, however he always made it sure that the engine he was developing wasn't tied to a specific architecture.
A version of RTCW specifically made to run optimally only on the K6-II and K6-III processors would likely give a measurable and might be even significant increase of the game's performance (Some of the easier optimization opportunities I see: Set the flags of the compiler to generate code for the specific architecture ("-march=k6-2 -m3dnow -mtune=k6-2" with the GNU Compiler Collection, I don't remember the specific flag for MSVC now, but it is basically free performance increase on that CPU architecture as it is done by the compiler not a human.), make buffers the size and use the ideal alignment in memory so that it is optimal for the cache lines of these CPUs, and if someone is familiar with the 3DNow! instruction set: use 3DNow! extensively in the code where it fits although this last one might be considered medium difficulty instead of easy 😀 ).

Optimization switches won't help inherent data bloat/overdraw issues. Also Carmack's less involved with RTCW's rendering - if anything, RTCW fixed some of his broken shit.