Some custom nodes hook iterables with side effects and in some cases
these can cause exception in workflows that belong in the past.
Guard against this so comfy stays up and does not trash a workflow
that was innocent based of a bug in a previous workflow.
AMD+windows sets an unusually low maximum GPU virtual memory quota
compared to other plaforms. For reference, the hardware limits on
XT9060 are 128TB. On Nvidia RTX3060 it is 1TB. On nvidia RTX5090
it is 128TB.
Bump this into the same league, but keep it to a modest 4TB pending
further usage case need + testing going even higher to the hardware
limits.
* model_prefetch: add friendly API to pause comfy compiler
* sparse_attention: Pause comfy compiler for un-freed pieces
The VSA plan and pool stats do not free within the transformer causing
a comfy-compiler rogue allocation and graph break.
Ideally these should be freed in the lifetime of the transformer run
(and comfy-compiler scope) but there isnt a convenient place for the
particular lifecylce intended by these variables. So go with the pause()
approach.
Comfy-aimdo 0.4.15 fixes the headroom on the NVML pressure mechanism.
Some system have only NVML as the pressure out and this could cause
shared VRAM spills. This bumps the headroom to 512MB.
I never reproduced this issue.
Comfy-aimdo 0.4.12 increases error logging reliablity to help root
cause os errors in some of the C APIs that are causing issues for
some users.
The log is also unified with python logging, so non-terminal users
see the logs properly.
Aimdo 0.4.13 fixes a bug in async-offload + MRU primary weights
allocation. https://github.com/Comfy-Org/ComfyUI/issues/15284
This priority scheme was broken in the case where you have pin
registration exhaustion while loading a VBAR that gets a big evicition.
The weight would stay in the loaded set but inherit the MRU priority
against other workflow models WRT pin registration which leads to async
offload without pinning.
Fix by universally promiting active pin registration above workflow
pins without concern for the weights/weights-loaded split. This diverges
from the actual budgeting where the split still makes sense.
Changes:
Remove sequential scan hint
Prefer NVML pressure on windows
Add async malloc clamp option (unused by comfy so far)
Workaround AMD windows GPU virtual address space leak
The largest change is the NVML pressure, which works around a cuMemGetInfo
drift from actual VRAM in some circumstances.
Windows has proven this logic works for a long time and there are
corner cases where this materialization actual consumes real RAM
on linux.
Its not as bad as the original windows commit charge surge, but
its still a detectable transient leak. So simplify and unify.
Some long running chaos testing on a 512GB RAM RTX6000 pro showed that this
is a little bit too low for common template workflows switching around. The
original number was just a guess from me, so go with the scientific result
instead.
These were alll non-dynamic (some non-ModelPatcher) code path calling
FreeMemory for management requiring up-front memory freeing. Convert it
to dynamic to avoid legacy free behaviour mixing into otherwise
dynamic workflows.
The aimdo 0.4.10 protocol causing startup failure to be too early and
before the aimdo version warning can happen. This causes user
confusion. Limp on with 0.4.9 as it will work and users will see the
version warning.
* main: implement --vram-headroom
Implement --vram-headroom for dynamic vram as a hybrid debug/diagnostic
option that can be used for people who still report shared VRAM spills.
They can trial and error the setting to maintain a bit more headroom to
avoid shared VRAM spills.
* main: implement --reserve-vram
Implement --reserve-vram as extra headroom on the simple method which
is semantically as close as possible to the stated functionality and
formet behaviour of non-dynamic VRAM.
Add this option for users who know they have so much ram they want
to pin everything or have a pagefile that outruns their disk speed.
The removes the RAM pressure caps completely and pins behind the
primary model load forcing all models to be permanently comitted
to RAM.
Some custom nodes .to weights completely out of load context which
can wreak havoc if its for a model that is not active. Detect this
condition and just let it fall-through to the non-dynamic loader
straight up.
Some custom nodes try to set this true globally. It messes with dynamic
VRAM with one-off spikes that can OOM but this is also very high risk
for windows where such allocations might get serviced by shared memory
fallback.
Trump it.
cleanup_models_gc can be called once per load_models_gpu via
free_memory, which in turn can de-activate an active model via
this reset_cast_buffers.
cleanup_models_gc() could also come via obscure garbage collector
paths so limit reset_cast_buffers to the post-node callsite instead.
* mm: split off registration helper to doer and headroom calc
* pinned_memory: implement registration comfy side
Move away from Aimdo buffer registrations which seem fraught with
danger and do it comfy side. Just start with the basic move.
* pinned_memory: do registrations as portable memory
* pinned_memory: discard async errors on registration fail
Like the good ol days.
* pinned_memory: implement abs shortfall retry
If pinned registration happens to fail despite the previous budget
ensures, consider the allocation shortfall, ensure it again, and
try again. This allows comfy pins to interoperate with other software
that might be doing substantive pinning.
* mm: re-instantate smart memory for VRAM
* mm: restore non-dynamic smart memory
By popular demand. We aren't quite ready for the deprecation as non
dynamic enabled GPUs and some high-vram custom model loader setups
prefer the old full hands on.
* memory_management: Add direct to read GPU mode
Make destination optional (or make it optionally GPU) and use aimdo
to file_read direct to GPU.
* ops: Remove stream pin buffers and use aimdo reads
This consumed too much RAM and its better to just take the hit on
the CPU syncing back the stream on a short ring buffer. Aimdo
implements this so just rip the stream pin buffer from comfy.
* model_management: all active pin registration movement
Its better to just let the active model load past the pin limit as
pins and let the pins move around. The saves the HDD and SATA
people disk traffic while only costing a few GPU syncs.
* utils: use aimdo file handle
This opens on windows with more favourable flags
* mp: only count the model proper for loaded_ram and vram
Exclude live loras from the numbers to avoid the case where the reported
loaded memory exceeds the size of the model.
This causes me confusion in the Kijai visualizer when it looked fully
loaded but was hitting disk due to this accounding disrepency.
* utils: add bit reverse utility
useful for max scattering something ordered.
* pinned_memory: Implement offload balancing
Use a max scatter alogorithm to prioritize pins of the same size such
that when doing a little bit of offloading it gets scattered, allowing
the prefetcher to more evenly swollow the offload.
* comfy-aimdo 0.4.7
Aimdo 0.4.7 implement VRAM buffer exhaustion predection to avoid
early speculative load of weights that definately wont fix once the
inference gets further in.
* model-prefetch: consolidate pin ensures on the sync point
This could happen mid prefetch block, cause a sync of the entire
block and lose overlap. Get ahead of the problem with a free down
at the natural compute stream sync point.
* mm: Put a 2GB min on the pin ceiling
This is reasonably bad if it starts causing swap pressure, moreso than
during normal ram-cache proceedings. Clamp it.
* add --fast-disk
Use the RAM right up to the wire as the community is bit accustomed too.
This trades off headroom for the case where large chunky intermediates
arrive and potenitally hits pagefile/swap, but a lot of people have
"it just fits" workflows out there, so strike a compromise with
75->90%.
Disable the incative cache for all but the very high RAM users.
* ModelPatcherDyanmic: purge stale vbar allocs on force cast
* ModelPatcherDynamic: restore backups before load
If doing a clean reload, mutative changes (lora application) could be
applied on-top of the already loaded weight. Restore from backup
unconditionally so that the new load is clean.
* model_management: disable non-dynamic smart memory
Disable smart memory outright for non dynamic models.
This is a minor step towards deprecation of --disable-dynamic-vram
and the legacy ModelPatcher.
This is needed for estimate-free model development, where new models
can opt-out of supplying a memory estimate and not have to worry
about hard VRAM allocations due to legacy non-dynamic model patchers
This is also a general stability increase for a lot of stray use cases
where estimates may still be off and going forward we are not going
to accurately maintain such estimates.
* pinned_memory: implement with aimdo growable buffer
Use a single growable buffer so we can do threaded pre-warming on
pinned memory.
* mm: use aimdo to do transfer from disk to pin
Aimdo implements a faster threaded loader.
* Add stream host pin buffer for AIMDO casts
Introduce per-offload-stream HostBuffer reuse for pinned staging,
include it in cast buffer reset synchronization.
Defer actual casts that go via this pin path to a separate pass
such that the buffer can be allocated monolithically (to avoid
cudaHostRegister thrash).
* remove old pin path
* Implement JIT pinned memory pressure
Replace the predictive pin pressure mechanism with JIT PIN memory
pressure.
* LowVRAMPatch: change to two-phase visit
* lora: re-implement as inplace swiss-army-knife operation
* prepare for multiple pin sets
* implement pinned loras
* requirements: comfy-aimdo 0.4.0
* ops: remove unused arg
This was defeatured in aimdo iteration
* ops: sync the CPU with only the offload stream activity
This was syncing with the offload stream which itself is synced with the
compute stream, so this was syncing CPU with compute transitively. Define
the event to sync it more gently.
* pins: implement freeing intermediate for pinned memory
Pinning is more important than inactive intermediates and the stream
pin buffer is more important than even active intermediates.
* execution: implement pin eviction on RAM presure
Add back proper pin freeing on RAM pressure
* implement pin registration swaps
Uncap the windows pins from 50% by extending the pool and have a pressure
mechanism to move the pin reservations om demand.
This unfortunately implies a GPU sync to do the freeing so significant
hysterisis needs to be added to consolidate these pressure events.
* cli_args/execution: Implement lower background cache-ram threshold
Limit the amount of RAM background intermediates can use, so that
switching workflows doesn't degrade performance too much.
* make default
* bump aimdo
* model-patcher: force-cast tiny weights
Flux 2 gets crazy stalls due to a mix of tiny and giant weights
creating lopsided steam buffer rotations which creates stalls.
* ops: refactor in prep for chunking
* mm: delegate pin-on-the-way to aimdo
Aimdo is able to chunk and slice this on the way for better CPU->GPU
overlap. The main advantage is the ability to shorten the bus contention
window between previous weight transfer and the next weights vbar
fault.
* bump aimdo
* pinning updates
* specify hostbuf max allocation size
There a signs of virtual memory exhaustion on some linux systems when
throwing 128GB for every little piece. Pass the actual to save aimdo
from over-estimates
* tests: update execution tests for caching
The default caching changed to ram-cache so update these tests
accordingly.
Remove the LRU 0 test as this also falls through to RAM cache.
If the same weight is used multiple times within the same prefetch
window, it should only apply compute state mutations once. Mark the
weight as fully resident on the first pass accordingly.
* mm: Use Aimdo raw allocator for cast buffers
pytorch manages allocation of growing buffers on streams poorly. Pyt
has no windows support for the expandable segments allocator (which is
the right tool for this job), while also segmenting the memory by
stream such that it can be generally re-used. So kick the problem to
aimdo which can just grow a virtual region thats freed per stream.
* plan
* ops: move cpu handler up to the caller
* ops: split up prefetch from weight prep block prefetching API
Split up the casting and weight formating/lora stuff in prep for
arbitrary prefetch support.
* ops: implement block prefetching API
allow a model to construct a prefetch list and operate it for increased
async offload.
* ltxv2: Implement block prefetching
* Implement lora async offload
Implement async offload of loras.
* pinned_memory: remove JIT RAM pressure release
This doesn't work, as freeing intermediates for pins needs to be
higher-priority than freeing pins-for-pins if and when you are going
to do that. So this is too late as pins-for-pins is model load time
and we dont have JIT pins-for-pins.
* cacheing: Add a filter to only free intermediates from inactive wfs
This is to get priorities in amongst pins straight.
* mm: free inactive-ram from RAM cache first
Stuff from inactive workflows should be freed before anything else.
* caching: purge old ModelPatchers first
Dont try and score them, just dump them at the first sign of trouble
if they arent part of the workflow.
Comfy-aimdo 0.3.0 contains several major new features.
multi-GPU support
ARM support
AMD support
Refactorings include:
Linkless architecture - linkage is now performed purely at runtime
to stop host library lookups completely and only interact with the
torch-loaded Nvidia stack.
Elimination of cudart integration on linux. Its no consistent with
windows.
Misc bugfixes and minor features.
Currently if the graph contains a cycle, the just inifitiate recursions,
hits a catch all then throws a generic error against the output node
that seeded the validation. Instead, fail the offending cycling mode
chain and handlng it as an error in its own right.
Co-authored-by: guill <jacob.e.segal@gmail.com>
This was doing an over-estimate of VRAM used by the async allocator when lots
of little small tensors were in play.
Also change the versioning scheme to == so we can roll forward aimdo without
worrying about stable regressions downstream in comfyUI core.
the mixed_precision ops can have input_scale parameters that are used
in tensor math but arent a weight or bias so dont get proper VRAM
management. Treat these as force-castable parameters like the non comfy
weight, random params are buffers already are.
* mm: Lower windows pin threshold
Some workflows have more extranous use of shared GPU memory than is
accounted for in the 5% pin headroom. Lower this for safety.
* mm: Remove pin count clearing threshold.
TOTAL_PINNED_MEMORY is shared between the legacy and aimdo pinning
systems, however this catch-all assumes only the legacy system exists.
Remove the catch-all as the PINNED_MEMORY buffer is coherent already.
There was an issue where the resample split was too early and dropped one
of the rolling convolutions a frame early. This is most noticable as a
lighting/color change between pixel frames 5->6 (latent 2->3), or as a
lighting change between the first and last frame in an FLF wan flow.
* sd: soft_empty_cache on tiler fallback
This doesnt cost a lot and creates the expected VRAM reduction in
resource monitors when you fallback to tiler.
* wan: vae: Don't recursion in local fns (move run_up)
Moved Decoder3d’s recursive run_up out of forward into a class
method to avoid nested closure self-reference cycles. This avoids
cyclic garbage that delays garbage of tensors which in turn delays
VRAM release before tiled fallback.
* ltx: vae: Don't recursion in local fns (move run_up)
Mov the recursive run_up out of forward into a class
method to avoid nested closure self-reference cycles. This avoids
cyclic garbage that delays garbage of tensors which in turn delays
VRAM release before tiled fallback.
* ltx: vae: add cache state to downsample block
* ltx: vae: Add time stride awareness to causal_conv_3d
* ltx: vae: Automate truncation for encoder
Other VAEs just truncate without error. Do the same.
* sd/ltx: Make chunked_io a flag in its own right
Taking this bi-direcitonal, so make it a for-purpose named flag.
* ltx: vae: implement chunked encoder + CPU IO chunking
People are doing things with big frame counts in LTX including V2V
flows. Implement the time-chunked encoder to keep the VRAM down, with
the converse of the new CPU pre-allocation technique, where the chunks
are brought from the CPU JIT.
* ltx: vae-encode: round chunk sizes more strictly
Only powers of 2 and multiple of 8 are valid due to cache slicing.
* ltx: vae: add cache state to downsample block
* ltx: vae: Add time stride awareness to causal_conv_3d
* ltx: vae: Automate truncation for encoder
Other VAEs just truncate without error. Do the same.
* sd/ltx: Make chunked_io a flag in its own right
Taking this bi-direcitonal, so make it a for-purpose named flag.
* ltx: vae: implement chunked encoder + CPU IO chunking
People are doing things with big frame counts in LTX including V2V
flows. Implement the time-chunked encoder to keep the VRAM down, with
the converse of the new CPU pre-allocation technique, where the chunks
are brought from the CPU JIT.
* ltx: vae-encode: round chunk sizes more strictly
Only powers of 2 and multiple of 8 are valid due to cache slicing.
* wan: vae: encoder: Add feature cache layer that corks singles
If a downsample only gives you a single frame, save it to the feature
cache and return nothing to the top level. This increases the
efficiency of cacheability, but also prepares support for going two
by two rather than four by four on the frames.
* wan: remove all concatentation with the feature cache
The loopers are now responsible for ensuring that non-final frames are
processes at least two-by-two, elimiating the need for this cat case.
* wan: vae: recurse and chunk for 2+2 frames on decode
Avoid having to clone off slices of 4 frame chunks and reduce the size
of the big 6 frame convolutions down to 4. Save the VRAMs.
* wan: encode frames 2x2.
Reduce VRAM usage greatly by encoding frames 2 at a time rather than
4.
* wan: vae: remove cloning
The loopers now control the chunking such there is noever more than 2
frames, so just cache these slices directly and avoid the clone
allocations completely.
* wan: vae: free consumer caller tensors on recursion
* wan: vae: restyle a little to match LTX
* ltx: vae: scale the chunk size with the users VRAM
Scale this linearly down for users with low VRAM.
* ltx: vae: free non-chunking recursive intermediates
* ltx: vae: cleanup some intermediates
The conv layer can be the VRAM peak and it does a torch.cat. So cleanup
the pieces of the cat. Also clear our the cache ASAP as each layer detect
its end as this VAE surges in VRAM at the end due to the ended padding
increasing the size of the final frame convolutions off-the-books to
the chunker. So if all the earlier layers free up their cache it can
offset that surge.
Its a fragmentation nightmare, and the chance of it having to recache the
pyt allocator is very high, but you wont OOM.
If a subclass BYO _load_from_state_dict and doesnt call the super() the
needed default init of these weights is missed and can lead to problems
for uninitialized weights.
* Implement seek and read for pins
Source pins from an mmap is pad because its its a CPU->CPU copy that
attempts to fully buffer the same data twice. Instead, use seek and
read which avoids the mmap buffering while usually being a faster
read in the first place (avoiding mmap faulting etc).
* pinned_memory: Use Aimdo pinner
The aimdo pinner bypasses pytorches CPU allocator which can leak
windows commit charge.
* ops: bypass init() of weight for embedding layer
This similarly consumes large commit charge especially for TEs. It can
cause a permanement leaked commit charge which can destabilize on
systems close to the commit ceiling and generally confuses the RAM
stats.
* model_patcher: implement pinned memory counter
Implement a pinned memory counter for better accounting of what volume
of memory pins have.
* implement touch accounting
Implement accounting of touching mmapped tensors.
* mm+mp: add residency mmap getter
* utils: use the aimdo mmap to load sft files
* model_management: Implement tigher RAM pressure semantics
Implement a pressure release on entire MMAPs as windows does perform
faster when mmaps are unloaded and model loads free ramp into fully
unallocated RAM.
Make the concept of freeing for pins a completely separate concept.
Now that pins are loadable directly from original file and don' touch
the mmap, tighten the freeing budget to just the current loaded model
- what you have left over. This still over-frees pins, but its a lot
better than before.
So after the pins are freed with that algorithm, bounce entire MMAPs
to free RAM based on what the model needs, deducting off any known
resident-in-mmap tensors to the free quota to keep it as tight as
possible.
* comfy-aimdo 0.2.11
Comfy aimdo 0.2.11
* mm: Implement file_slice path for QT
* ruff
* ops: put meta-tensors in place to allow custom nodes to check geo
Comfy Aimdo 0.2.10 fixes the aimdo allocator hook for legacy cudaMalloc
consumers. Some consumers of cudaMalloc assume implicit synchronization
built in closed source logic inside cuda. This is preserved by passing
through to cuda as-is and accouting after the fact as opposed to
integrating these hooks with Aimdos VMA based allocator.
Pytorch only filters for OOMs in its own allocators however there are
paths that can OOM on allocators made outside the pytorch allocators.
These manifest as an AllocatorError as pytorch does not have universal
error translation to its OOM type on exception. Handle it. A log I have
for this also shows a double report of the error async, so call the
async discarder to cleanup and make these OOMs look like OOMs.
Comfy-aimdo 0.2.9 fixes a context issue where if a non-main thread does
a spurious garbage collection, cudaFrees are attempted with bad
context.
Some new APIs for displaying aimdo stats in UI widgets are also added.
These are purely additive getters that dont touch cuda APIs.
Sync the compute stream before freeing the cast buffers. This can cause
use after free issues when the cast stream frees the buffer while the
compute stream is behind enough to still needs a casted weight.
* mp: respect model_defined_dtypes in default caster
This is needed for parametrizations when the dtype changes between sd
and model.
* audio_encoders: archive model dtypes
Archive model dtypes to stop the state dict load override the dtypes
defined by the core for compute etc.
Comfy-aimdo 0.2.7 fixes a crash when a spurious cudaAsyncFree comes in
and would cause an infinite stack overflow (via detours hooks).
A lock is also introduced on the link list holding the free sections
to avoid any possibility of threaded miscellaneous cuda allocations
being the root cause.
* ops: dont unpin nothing
This was calling into aimdo in the none case (offloaded weight). Whats worse,
is aimdo syncs for unpinning an offloaded weight, as that is the corner case of
a weight getting evicted by its own use which does require a sync. But this
was heppening every offloaded weight causing slowdown.
* mp: fix get_free_memory policy
The ModelPatcherDynamic get_free_memory was deducting the model from
to try and estimate the conceptual free memory with doing any
offloading. This is kind of what the old memory_memory_required
was estimating in ModelPatcher load logic, however in practical
reality, between over-estimates and padding, the loader usually
underloaded models enough such that sampling could send CFG +/-
through together even when partially loaded.
So don't regress from the status quo and instead go all in on the
idea that offloading is less of an issue than debatching. Tell the
sampler it can use everything.
Define a threshold below which a weight loading takes priority. This
actually makes the offload consistent with non-dynamic, because what
happens, is when non-dynamic fills ints to_load list, it will fill-up
any left-over pieces that could fix large weights with small weights
and load them, even though they were lower priority. This actually
improves performance because the timy weights dont cost any VRAM and
arent worth the control overhead of the DMA etc.
* respect model dtype in non-comfy caster
* utils: factor out parent and name functionality of set_attr
* utils: implement set_attr_buffer for torch buffers
* ModelPatcherDynamic: Implement torch Buffer loading
If there is a buffer in dynamic - force load it.
* model_management: Remove non-comfy dynamic _v caster
* Force pre-load non-comfy weights to GPU in ModelPatcherDynamic
Non-comfy weights may expect to be pre-cast to the target
device without in-model casting. Previously they were allocated in
the vbar with _v which required the _v fault path in cast_to.
Instead, back up the original CPU weight and move it directly to GPU
at load time.
Comfy Aimdo 0.2.4 fixes a VRAM buffer alignment issue that happens in
someworkflows where action is able to bypass the pytorch allocator
and go straight to the cuda hook.
This was previously considering the pool of dynamic models as one giant
entity for the sake of smart memory, but that isnt really the useful
or what a user would reasonably expect. Make Dynamic VRAM properly purge
its models just like the old --disable-smart-memory but conditioning
the dynamic-for-dynamic bypass on smart memory.
Re-enable dynamic smart memory.
* sd: add support for clip model reconstruction
* nodes: SetClipHooks: Demote the dynamic model patcher
* mp: Make dynamic_disable more robust
The backup need to not be cloned. In addition add a delegate object
to ModelPatcherDynamic so that non-cloning code can do
ModelPatcherDynamic demotion
* sampler_helpers: Demote to non-dynamic model patcher when hooking
* code rabbit review comments
Allow non QuantizedTensor layer to set want_requant to get the post lora
calculation stochastic cast down to the original input dtype.
This is then used by the legacy fp8 Linear implementation to set the
compute_dtype to the preferred lora dtype but then want_requant it back
down to fp8.
This fixes the issue with --fast fp8_matrix_mult is combined with
--fast dynamic_vram which doing a lora on an fp8_ non QT model.
Some custom node packs are naughty, and violate the
dont-load-torch-on-load rule. This causes aimdo to lose preference on
its allocator hook on linux.
Go super early on the aimdo first-stage init before custom nodes
are mentioned at all.
Comfy Aimdo 0.2.2 moves the cuda allocator hook from the cudart API to
the cuda driver API on windows. This is needed to handle Windows+cu13
where cudart is statically linked.
rattus