<div dir="auto">1. So make gtt = ram/2. There's your 50%.<div dir="auto"><br></div><div dir="auto">2. Not our problem.</div><div dir="auto"><br></div><div dir="auto">Marek</div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Fri., May 20, 2022, 05:42 Christian König, <<a href="mailto:ckoenig.leichtzumerken@gmail.com">ckoenig.leichtzumerken@gmail.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div>
In theory we should allow much more than that. The problem is just
that we can't.<br>
<br>
We have the following issues:<br>
1. For swapping out stuff we need to make sure that we can allocate
temporary pages.<br>
Because of this TTM has a fixed 50% limit where it starts to
unmap memory from GPUs.<br>
So currently even with a higher GTT limit you can't actually use
this.<br>
<br>
2. Apart from the test case of allocating textures with increasing
power of two until it fails we also have a bunch of extremely stupid
applications.<br>
E.g. stuff like looking at the amount of memory available and
trying preallocate everything.<br>
<br>
I'm working on this for years, but there aren't easy solutions to
those issues. Felix has opted out for adding a separate domain for
KFD allocations, but sooner or later we need to find a solution
which works for everybody.<br>
<br>
Christian.<br>
<br>
<div>Am 20.05.22 um 11:14 schrieb Marek
Olšák:<br>
</div>
<blockquote type="cite">
<div dir="auto">Ignore the silly tests. We only need to make sure
games work. The current minimum requirement for running modern
games is 8GB of GPU memory. Soon it will be 12GB. APUs will need
to support that.
<div dir="auto"><br>
</div>
<div dir="auto">Marek</div>
</div>
<br>
<div class="gmail_quote">
<div dir="ltr" class="gmail_attr">On Fri., May 20, 2022, 03:52
Christian König, <<a href="mailto:ckoenig.leichtzumerken@gmail.com" target="_blank" rel="noreferrer">ckoenig.leichtzumerken@gmail.com</a>>
wrote:<br>
</div>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Am 19.05.22
um 16:34 schrieb Alex Deucher:<br>
> The current somewhat strange logic is in place because
certain<br>
> GL unit tests for large textures can cause problems with
the<br>
> OOM killer since there is no way to link this memory to a<br>
> process. The problem is this limit is often too low for
many<br>
> modern games on systems with more memory so limit the
logic to<br>
> systems with less than 8GB of main memory. For systems
with 8<br>
> or more GB of system memory, set the GTT size to 3/4 of
system<br>
> memory.<br>
<br>
It's unfortunately not only the unit tests, but some games as
well.<br>
<br>
3/4 of total system memory sounds reasonable to be, but I'm
100% sure <br>
that this will break some tests.<br>
<br>
Christian.<br>
<br>
><br>
> Signed-off-by: Alex Deucher <<a href="mailto:alexander.deucher@amd.com" rel="noreferrer noreferrer" target="_blank">alexander.deucher@amd.com</a>><br>
> ---<br>
> drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 25
++++++++++++++++++++-----<br>
> 1 file changed, 20 insertions(+), 5 deletions(-)<br>
><br>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c<br>
> index 4b9ee6e27f74..daa0babcf869 100644<br>
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c<br>
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c<br>
> @@ -1801,15 +1801,30 @@ int amdgpu_ttm_init(struct
amdgpu_device *adev)<br>
> /* Compute GTT size, either bsaed on 3/4th the size
of RAM size<br>
> * or whatever the user passed on module init */<br>
> if (amdgpu_gtt_size == -1) {<br>
> + const u64 eight_GB = 8192ULL * 1024 * 1024;<br>
> struct sysinfo si;<br>
> + u64 total_memory, default_gtt_size;<br>
> <br>
> si_meminfo(&si);<br>
> - gtt_size =
min(max((AMDGPU_DEFAULT_GTT_SIZE_MB << 20),<br>
> - adev->gmc.mc_vram_size),<br>
> - ((uint64_t)si.totalram *
si.mem_unit * 3/4));<br>
> - }<br>
> - else<br>
> + total_memory = (u64)si.totalram *
si.mem_unit;<br>
> + default_gtt_size = total_memory * 3 / 4;<br>
> + /* This somewhat strange logic is in place
because certain GL unit<br>
> + * tests for large textures can cause
problems with the OOM killer<br>
> + * since there is no way to link this
memory to a process.<br>
> + * The problem is this limit is often too
low for many modern games<br>
> + * on systems with more memory so limit the
logic to systems with<br>
> + * less than 8GB of main memory.<br>
> + */<br>
> + if (total_memory < eight_GB) {<br>
> + gtt_size =
min(max((AMDGPU_DEFAULT_GTT_SIZE_MB << 20),<br>
> +
adev->gmc.mc_vram_size),<br>
> + default_gtt_size);<br>
> + } else {<br>
> + gtt_size = default_gtt_size;<br>
> + }<br>
> + } else {<br>
> gtt_size = (uint64_t)amdgpu_gtt_size
<< 20;<br>
> + }<br>
> <br>
> /* Initialize GTT memory pool */<br>
> r = amdgpu_gtt_mgr_init(adev, gtt_size);<br>
<br>
</blockquote>
</div>
</blockquote>
<br>
</div>
</blockquote></div>