Will a dedicated GPU speed up my ON1 2023 significantly? is it worth the cash?
Hi all. Happy New Year.
I've been using ON1 2020, 2022 and now 2023 as a great alternative to LR and bits of PS. I bought a gaming type windows PC a year ago to speed up my workflow as I regularly edit sets of several hundreds photos. I use HDR, NR etc.
ON1 2023 is usable but noticeably slower than 2022. It can take several seconds to switch between images or between edit and browse mode. I like the new tack sharp and NR, but these too are quite slow. I run 3 monitors and use ON1 across two of them. My PC spec is: AMD Ryzen 5 5600G processor with onboard Radeon 7 graphics; 500GB SSD (for OS & system files), 2GB SSD for storage & photos; 32GB DDR4 RAM; 650W power supply; MSI MPG B550 gaming plus motherboard.
At the time I didn't buy a separate graphics card. I am now considering buying either an RX 6600 XT or RX 6700 XT AMD Radeon graphics card. There are many such cards at a variety of prices.
As far as I can tell, from online searches and comparisons, they all seem to have a very similar performance from the basic AMD 6600 or 6700 GPU within.
I have three questions:
1) Does anyone know (or how can I estimate) how much either of these processors might speed up ON1 2023? I.e. will I get a good and noticeable bang for my bucks (£250 - 400 GBP, which is $300 - 500 USD).
2) What else should I try first before buying a GPU? I.e. how else can I optimise ON1 2023 and/or my PC? I have played with the system preferences in ON1 2023. It is running on fast SSDs. How can I prioritise how ON1 2023 gets priority of system resources as a process?
3) I have read that using an nvidia RTX card is not a good idea as it might clash with my AMD processor. That is why I am looking at RX 6##0 XT cards. Does anyone have any experience with these cards and ON1 2023, or want to suggest any other cards I should consider? Many thanks.
PS Brian Lawson has suggested the following to me on another forum:
* moving the program's Scratch space to its own fast SSD.
* put my PerfectBrowseCache there too
-
Hello and Happy New Year!
I have stayed away from all in one processor/graphics card due to possible issues with performance. At least in the past the GPU and possibly the CPU performance has been compromised but my knowledge is dated by a few years. I also don't think the CPU and GPU combination matters much as I believe a lot of gamers will buy the Ryzen CPU and pair it with the Nvidia GPU. One thing used to be true is the combo CPU/GPUs used the same memory and you were (are) splitting the amount of memory in the computer between the 2; you won't have that with a GPU with on board memory - this may actually be your problem!
My own WIndows computer runs with a Ryzen 3700X with 32 Gb memory and a RX580 with 8 Gb memory and this is what I have found. When the system preference was at 50% GPU usage (4 Gb) the program ran slow, as soon as I put it to max (VRAM at 80%) the system was fine... my takeaway was 2023 needs a lot of power from the GPU to run smoothly. I tend to watch my system resources with task manager to see what is being affected. I would advise trying that to see where your being hung up and try tweaking a little. I can say that watching task manager was interesting with 2023 as at 50% the GPU was only used for a second or less but at max it was being used for maybe 4 or 5 seconds at a time.
Brian has been saying use a separate scratch disk for years and I do have RAW going to it's own SSD scratch disk based on his suggestion. Other things that I have done is use Windows performance enhancement for Ryzen CPUs, if you have the AMD GPU software (adrenaline) point it to 2023 as a game to have it exclusively run it. I have it pointed to all my photo software. One thing I have found out today on a different forum for a different photo program is AMD may (or does) have issues with OpenCL and Nvidia does not ... I'm not an IT person but it's out there in the computer world. :)
My thoughts are if you try tweaking and get nowhere then I definitely would go with a new graphics card. I probably would want a card with minimum 8 Gb but I'm thinking 10+ Gb to future proof. 650W power supply should be OK but a more powerful GPU has more power cables going to it than just the motherboard so be aware of that. If you do put a graphics card in you need to somehow tell Windows or the motherboard to use that card vs the onboard GPU.
I hope this helps!
0 -
I only have two M.2 slots and already two M.2 compatible SSDs in them. One is the C: drive (Kingston 500GB SSD), the other is a 2TB Seagate FireCuda 510 SSD.So what would be the best solution for a scratch disk?Should I buy another M.2 SSD, perhaps another 500GB SSD to use for the On1 Cache, Scratch disk (and also my Microsoft OneNote Cache, as that hangs too). This would mean moving the 2TB Seagate SSD to a Sata port if that is possible..FYI:1) Looking st the System Performance in Task Manager, it seems to be the CPU and sometimes the SSDSs that are maxing out. The AMD onboard GPU is rarely maxing out, so it would seem that buying a discrete GPU may not help much, although perhaps this would take some load of the CPU?2) I have moved my ON1 Cache and Scratch folders to the E: drive (Seagate 2TB). This seems to have speeded things up a bit. I also moved the main OneNote files off the C: drive.Maybe I just need a faster CPU? The AMD Ryzen 5 5600G processor is supposed to be fairly quick. Or maybe I need a PCI 4.0 SSD as the Firecuda is only PCI 3.0?0
-
Andrew,
How much memory do you have in the computer? The specs of your CPU are comparable to my 3700X except I have 2 more cores and 4 more threads - not a big deal. The GPU is what 2023 mainly needs and if you aren't using it a lot then it may be indicating there isn't a lot of memory available to do the task. I am only saying that due to my experience of 4 Gb vs the 80% VRAM usage, I may be totally off base. As far as drives, there are 2 different types of M.2 drives, PCI and SATA and the PCI is the fastest available. A PCI 4.0 anything will not run at the speeds said unless the motherboard can handle 4.0 - I don't know your system so I can't comment if it will or won't work.
Going back to drives, I'm going to say that I have all my data on a spinning drive and all the other stuff on faster drives. I have a M.2 PCI drive for my boot up and programs and a SSD for the scratch disk. I would imagine a B550 board can handle another SATA SSD or spinning drive, data does not have to be super fast and the amount of storage available on a spinning drive vs price is great, just make sure you get a 7200RPM drive for fastest retrieval if you go that way.
Some other things - I'm running Windows 10 and without 2023 open it is running 6 Gb of memory. Turning 2023 on now has about 8 Gb memory without doing anything, I have seen RAW 20XX use a lot of memory when we clone, add layers and what not but haven't used 2023 enough to track its memory usage. I personally have gotten a huge file using a lot of layers that an earlier version on my computer with less memory couldn't handle. From what I've read/heard the GPU is really what crunches numbers for the computer and if I got it right the GPU memory is quicker than the system memory because of the way it is set up and handled. Something that is often not thought about - heat from the CPU/GPU. If it gets too hot it will throttle down to protect the chip. If I run Cinebench on my computer my CPU will get to 80 F and stay there without a GPU installed on it.
If you don't have a lot of memory in the computer that may be the first place to look at. Going from 8 to 16 or more may be that's all you need. Having 16 and having the computer at 8 with 2023 open only leaves you with 8 more including video processing. If you are running 16 already, go to 80% VRAM in 2023 and see what happens. If you have enough memory and it still sluggish I would say then video card. I'm not thoroughly convinced a faster buss speed, going from PCI 3.0 to 4.0, is going to do much. Overclocking won't do much either.
I have a comment and it may seem off the wall but I am SO guilty of this myself ... how much money should we throw at a $80 program to have it work properly on our computer? Being guilty myself of this really has me thinking as to what happens when... I don't want to spend $1000 every few years for a computer to run a program. I bought this computer 2 years ago for a good amount, last Christmas I asked for 16 Gb of memory to get to 32Gb and I asked for a M.2 PCI drive because I had the slot. I really only surf the web, do some home spreadsheets and documents, watch some YouTube channels and do photography and out of that RAW was having issues with my last computer. I will say I am somewhat of a techie when it comes to computers but not super into it and I don't like spending money on computers that work great except for a single problem; but I did!
These are my thoughts, not sure if my logic make sense or not. I can say that on my old computer going from a 516 Mb to 2 Gb memory card did make a difference in the past and going from using 50% to 80% of VRAM on this computer made a difference in 2023.
0 -
Thanks Vinny
I have 32GB of Kingston Fury Beast DDR4 3200KHz RAM
The AMD Ryzen 5 5600G is 6 core, 12 thred running at 3.9 FGHz. It has 19MB cacheThe MSIB550 board can handle 6 SATA SSDs - I have one which I use as a backup. This is a 6GB Seagate Barracuda it is SATA 6 Gb/s 5400 rpm. I guess I could try using this as the scratch disk. I have seen reviews pairing the 5600G with a B450 board (as adequate), so the B550 should hopefully be up to the job.
The MSI B5500 board has 2 x M.2 slots for storage and 6 x SATA 6GB/s ports. It also has the following expansion slots:
*1 x PCIe 4.0/3.0 x16 slot (PCI_E1)
*1 x PCIe 3.0 x16 slot (PCI_E3)
* 2 x PCIe 3.0 x1
I have moved the ON1 cacheand scratch disk off my C: drive:
* my OS is running off SSD 1 (M.3 500GB)
* ON1 data is residing on SSD 2 M.2 2TB, which is whwere I have moved the Scratch disk and Cache for now.
* I could try moving the Scratch disk and Cache to HDD1 (5400 rpm)
Or maybe the ON1 data should be on HDD1 and the Scratch/Cache on SSD 2? I could experiment...
Or, can I use the PCIe 4.0 expansion slot for a fast SSD rather than keeping it available for a GPU? Given that I have 32GB of fast RAM and the GPU doesn't seem to be the bottleneck...
I have a decent 650 W PSU and a couple of extra fans, so hopefully heat and power aren't an issue.
Like you, I don't want to throw too much money at this. I bought a good spec custom gaming PC and apart from not buying a separate GPU (due to the high prices), I didn't skimp and thought that the 5600G was a good and reasonably future-proofed option. It seemed to be the case until 2023 required A LOT more resources.0 -
OK, you gave out some more info. As far as the cost, my son wanted a new GPU last year but actually bought an open box computer for a little more than the cost of a GPU ... prices were crazy!
32 Gb should be plenty memory, can you dedicate 8 GB strictly for video? I think in WIndows you can tell it to do that. Have you updated the driver? I had an issue with the AMD software not seeing an update so I went to AMD directly and downloaded it. Did you tell Windows to use the Ryzen performance mode? Did you point the AMD graphics software to 2023 to use it exclusively. Playing around with the 2023 preferences - did you put the system VRAM resources to 80% and play around with the check boxes? Do not have a scratch disk on slow spinning hard drive. Typically you want to put the scratch and cache on a drive that doesn't house the program but I would think a M.2 PCIe drive should be able to handle the data transfer of both easily. I'm running a B450 board so B550 is definitely fine.
The heating issue is not noticeable unless you are looking for it. Case fans won't cool the CPU well. My computer was store bought so I have the lowest type of CPU cooler that was installed. I have played around with the fan speeds on the CPU but other than getting a new CPU cooler there's nothing I can do. I will run the AMD overclocking program along with Cinebench to see where the temp goes. I do hear my CPU fan accelerate at times when I use 2023 so I know my CPU is getting hotter.
At this point other than playing around with system settings in 2023 and WIndows I'm not sure what else may help without the purchase of the discrete GPU, but what happens if it still has issues? I have been around enough here to say that at times it seems that RAW just doesn't play nice on some computers. A couple of years ago a person had a monster AMD system and RAW refused to work speedily and it seemed the person knew computers really well.
I wish I can be of more help but I don't wat to say "yes, buy a GPU and your problems will go away". It may as it does seem people with low memory GPUs even if it meets the minimum requirements are having issues with the program. Maybe after trying the above tweaks buy it from a store you can return it if it doesn't work.
0 -
Andrew,
I’m pretty much in the same boat that you are. I have the Ryzen 5700G, which has 2 more cores and 4 more threads than the 5600G, and is just slightly faster. I have a B450 motherboard from Asrock, a generation older than yours. I upgraded to this APU from the 2200G. While this was a big step up in both CPU and iGPU power, it still leaves PR moving far slower than I consider acceptable, even with PR 2022. I haven’t spent much time with PR 2023 yet, but it sounds like it needs a lot more GPU than any current integrated GPU can provide. I have also been looking at the AMD RX 66xx or 67xx cards.
(The Ryzen 2200G and an even older CPU had no problem with Lightroom or processing 4K video, so PR is clearly more compute intensive than either of the other programs .)
On my Ryzen 5700G, I saw no noticeable performance changes moving the GPU strength slider in On1.
I haven’t heard anything specific about AMD processors not working well with Nvidia cards, but I also have no experience using PR with either maker. I’ve always run AMD processors, and it seems to me like a large percentage of complaints about PR’s performance come from AMD users, including myself. (I wonder if they are compiling specifically for Intel processors, and ignoring AMD optimizations.) On1 states explicitly on their website that they haven’t tested with any AMD video cards made in the last 4+ years, which is really unnerving.
To Vinny’s question, it doesn’t seem that you can allocate a specific amount of memory to the integrated GPU on all motherboards, and it’s not clear that you would need to. It’s supposed to be handled dynamically on newer Ryzen models. With 32GB of RAM and little else running on your system, RAM allocated to graphics shouldn’t be a problem. However, system RAM is MUCH slower than VRAM, so not having a graphics card does affect memory access speeds, along with the processing power a dedicated GPU can supply.
You probably DON’T want to use the spinning disk as a cache for PR. From the amount of disk writes that everyone says it does for caching, fragmentation on the disk will really hurt its throughput. An SSD isn’t impacted by fragmentation, so that's your best bet for PR's caching needs
My OS, Windows 10, is on an NMVe. My scratch space and caches are all on an SSD, and my images are stored on a physical disk.
I’m not positive about the specs for your motherboard, but a quick search shows that it has only 2 M.2 slots, so you may not have room for a 3rd. My scratch space drive is just a plain SATA SSD, not an M.2 device; this is still much quicker than a physical spinning drive. Despite this and following all of the suggested performance settings for the OS and PR, I’m still unhappy with the performance of PR. I’m also hoping that a GPU solves this once and for all for me.
Like Vinny, I also question the need to throw more money at what is truly a fairly high-end computer to get it to make an $80 program work acceptably. What happens if after spending $300-$500 on a GPU, the performance is still not acceptable? And even if it is, the $300 card might not be powerful enough in a couple more years if more AI processing is added to PR as many expect. On1 won’t publish any information on expected performance for GPUs.
I don’t fully understand the complexities of how PCIe slots are allocated on motherboards, but as you add more devices, you can reduce the performance of others, or end up disabling some altogether. So adding a PCIe SSD could limit the performance of a future GPU addition.
I plan on adding a video card in the near future. I'm really hoping that it helps.
0 -
Hi Carl. Thanks. I concur with what you wrote. I am on the fence as to whether or not to buy another disk and a GPU. Not convinced the GPU will have a big impact unless it does so by freeing up the CPU.
0 -
Hey guys,
Carl, Thanks for adding to this!
Although I am not 100% sure that a discrete GPU will be successful I do think it will be. I also think that Nvidia was the go to GPU for gaming for quite a while and AMD is now coming into the light. I am not a complete computer geek but have read that the newer CPUs and GPUs in AMD can talk to each other to make the computer more cohesive but not sure if your chips and motherboards qualify for that. Getting either manufacturer should work fine with an AMD CPU. Unfortunately the only thing we can use to see how the CPU/GPU combo works together is gaming benchmarks; I wouldn't be too concerned that the frame rate on a game is 10 FPS slower on one GPU vs another.
My suggestion is to buy one where it can be returned if it doesn't work as it should, a place like Best Buy or Amazon may not have a restocking fee. I looked up on Newegg and they have a RX6700 XT for $419 that has 12Gb of DDR6 memory, DDR6 I believe is 2x as fast as DDR5 and the amount of memory should be enough for a while (hopefully ;-) ).One installed then you have to point the computer to it, I found this: https://www.technewstoday.com/how-to-use-dedicated-graphics-card-instead-of-integrated/#:~:text=Switching%20to%20Dedicated%20GPU%20Without%20the%20GPU%20Software,this%20for%20all%20the%20programs%20of%20your%20choice.
I didn't read it through so hopefully it is detailed enough to follow. I will also say to make sure your GPU software points to RAW, I have 2021 & 2023 installed and it points to both of them. It also points to all my photo software. In AMD you do it under games, in Nvidia it is under 3D.
If the initial installation of 2023 is any indication of how well it will run with a dedicated GPU and more memory I think you both will be pleased, going from the program only using 4 Gb to 6.4 Gb (80% VRAM) made a huge difference. My RX580 is old now in technical terms and that Newegg card I mentioned is supposed to be 100+% faster on doing things per online benchmarks. Being faster and using 80% VRAM (9.6 Gb) should make a big difference. But I definitely would try and get it somewhere that it can be returned without a restocking fee "just in case"!
Good luck with this!!!
0 -
Hi Carl & Vinny
Thanks again. ON1 sent me instructions on how to update the GPU driver, which I followed. It didn't improve things. In fact it stopped my AMD software from working. So, yesterday I uninstalled the AMD driver and software and reinstalled it from the AMD website. Both are now working (though subjectively, ON1 2023 feels a little slower?).
I have pointed the new AMD Adrenaline software at ON1 2023, by adding it as a game. There doesn't seem to be any clear way to optimise performance for ON1 2023. Is it just a case of adding it?Last year I found an article which explained how to prioritise a process on the CPU. This was a function within Windows 10. I haven't managed to find how I did this although was was able to go into the option to make ON1 2023 a High Priority. Not sure whether this helps.
Cheers, Andrew
0 -
Andrew,
I looked at my Adrenaline settings under gaming and the only thing I have enabled is Surface Format Optimization under the advanced setting. I think when you point Adrenaline to ON1 as a game it uses it's power to make it run as a game vs just a program; I may be wrong on this though. I used to point the Nvidia software to ON1 in it's 3D area and I did see a difference in performance. Personally, I want to give ON1 all the edge for speed I could just in case it was going to chug along.
I think what you are seeing what Carl said about the VRAM memory being quicker and crunches the numbers faster. The crypto people were buying them up for this reason. 16 GB of computer memory is slower than 16 Gb VRAM on a GPU.0 -
After much (internal and external) debate, I have decided to buy a 500GB SATA SSD to use as a scratch disk and cache, plus a dedicated GPU - PowerColor Fighter AMD Radeon RX 6600 Graphics Card with 8GB dedicated GDDR6 Memory, 28 Compute Units, 32 MB of all new AMD Infinity Cache, Clock 2044 MHz | OC Clock 2491 MHz and 1792 cores.
I bought the GPU as much to try to take the apparent load of the CPU (AMD Ryzen 5 5600G), which gives the impression of being the bottleneck while using ON1 2023. I might be wrong. Hopefully, these two things will speed things up significantly.
The SSD is a Crucial BX500 480GB 3D NAND SATA 2.5 Inch Internal SSD - Up to 540MB/s bought from Amazon.It would be really nice if ON1 could give a lot more clarity on these issues, given the processor/disk/GPU hungry direction that the software is going. Surely, there could/should be advice and maybe even comparisons for both Intel and AMD CPUs and AMD and Nvidia GPUs.
Wish me luck0 -
I think you will find that the GPU will help a lot, just make sure you now tell Windows to use the new GPU vs the on board one. I have a 8 Gb older card and it works great so a newer card should be better. I will say to make sure you tell the ON1 system resources to use 80% VRAM on the new card.
The scratch disk being by itself should also help somewhat, I think most of the improvement will come from the dedicated GPU.
Good luck! I hope it exceeds your expectations!!!
0 -
Hi
I would be insterested to know if using other drives are really helping.
I can manage to test on another NVMe PCIe4 on a "PCIe 2" slot on my main board ( system disk is on a NVMe pci e 3 slot).
My two cents with GPU : for a 45 Mpix image (Raw from Nikon), testing a Nvidia GTX1060 with AI Denoise takes around 12 s to apply filtering. Using a RTX 3070ti 8GB vram drop down calculation to around 4 seconds with ON1raw2022.
similar result with ON1raw 2023 and RTX card (I did not reproduce test with GTX card). Different behaviour in a sense that complete calculation is applied when hitting Apply button at the end of the process in ON1 2023.
- When using AI denoise and / or AI tack sharp on a picture, then you are going back to browse, culling photos , when you are reaching one with AI stuff already applied will take more time to display with a little popup window saying "loading photo" after a few seconds (stangely, because AI stuff should not be visible in browse mode).
-Main performance problems come when doing local /masking brush (getting slooowww, stuttering, buggy) or culling among many photos getting slower and slower (huge RAM eater, never give back resource).
My base configuration is AMD Ryzen 2700x (8cores around 4GHz - 2017) /32GB DDR4/ 512 GB NVMe system drive on a PCIe 3 M.2 slot- Asus X470 Main board - APP, OS and Scratch zone in appdata default location, Imported files are also on same drive (will have to change that !) , and all setting at 80% , video card strengh at max.
All with latest BIOS/AMD chipset driver/ Nvidia RTX creator driver from this month, APP reinstalle d in Admin mode , system in performance mode, CUDA core are affected to ON1 app in Nvidia control center.
Best regards.0 -
It seems counter-intuitive but you might want to lower the Video Card strength. That slider tells the program how much GPU to expend on rendering the image. The GPU also tracks the brush movements. The more time it spends rendering what you're brushing over, the less time it has to track the brush strokes.
0 -
Thanks Brian
...Combined with temporary removal of fast preview also seem to help when picture is zoomed and brush is small (the smaller the slower it is usually).
Just tried to put Browse cache and scratch on other NVme drive and noticed many read /write activities, especially in perfectbrowse cache, with relatively low file size but numerous files. This has an impact on drive performance usually. Separating system disk and data/scratch disk is a help to I/O queue management. I will try this more extensively with heavy culling but at first sight it seems better.
SO at the moment :
- All drivers up to date as already mentionned. "AMD power management" set to balanced after AMD chipset driver update (instead of windows "performance" profile which can be counter productive on AMD platform) , driver from AMD more recent than the one from Asus (that profile was not showing up in windows power management prior update ! ).-Graphic adapter driver up to date (from NVidia, not the builder) and ON1 reinstalled as admin user
-system perf/Ram set to 80% in ON1 setting - Graphic strengh set to around 60%-70% (estimate)
-Browse cache and scratch on other drive (fast one NVMe, I would avoid USB attachement on a worksation, for laptops, modern ones often have extra NVMe slot so )
- Bios : cpu setting and memory setting should be optimal (leaving most things in Auto freq and memory profile set at its nominal speed 2666MHz for 4,1 GHz cpu speed - no manual overclocking)
This hopefully should save me some money and earth resources if I don't replace the whole workstation ( I may still replace cpu with a more recent AMD 5900x which is supported by my MB, but limiting the cost of a full replacement).0 -
Well.... After investing in what I perceived to be a well specified gaming PC a year ago, I have to say I am very disappointed.
Having gone the extra mile to add a dedicated GPU and a scratch disk/cache SSD, the result is that ON1 2023 is still running just as slowly and there has been little or no discernable improvement.
The major workload and the bottleneck still appears to be the CPU (AMD Ryzen 5 5600G).
All this talk about AI functionality and needing a dedicated GPU seems to be very misplaced.
Why is it that ON1 2023 has a significantly longer delay when switching from Edit to Brose mode? What does that have to do with AI? Why is just that element so much slower than previous versions of ON1?
It is frustrating, as otherwise, ON1 functionality is great. The new functionality is useful and convinced me to upgrade, but not at the expense of core functionality and speed....
I really think that ON1 need to give this a lot more consideration and need to provide a lot more information, help, much better details of recommended hardware specs and maybe even the option to switch off certain functionality if it isn't required so as not to sacrifice so much speed.1 -
The delay in moving from the Browser to the Editor is because the new SSAI feature analyzes the photo to create the masks for that image when it is first opened. If the image has been previously edited with NoNoise AI or Tack Sharp those AIs also have to do their thing before the photo can be presented on screen for editing.
0 -
The delay in moving from the Browser to the Editor is because the new SSAI feature analyzes the photo to create the masks for that image when it is first opened. If the image has been previously edited with NoNoise AI or Tack Sharp those AIs also have to do their thing before the photo can be presented on screen for editing.
0 -
Surely, the SSAI feature only needs to be performed once. If the image hasn't been edited, then can't it remember the previous mask? It seems to be slow every time.
0 -
No, my understanding is that it is performed each and every time.
0 -
Seeing the flurry of additional comments, I'm not sure if I can add anything but christophe ducrocq mentioned a 45 Mpix image which could be an issue for some setups. A Ryzen 2700 or a 8 Gb GPU may not be sufficient for what 2023 needs to deal with a file of that size. My D7200 is 24 Mpix and 4 Gb (50%) GPU memory was sluggish and seems OK when at 80%. If it is linear then a 45 Mpix camera will possibly need a GPU with more than 8 Gb and with that much data - more cores, faster CPU and/or more cache may be needed to speed things up.
Andrew - Are you seeing that Windows and for that matter ON1 is using your new dedicated GPU? I can't see 2 less cores on the CPU being that big of a deal but I could be wrong. I do hear my CPU fans ramp up at times but maybe for 4 or 5 seconds and when I do track it doing any heavy or long photo manipulation it doesn't seem to affect the CPU all that much. I can say that cataloging is CPU intensive vs GPU but ON1 has always been like that.christophe ducrocq Interesting about the AMD performance in power management as I have Windows running that since it is supposed to help with AMD performance. I also have my bios tweaked a little to run the CPU and memory a little faster than stock just because this is the first computer I can tweak! :-)
I will say after seeing threads like this I question myself as to how much longer will I upgrade software. Having a 2 year old computer becoming "obsolete" due to software requirements is a little over the top for me. I don't seem to have major issues yet with 2023 but I did have to upgrade for 2021 because I just - had to - upgrade the software. Hopefully there is a simple solution for you guys and things start working normally.
0 -
Vinny Giannino,
GPU is efficient on AI as it devided by around 4 the time needed to apply AI Denoise (on12022 -tack sharp was not yet here, but I have consistent time with ON12023 and Denoise+ tackSharp), when I switched from and old GTX1060 to a RTX3070ti . This is more or less the same ratio if you check CUDA cores available on both cards. I won't give this as a rule but, for me, the more cuda you have the less you wait AI Denoise/TAcksharp.It is really necessary to boost GPU, and I would have probably bought a 3080 if it was not so expensive (and now 40xx series are out 4070 ...). From a GPU point of view, Monitoring ressource don't show a saturated activity so far (and hope it will not for some time due to high prices!).
As far as I could see 8GB Vram is enough, I never saturate it.
For CPU (and yes my files are big :) , All 8 cores (dual threaded) are used periodically for a few seconds (can hear the fans spining up and monitoring CPU resource is just showing a spike in activity). Most of the time this generation cpu is enought but app seem to use all what's available when applying filters/similating presets or generating previews....The more you have the more it works, so reducing this number is not a good idea (this would lead me to Intels 13700 or for really big loads13900. But 13600 is not bad too). So, I agree with you (and I somehow figured it when I switched to 45Mpix files), 5 years is getting old for my 2700x CPU.
Also, I think the problem is more in the IO bus design of PCs. My MB is PCiev3 based (and PCIe v2 for some slots), My graphic adapter is PCIe4 (so it is not completely used). Doing computation with efficiency between CPU and GPU need more than this quite limited PCIe bus - in Pro Servers, those for deep learning and calculation, there is some specific "NVLink" bus between CPU and GPU, 5 to 20 time faster and with Unified memory access to reduce latency drastically). Load /store in memory operations on PC are not very efficient.
So switching to a more recent CPU (with more cores, Intel ones), new PCIe4/5 , new DDR5 (although it is not yet very fast and don't give yet big difference in performances compared to DDR4), all this combined should help for CPU based activities (and there is a lot) and probably on CPU to GPU transfert (Game performance is increasing FPS for each new generation of CPU/GPUso it should be good for us).
Speading IO over several disks is also a good thing too, as DiskDrive don't like numerous small files access and I could see that Browse cache is accessing drive a lot ( small bandwidth but continuous access which can be painfull for older disk technology). NVMe drives are the most efficient in bandwidth and IO access (high number , better queue depth management and low latency) so I would recommand using at least 2 of them (New ATX MB have 3 or 4 NVMe M.2 ports ) , one for OS /apps + image import , the other for scratch/cache disk, and if you have money another one for pictures (otherwise use classical SSD drives or fast 7200rpm drive for capacity). Avoid USB as much as possible, it is the weakest possible link from IO performance point of view.
I will probably perform some testing on a more recent configuration soon ( CPU i12700H + same RTX3070ti +32GB and 1TB NVMe drive in a laptop). Hopefuly, I should see better performances. In that case, I may probably switch to a more recent cpu /MB for my own usage.0 -
Vinny Giannino - Yes, Windiows and ON1 appear to be using the new GPU. I loaded the latest AMXD software and made sure as best I could. Using performance monitoring in task manager, I can see the dedicated GPU is being used. That said, most of the workload and seemingly the main bottleneck is the CPU, in my case the AMD Ryzen 5 5600G.
Perhaps an AMD Ryzen 7 5700X would help, especially as it is PCIe 4 rather than 3, which might speed up some throughput. I don't know if it would benefit me unless I upgraded the motherboard too. I'm not wanting to throw any more money at this issue.
Should you have to have a $2000+ state of the art machine to edit a few photos?
My 1 year old PC was fairly state of the art at the time and very ell speced in terms of memory and SSDs...0 -
Yes we are playing a game without knowing the rules. We can only speculate with a cosly method.
-How much CPu simultaneous thread ON1 need to work correctly ? ( I would say at least 24 because 16 are used for sure)
-What bandwidth between CPU and GPU ? same question for DIsks? How many Cuda (GPU) cores (for sure they are used for AI stuff ; more than 6K is correct so a 3070Ti or more ).
Saying a quad core +16GB and 4GB vram GPU as a general rule is not precise enough and should be documented, especially if ones want adoption from pro users, for whom time is money saved / dedicated to new project. I am pretty sure someone in dev team has those kind of information to build their own testing machine.0 -
christophe ducrocq it sounds like you have way more knowledge than I do about this stuff. I think there is a lot of truth in what you are saying. I went with a Ryzen 3700x due to it benchmarking better in games than a 2700x figuring it would do better with ON1. It has more on board cache which theoretically means it can process things faster. Years ago I asked ON1 support if they were developing the program for multi core CPUs and they said they were. AMD was/is known for better multiform utilization then Intel which is another reason I went with AMD. As far as drives and other things I guess it can make a difference depending on what you want to do. For just storage even a USB drive can be good enough but not for reading and writing all the time. M.2 drives come in 2 types PCIe and SATA with SATA being slower. My storage consists of a M 2 PCIs drive for boot up and programs, a SATA SSD drive for a scratch disk and a spinning HD for data. Any lag in reading data is from the spinning disk in my system. I did have issues as I said when the ON1 system adjustments weren't adjusted correctly but I haven't done any heavy manipulation on photos recently. I have seen my memory usage go up significantly when using layers but at this point I have 32 Gb of memory. One thing I will say that I bought my Windows machine a couple of months before Apple launched the M1 mini and I may have gotten that instead of a Windows computer if it was available.
Andrew Nicholson I agree, my computer is only about 2 YO and it's crazy that I may have issues with software needing more power. So far I haven't had a major issue but I'm keeping my fingers crossed I don't. But with that said I have read that Affinity has issues with AMD GPUs and I own their V2 suite. I need to follow my own advice of be happy with what you have on the computer, test new software to see if it's compatible with my system and if not don't get it. I jumped into 2023 before it was released and luckily it is working for me. I wish I could give you advice on how to fix your issue but I have not seen bottlenecks in my CPU except when it catalogs. After that the CPU/GPU combo seems to working OK.
0 -
Andrew,
There is also a feature for newer processors that is supposed to speed up GPU processing by enabling direct memory access between the CPU and GPU. In AMD processors, it's call SAM - Smart Access Memory. In the Intel environments it's called Re-Size BAR Support. This should be a benefit to everyone with newer hardware, but I haven't heard much about it's effectiveness yet.
Depending on it's age, you might need to upgrade your BIOS. Perform in internet search or check with your motherboard maker for how to enable it.
The other setting that should be enabled on Windows is in the "Graphics Settings". Make sure to add the PR Executable to the list and set it to "High Performance".
0 -
Thanks Carl Traub. I have implemented these based on your suggestion. I had forgotten where/how to find the "Graphics Settings to "High Performance" - as I did this on a previous PC.
A quick test didn't seem to have made a big difference. Time may tell as I get into editing a lot of new images.
This was very helpful advice - what I was ideally looking for from ON1 themselves.0 -
Andrew Nicholson After reading more about the SAM and re-Bar technologies, it sounds like products will need to be modified to take advantage of it. I couldn't tell if it was intended only for games, though it wouldn't seem like it was. It provided about a 10% boost to games in one article that I read , although it greatly reduced the rate of dropped frames. But it might not provide much of a boost to apps that don't do the kind of continuous VRAM writing that games do. It's probably not something that ON1 will focus on.
I don't know why ON1 doesn't add this "High Performance" setting to the installer - it's a simple registry setting.
0
Please sign in to leave a comment.
Comments
28 comments