TL;DR
Play games on Amazon Luna for your next adventure, included with Prime
- A rotating selection of games, no download needed
- Play on TV, laptop or phone
- Fast, free delivery for your gear, too
An XDA writer ran Crysis Remastered from a temporary RAM disk on a Lenovo Yoga 9n Gen 11 with Nvidia’s RTX Spark, using memory reserved from the system’s shared pool for the GPU. The experiment worked, but Windows on Arm’s Prism emulator still read translated code from the SSD, so the game was not running entirely from memory.
An XDA writer ran Crysis Remastered from a temporary RAM disk on a Lenovo Yoga 9n Gen 11 equipped with Nvidia’s RTX Spark, testing what it means to put a game “inside” a GPU that shares memory with the computer’s processor. Resource Monitor showed the game reading files from the RAM disk, but Windows on Arm’s Prism emulator also accessed translated code stored on the SSD.
The laptop has 64GB of shared LPDDR5X memory, rather than separate system RAM and graphics memory. A setting in Nvidia’s app determines how much is reserved for the GPU; changing the allocation requires a restart. In the writer’s test, a 56GB GPU allocation left 8GB reserved for the GPU and 38.1GB available to Windows. With a 48GB/16GB split, Windows reported only 30.3GB available, while about 16GB was already in use.
To place the game in memory, the writer used SoftPerfect RAM Disk, which supports Arm64 Windows, to create a 23GB temporary drive. The game occupied 19.9GB. After copying the files to the RAM disk, the writer used a Windows junction so the EA app could continue treating the game as if it were in its usual Program Files location. A RAM disk is volatile: its contents disappear after a reboot.
While the game loaded, Windows Resource Monitor showed reads from the RAM disk’s Crysis Remastered folder. The writer then played part of the first level, saved and reloaded the game, reporting that it behaved like a normal installation. With the game at its pause menu, Task Manager showed 42.5GB of 54.8GB in use and 12.2GB available; the SSD showed 2% activity. Those are observations from this laptop and test, not a general performance result for other systems.
Hardware experiment · Published October 26, 2026
Crysis Inside My GPU Is the Weirdest Thing I’ve Tried This Year
An XDA writer put Crysis Remastered on a temporary RAM disk in a Lenovo Yoga 9n Gen 11 with Nvidia’s RTX Spark. The game files came from shared memory reserved for the GPU, while Windows on Arm still fetched translated code from the SSD.
The setup
One pool of memory, two jobs
The laptop has no separate physical VRAM store. A reserved slice of its shared memory is labeled “Dedicated GPU memory,” while Windows uses the remainder.
CPU + GPU memory
The Arm CPU and Blackwell GPU share LPDDR5X memory rated up to 300 GB/s, according to the source account.
A reservation with a cost
Nvidia’s app adjusts the GPU reservation, and changing it requires a restart. More reserved memory leaves less available to Windows.
RAM disk, not an install drive
SoftPerfect RAM Disk created a temporary 23 GB drive. Its contents disappear after reboot, so this was an experiment rather than a permanent installation.
Memory allocation
The split changes what Windows can use
In the writer’s test, background system use made the allocation less straightforward than the headline numbers suggest.
Two GPU reservation settings
How it worked
Files in memory, launched as usual
A Windows junction let the EA app keep using the game’s usual install path after the files were moved to the RAM disk.
Create
Set up a 23 GB temporary drive with Arm64-compatible SoftPerfect RAM Disk.
Copy
Move the 19.9 GB Crysis Remastered installation onto the drive.
Link
Use a Windows junction at the expected Program Files location.
Play
Launch through the EA app; the writer played part of a level, saved, and reloaded.
The catch
“Inside the GPU” did not mean “off the SSD”
Resource Monitor showed reads from the RAM disk’s Crysis Remastered folder. It also showed a cached Nvidia library being read from XtaCache on C: while Prism translated x86 code for Windows on Arm.
“The game files were in memory, but some of what actually ran was coming from a cache on the SSD.”— XDA writer Joe
- The GPU’s reserved memory is part of one shared physical pool, not a separate drive for game installs.
- Windows background use reduces the memory available for the game and other work.
- Resource Monitor showed both RAM disk reads and SSD cache activity.
- The setup demonstrates shared memory and emulation; it does not establish a performance gain.
What the numbers say
A storage benchmark, not a frame-rate result
The reported CrystalDiskMark figures compare storage reads. The article does not give a controlled game loading-time or gameplay comparison.
The reported test used three passes and a 1 GiB file; the SSD was 89% full. The source excerpt ends partway through its benchmark table, so remaining results and a full comparison are unavailable.
Key questions
What this experiment does—and doesn’t—show
Was the game installed in physical GPU VRAM?
No. The laptop shares LPDDR5X memory between CPU and GPU. The writer reserved part of that pool and created a RAM disk in system memory.
Did the game run from the RAM disk?
The game files were copied there, and the writer reported playing part of a level, saving, and reloading successfully through the usual launcher path.
Were SSD reads eliminated?
No. Prism’s translated code cache, XtaCache, remained on the C: drive, and Resource Monitor showed SSD activity.
Does this prove RAM disks make games faster?
No. The report includes storage benchmark readings but no controlled comparison of game loading times or frame rates.
What happens after a reboot?
The RAM disk is volatile, so its contents disappear when the computer restarts.
What would clarify the practical impact?
Controlled RAM disk and SSD comparisons with matched memory settings and software conditions, accounting for Prism and Windows memory use.
The Limits of a GPU RAM Disk
The experiment makes a hardware distinction easier to see: on this system, “GPU memory” is a reserved portion of one shared pool, not a separate physical store into which a game can simply be installed. Giving more memory to the GPU reduces what Windows can use, and the writer found that background system use made the split less straightforward than the numbers might suggest.
It also shows why moving game files to memory does not necessarily remove storage activity. Because the laptop runs Windows on Arm, Prism translates x86 code and stores translated material in XtaCache on the C: drive. Resource Monitor showed a cached Nvidia library being read there as the game ran. The test is a practical demonstration of shared memory and emulation, not evidence that RAM-disk gaming will improve frame rates or that all game activity can be moved off an SSD.
Products Worth Considering
LG UltraGear 27-Inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1
Samsung 32-Inch Odyssey OLED G8 (G80SH) 4K Gaming Monitor
Bose SoundLink Flex Bluetooth Speaker (2nd Gen), Portable Waterproof Speaker, Black
As an affiliate, we earn on qualifying purchases.
How the Crysis Test Was Set Up
“Can it run Crysis?” has been a PC hardware joke since the original game arrived in 2007. The XDA writer framed this test as a deliberately unusual version of that question: instead of benchmarking the game on Nvidia’s RTX Spark, the aim was to put the installed game files in memory allocated to the GPU on a Lenovo Yoga 9n Gen 11.
The machine’s Arm CPU and Blackwell GPU share LPDDR5X memory rated for up to 300GB/s of bandwidth, according to the source account. Nvidia’s app provides a slider for the GPU reservation, with the writer’s unit allowing allocations as low as 4GB. Task Manager labels the reserved portion “Dedicated GPU memory,” while Windows groups it under hardware-reserved memory. The test also involved the original 2007 Crysis, which the writer said launched through Prism, but the RAM-disk run described in detail was Crysis Remastered.
““Can it run Crysis?””
— XDA writer Joe
What the Test Cannot Establish
The source describes a single laptop and one setup; it does not establish how the same process would work on other shared-memory systems, or whether it changes game performance. The account gives selected CrystalDiskMark results, including RAM-disk read speeds of 28,763 MB/s versus 7,293 MB/s for the SSD in the SEQ1M Q8T1 test, but those are storage benchmark readings, not measured frame-rate gains. The test used three passes and a 1GiB file, and the SSD was 89% full.
The source excerpt ends partway through its benchmark table, so the remaining test results and any full comparison are unavailable here. It also does not report a controlled comparison of loading times or gameplay performance between the RAM disk and SSD. The writer’s observations about memory use and disk access should not be treated as a broad specification for every RTX Spark system.
Further Tests Would Need Comparisons
The account documents a completed experiment rather than announcing a follow-up. A clearer measure of practical impact would require controlled load-time and gameplay comparisons between the RAM disk and SSD, with the same memory allocation and software conditions. Such testing would also need to account for Prism’s translation cache and Windows’ own memory use.
For now, the demonstrated result is narrower: Crysis Remastered could be launched through the EA app while its game files sat on a temporary RAM disk, and the system still accessed translated code from the SSD. Readers attempting the setup would also need to account for the RAM disk’s volatility and the GPU reservation’s direct effect on memory available to Windows.
Key Questions
Was Crysis Remastered installed in physical GPU VRAM?
No separate VRAM was used. The laptop shares LPDDR5X memory between its CPU and GPU; the writer reserved part of that pool for the GPU and created a RAM disk in system memory.
How did the game run from the RAM disk?
The writer copied the game files to a 23GB SoftPerfect RAM disk and used a junction so the EA app could access them at the expected install path. The writer reported playing part of a level, saving and reloading successfully.
Did the experiment eliminate SSD reads?
No. Resource Monitor showed reads from the RAM disk, but also activity from Windows’ XtaCache, which stores code translated by Prism for Windows on Arm.
Does the test prove games run faster from a RAM disk?
No. The report includes storage benchmark readings and system observations, but no controlled comparison of game loading times or frame rates. It does not establish a gameplay performance advantage.
What happens to the game files after a reboot?
The RAM disk was configured as volatile storage, so its contents disappear when the computer restarts. The writer used it for an experiment, not as a permanent game installation.
Source: XDA Developers
