Blender 3.3 Released With Intel oneAPI Backend, Improved AMD HIP Support

[qarium]

do you have a link to said patches? and I dont have a misunderstanding, if its not something thats supported in the end it doesn't matter.

https://github.com/RadeonOpenCompute/ROCm/issues/1745

the question is what is better driver from amd.com ? or driver in your distro ?

i would say driver in your distro is better. and distros like Arch already patched the polaris support in...

[Quackdoc]

https://github.com/RadeonOpenCompute/ROCm/issues/1745

the question is what is better driver from amd.com ? or driver in your distro ?

i would say driver in your distro is better. and distros like Arch already patched the polaris support in...

okay, so the patch linked that arch merged is this, it's literally as I said, a flag change, there is no difference, this isn't a feature patch it's changing default behaviour. it makes no difference, and Like I said, if it's not supported it's not supported so it wouldn't change anything even if you were right, which according to this patch, you aren't it's a damn flag change. this is nvidia levels of fuckery.

#From xuhuisheng
#at https://github.com/RadeonOpenCompute/ROCm/issues/1659#issuecomment-1041026624

diff --git a/utils/flags.hpp b/utils/flags.hpp
index 8f0228cc..2eaa47c5 100644
--- a/utils/flags.hpp
+++ b/utils/flags.hpp
@@ -245,7 +245,7 @@ release(bool, ROC_SYSTEM_SCOPE_SIGNAL, true, \
"Enable system scope for signals (uses interrupts).") \
release(bool, ROC_SKIP_COPY_SYNC, false, \
"Skips copy syncs if runtime can predict the same engine.") \
-release(bool, ROC_ENABLE_PRE_VEGA, false, \
+release(bool, ROC_ENABLE_PRE_VEGA, true, \
"Enable support of pre-vega ASICs in ROCm path") \
release(bool, HIP_FORCE_QUEUE_PROFILING, false, \
"Force command queue profiling by default") \​

[qarium]

okay, so the patch linked that arch merged is this, it's literally as I said, a flag change, there is no difference, this isn't a feature patch it's changing default behaviour. it makes no difference, and Like I said, if it's not supported it's not supported so it wouldn't change anything even if you were right, which according to this patch, you aren't it's a damn flag change. this is nvidia levels of fuckery.

#From xuhuisheng
#at https://github.com/RadeonOpenCompute/ROCm/issues/1659#issuecomment-1041026624
diff --git a/utils/flags.hpp b/utils/flags.hpp
index 8f0228cc..2eaa47c5 100644
--- a/utils/flags.hpp
+++ b/utils/flags.hpp
@@ -245,7 +245,7 @@ release(bool, ROC_SYSTEM_SCOPE_SIGNAL, true, \
"Enable system scope for signals (uses interrupts).") \
release(bool, ROC_SKIP_COPY_SYNC, false, \
"Skips copy syncs if runtime can predict the same engine.") \
-release(bool, ROC_ENABLE_PRE_VEGA, false, \
+release(bool, ROC_ENABLE_PRE_VEGA, true, \
"Enable support of pre-vega ASICs in ROCm path") \
release(bool, HIP_FORCE_QUEUE_PROFILING, false, \
"Force command queue profiling by default") \​

this is the patch who does enable it right this patch is a flag to enable it...
but think again is this the patch who does make it possible ? of course not many other patches makes it possible and this one is only enable it...

"this is nvidia levels of fuckery."

wrong amd published a complete guide in how to programm an driver for polaris because of this you can write your own driver.
believe it or not many polaris people in the opensource community did make many patches to the opensource version of rocm/hip..
and the patch i showed to you is only the final flag to enable this on polaris hardware.

this has nothing to do with nvidia fuckery...

[Quackdoc]

this is the patch who does enable it right this patch is a flag to enable it...
but think again is this the patch who does make it possible ? of course not many other patches makes it possible and this one is only enable it...

"this is nvidia levels of fuckery."

wrong amd published a complete guide in how to programm an driver for polaris because of this you can write your own driver.
believe it or not many polaris people in the opensource community did make many patches to the opensource version of rocm/hip..
and the patch i showed to you is only the final flag to enable this on polaris hardware.

this has nothing to do with nvidia fuckery...

is this a joke? its not supported, this is a do what you want but dont expect support cop out, maybe this is fine in the basements of neckbeards jacking off to anime tiddies all day. but this doesn't fly in the real world, either it's supported or it's not. end of story, no, "just do it yourself" no "rely on the community" it is either a supported use case or it isn't,

who gives a crap if AMD published a guide about it, do you think the clientele cares? you think a dudes corporate boss cares? hell no. the real world doesn't work like that, things don't magically get good because it's open source, unless rhel or someone who matters commits to supporting this, it doesn't make a lick of difference.

[qarium]

is this a joke? its not supported, this is a do what you want but dont expect support cop out, maybe this is fine in the basements of neckbeards jacking off to anime tiddies all day. but this doesn't fly in the real world, either it's supported or it's not. end of story, no, "just do it yourself" no "rely on the community" it is either a supported use case or it isn't,
who gives a crap if AMD published a guide about it, do you think the clientele cares? you think a dudes corporate boss cares? hell no. the real world doesn't work like that, things don't magically get good because it's open source, unless rhel or someone who matters commits to supporting this, it doesn't make a lick of difference.

if Nvidia does something like this because of closed source it is very bad for the customer.
if amd does this it is maybe not good either but it is less harmfull for you because you and others can still write the driver for yourself.

"maybe this is fine in the basements of neckbeards jacking off to anime tiddies all day."

red-hat makes billions of dollars every year by this just call red-hat and they make sure you get a professional polaris rocm/hip driver.

"but this doesn't fly in the real world"

red-hat earns money like this for like 20+ years...

"either it's supported or it's not. end of story, no, "just do it yourself" no "rely on the community" it is either a supported use case or it isn't,"

its a problem how companies make money AMD earns money by selling you the gpu and lets say 4-5 years later non of the money you paid is left in the pocket of AMD and then they drop official support.

but redhat works very different than AMD redhat makes money like this: after 4-5 years the angry polaris customer discovers that his gpu is no longer officially supported then he try it by himself the opensource version and is angry because it has bugs or does not work at all then they make a contract with redhat and then redhat pay the developer to fix the rocm-hip polaris support.

"who gives a crap if AMD published a guide about it"

i cann tell you this because i have good memory bridgman from amd one time said that the programming guide for their gpus where where popular and where downloaded over 75000 times in the first days of the release.
and you can be sure only experts like developers do download this or students in university who want to learn it.
and 75000 developers would outgun anything what nvidia has ...

"do you think the clientele cares?"

75000 downloads of the programming guide PDF in the first week...
and redhat did get 1000 of new contracts about driver development on AMD hardware.
so the answer to your question is. YES...

"you think a dudes corporate boss cares?"

the corporate boss at IBM/Redhat say: YES...

"things don't magically get good because it's open source"

in reality things become magically good by open-source.

"unless rhel or someone who matters commits to supporting this, it doesn't make a lick of difference."

IBM/Red-Hat really want a call from you to negotiate a contract to make the development of the ROCm-HIP Polaris driver.
but i can tell you this they already have customers like this.

[Quackdoc]

red-hat makes billions of dollars every year by this just call red-hat and they make sure you get a professional polaris rocm/hip driver.

where has rhel pleged to support polaris rocm?

"but this doesn't fly in the real world"

red-hat earns money like this for like 20+ years...

yeah, this is their buisness....

"either it's supported or it's not. end of story, no, "just do it yourself" no "rely on the community" it is either a supported use case or it isn't,"

its a problem how companies make money AMD earns money by selling you the gpu and lets say 4-5 years later non of the money you paid is left in the pocket of AMD and then they drop official support.



but redhat works very different than AMD redhat makes money like this: after 4-5 years the angry polaris customer discovers that his gpu is no longer officially supported then he try it by himself the opensource version and is angry because it has bugs or does not work at all then they make a contract with redhat and then redhat pay the developer to fix the rocm-hip polaris support.

so where is rhel or someone pledging to support this? if not then it doesn't fucking matter, not our business to pay rhel to support AMD's shitty projects when we can just buy nvidia and get support that dwarfs it

"who gives a crap if AMD published a guide about it"

i cann tell you this because i have good memory bridgman from amd one time said that the programming guide for their gpus where where popular and where downloaded over 75000 times in the first days of the release.
and you can be sure only experts like developers do download this or students in university who want to learn it.
and 75000 developers would outgun anything what nvidia has ...


"do you think the clientele cares?"

75000 downloads of the programming guide PDF in the first week...
and redhat did get 1000 of new contracts about driver development on AMD hardware.
so the answer to your question is. YES...

The answer is a hard no. we don't live in a world where everyone can pay rhel or collabora to support what they want. unless rhel is pledging to support polaris, it doesn't matter one lick, 75000 downloads of a programming guide means very, very little. 1000 contracts is very, very few, last I tested rx580 support didn't work on rocky, which means it likely doesn't work on rhel, so no one is supporting it there

"you think a dudes corporate boss cares?"

the corporate boss at IBM/Redhat say: YES...

im glad we live in a world where everyone has their own personal rhel teams working on whatever they need.

"things don't magically get good because it's open source"

in reality things become magically good by open-source.

"unless rhel or someone who matters commits to supporting this, it doesn't make a lick of difference."

IBM/Red-Hat really want a call from you to negotiate a contract to make the development of the ROCm-HIP Polaris driver.
but i can tell you this they already have customers like this.

yeah well, until we see support for it out of box, it's unsupported. until then, i'm migrating all of my clientele to nvidia.

EDIT: and mind you these issues are JUST rocm, there are still issues with things like the hevc amf, and missing vulkan extensions

[qarium]

where has rhel pleged to support polaris rocm?
yeah, this is their buisness....
so where is rhel or someone pledging to support this? if not then it doesn't fucking matter, not our business to pay rhel to support AMD's shitty projects when we can just buy nvidia and get support that dwarfs it
The answer is a hard no. we don't live in a world where everyone can pay rhel or collabora to support what they want. unless rhel is pledging to support polaris, it doesn't matter one lick, 75000 downloads of a programming guide means very, very little. 1000 contracts is very, very few, last I tested rx580 support didn't work on rocky, which means it likely doesn't work on rhel, so no one is supporting it there
im glad we live in a world where everyone has their own personal rhel teams working on whatever they need.
yeah well, until we see support for it out of box, it's unsupported. until then, i'm migrating all of my clientele to nvidia.
EDIT: and mind you these issues are JUST rocm, there are still issues with things like the hevc amf, and missing vulkan extensions

right you can always choose to just buy nvidia and use CUDA they have a defacto monopole on GPU Compute.

to make business agaist this monopole is very very hard believe this.

and again keep this in mind: i did read multible persons who did report here in the forum that they did successfully run gpu compute apps on Polaris with the opensource version of ROCm-HIP...

"75000 downloads of a programming guide means very, very little."

the people at amd where in fact very surprised by this big attention. they thought they will have 1000-2000 downloads... and in the end it was much much more.

"1000 contracts is very, very few,"

last time i checked the price the cheapest contract per computer was 300-400€ per year...
1000 contracts means 400 000€ in one single week? i would say not bad.

"and mind you these issues are JUST rocm, there are still issues with things like the hevc amf, and missing vulkan extensions"

these kind of problems are everywhere... most of the time believe this only the big monopoles have the resources to invest a signifikant amount of money to fix it. but by this logic you always use a intel cpu and a nvidia gpu and a microsoft operating system...

to make a successfull bussiness outside of these monopoles is very hard work and you did discover this does not only mean hard work for amd but instead also hard work for you as a customer.

but hey i use linux and amd hardware... i don't care about intel or nvidia or microsoft.

[Quackdoc]

right you can always choose to just buy nvidia and use CUDA they have a defacto monopole on GPU Compute.

to make business agaist this monopole is very very hard believe this.

and again keep this in mind: i did read multible persons who did report here in the forum that they did successfully run gpu compute apps on Polaris with the opensource version of ROCm-HIP...

"75000 downloads of a programming guide means very, very little."

the people at amd where in fact very surprised by this big attention. they thought they will have 1000-2000 downloads... and in the end it was much much more.

"1000 contracts is very, very few,"

last time i checked the price the cheapest contract per computer was 300-400€ per year...
1000 contracts means 400 000€ in one single week? i would say not bad.

"and mind you these issues are JUST rocm, there are still issues with things like the hevc amf, and missing vulkan extensions"

these kind of problems are everywhere... most of the time believe this only the big monopoles have the resources to invest a signifikant amount of money to fix it. but by this logic you always use a intel cpu and a nvidia gpu and a microsoft operating system...

to make a successfull bussiness outside of these monopoles is very hard work and you did discover this does not only mean hard work for amd but instead also hard work for you as a customer.

but hey i use linux and amd hardware... i don't care about intel or nvidia or microsoft.

this monopoly is because AMD and intel don't have any competitve products, and if this is how fast AMD is willing to depreciate their support, they just keep exacerbating the problem

[qarium]

this monopoly is because AMD and intel don't have any competitve products, and if this is how fast AMD is willing to depreciate their support, they just keep exacerbating the problem

right... but on the other hand business is business and as soon as you as a company pay more than you earn it is no longer a business instead it is charity...

in the market AMD is really in means not the gpu compute market instead of the gaming market most people buy new gpu after 5 years.

if you have a desktop and you really want ROCm-HIP Blender 3.3 support you can buy a used Vega64 on ebay for like 250€ ... for most of the peoplem not a big deal.

i for myself i plan to upgrade from vega64 to a amd 6800xt or something like that.