probably not, this company aint trustworthy
1 Like
"This converts radv_CmdBuildAccelerationStructuresKHR to a simple shim that pushes the actual build commands to a queue, where they are accumulated and dispatched as late as possible.
This helps especially with games that don't do any build command batching of their own. For example, it triples the performance of Hitman 3."
This is Vulkan. Cycles uses HIP to render. The 3x boost is in a single game. As the quote says, the boost is from batching commands to build the BVH, which even if Cycles did use Vulkan (very unrealistic), BVH building is not the majority of time most people wait for with Cycles, and even if it were, I have to wonder if batching would even make enough of a difference for Cycles, or isnât already implemented where it makes sense for it to be.
2 Likes
Thanks, yeah the question was related only to Vulkan.
I have moved away from AMD, but I just found that on Linux, now Blender has âHIP RTâ. I thought it was Windows-only? Linux support was added in Blender 3.6.2?
Also, I know that HIP + RDNA2 + Linux has been crashing in viewport for over a year. How is the situation with RDNA3? Does it not crash on Linux?
Would be curious to know the current HIP state on Linux as well.
As far as I am aware HIP RT is windows only and RDNA3 suffers from the same crash as RDNA2.
Both HIP RT and the cycles crash issues might be resolved in 4.0. However itâs now years after RDNA2âs release and Blender or AMD havenât been able to resolve the technical issues. Smart to move away from AMD if you use Blender (Windows or Linux).
2 Likes
It makes me think that, when there is a technical issues that canât be solved reasonably quickly, there is probably a hardware issue in the GPU itself. Itâs not an issue that can be fixed with a software/driver upgrade, etc.
This is why I havenât used a AMD GPU in years, they are fine for gaming, but donât expect them to be very good for other applications. Can you even run AMD GPUs headless? That was the major problem I had about 12 years ago with them. Hopefully that has changed by now?
Whatever the issue is AMD couldnât get Cycles to work at all or without major issues on Linux in Blender for the entire 3.X series (3.0, 3.1, 3.2, 3.3, , 3.4, 3.5, 3.6) .
AMD can run headless. Previously âROCmSupportâ confusingly said they would not support Blender as it used a GUI in this GitHub Issue, its all a bit of a messâŚ
2 Likes
Hello, Iâve been following this discussion for a bit,
Iâm getting Amd rx 480 sapphire nitro 4gb, Iâm wondering if it is even supported with blender? or will blender forcibly use the CPU?
Your gpu is gcn4. HIP currently supports gcn5 and above and for HIP-RT you need an RDNA2 as those were the first ones with hardware raytracing. But for working in blender itself a gcn1 is enough.
2 Likes
When buying an older card, an Nvidia card would be a better choice. If possible look for 1060 or something newer like a 1650. When Blender 3.0 came out, there was a cycles rewrite, resulting in a hard cut off for older AMD hardware. There was no such cut off for Nvidia cards.
2 Likes
So again sorry, will it be in any use or blender will ignore it in rendering even if I chose gpu complete?
you think Nvidia would be better in the same price level?
Since Blender 3.0, the oldest AMD card that is supported should be the Vega 56/64. RX 480 will not be able to render at all. What cards from Nvidia are in your budget?
1 Like
Consider that if you pick a slow card because of price you will be spending much more money on energy bill that otherwise with a faster card you would not. To add to this also other concurrent spending: lights turned on, etc.
For same work quantity.
1 Like
Pretty much any Nvidia card would likely be better. For rendering, if you can manage it, then even the lowest end RTX card you can get would be light years ahead in Cycles.
3 Likes
Avoid AMD GPUs for blender and Content Creation in general. The Software Support is TERRIBLE.
To AMDâs defense, Iâve gotta say that outside Blender open data benchmarks, the âreal worldâ benchmarks seems to justify pricing of most of GPUs regularly tested in Blender, if nothing else. Sure, there are some real treats for their price (4070 for example), but overall it doesnât seem to be that bad. Not for itâs price.
I just dived into rendering in general, so I know very little about all this yet. But when watching numbers from actual scene testing and when comparing them to my 7900 XTX performance, I was quite a surprised (after initial sh*t talk in regard to AMD). (edit.: I was talking about comparison of 7900 XTX to 4090 mostly)
See for yourself:
1 Like
It is good to finally have all three companies offering a product. Gone are the days when you had to pick your step on an Nvidia ladder and accept the mandated âpenaltyâ of arbitrary product segmentation if you canât stretch to buy the top card. Now you can match hardware quite closely to the desired budget and personal workload.
I am sure the 4090 is a fine card, but most of the time, you canât find one below 1800 euros. A 7900XTX offers good speed and, crucially, the same amount of memory as Nvidia counterpart, while being nearly half the price. Also, with todays denoising technologies, the rendering can be completed relatively quickly. Most of my daily renders take only around 3 minutes per shot and a lot of that time (about 40s) is taken by the BVH building. A card thatâs twice as expensive, would realistically shave, maybe a little over a minute, of the render time, which does not seem like a good return on investment.
Other people may find value in having the fastest, even if the most expensive, hardware. It is all very individual. Overall, itâs good to have a market situation that is no longer in the grip of a single vendor.
2 Likes
The video has 4 scenes tested where the 4070 was 2.1, 1.8, 1.3, 2.8 times faster than the 7800XT. Blender open data tests up to 7 scenes (maybe?) and states 2.5 times âfasterâ in 3.6. The results appear to be similar given that all scenes are different.
Half the performance per dollar in cycles is a tough sell. However for hobbyists it probably doesnât matter.
1 Like