[peterschmidt85]

Offers can report VRAM that differs from the GPU SKU's known capacity. This makes expected-VRAM filters unreliable and splits equivalent offers into separate groups when grouping by GPU.

• Normalize full-GPU, per-device VRAM against KNOWN_ACCELERATORS, accepting a correction only when exactly one capacity matches the existing ±7% window.
• Round reported VRAM to whole GiB when the model is unknown or no unique capacity matches, with warnings identifying the offer, original measurement, and returned value. Partial allocations keep their assigned VRAM.
• Correct DigitalOcean's combined-VRAM handling and the MI325X registry capacity, and add common RTX models found in real offers.

Examples, in GiB per GPU:

GPU	Backend	Before	After
B300	AWS	268.59	270
B300	OCI	262.5	270
B300	RunPod	288	270
B300	Verda	268	270
L40S	AWS	44.70	48
MI325X, 8 GPUs	DigitalOcean	2,048	256

Compatibility:

• SSH GPU discovery uses the existing helper; shim formats and configuration stay unchanged.
• Kubernetes continues using the registry for GPU memory; MI325X detection now reports 256GB instead of 288GB. No dstack code or configuration changes are required.
• Slurm syntax and P100/A100 mappings are unchanged. In gpu_partitions, gpu: MI325X:288GB will fail to resolve; change it to gpu: MI325X:256GB. Bare gpu: MI325X needs no edit and now resolves to 256GB.

[jvstme]

(nit) Is this necessary? I think we usually only export gpuhunt utility functions if we need to call them in dstack

[jvstme]

(nit) You've switched from correct_gpu_memory_gib to get_gpu_memory_gib throughout the codebase. The only behavioral difference is that get_gpu_memory_gib doesn't round the memory of unknown GPUs to a whole number. Wouldn't it make sense to round it? I don't think any GPU models have a nominal memory that isn't a whole number, so we probably don't want to display fractional values in dstack.

If you decide to round everywhere, I don't see a reason to keep both correct_gpu_memory_gib and get_gpu_memory_gib. I'd keep only correct_gpu_memory_gib.