How much does my GPU matters?

[nitro912gr]

Hello, I'm looking into a possible upgrade for my system and I was thinking to left the GPU out of it becauseI have an amd radeon 7850 1GB and it is still a good enough GPU (entry for gaming, but still strong overall).

 

But since I don't know how and how much does affinity utilize the GPU I'm not sure if this is the right way to go, or should I upgrade my GPU with the rest of the system?

 

 

Does it matter for affinity?

 

(as far as I'm consern 7850 is close to a more modern radeon 460, so I have to get a GPU from 470 and above to make any difference)

 

 

 

[verysame]

From my experience, it doesn't.

I tested it with a low-profile GPU (but good video card Quadro k4000 and Quadro 5000) and with hi-end profile GPU (dual 1080Ti) and I haven't noticed any significant difference.

At some point, I believe I even read a post by someone here about how GPUs are currently not being used by Affinity.

So, my advice would be: if you are happy with the your current video card, keep it. If you are thinking to upgrade to something better, I would go with the 1080ti. Those are beautiful cards and really fast.

 

Just my 2 c.

[nitro912gr]

Thanks, so if using a less than 1 TFlop GPU against a beast with more than 11 TFlops and doesn't show any improvements I don't expect to see any difference if I go from almost 2 Tflops of 7850 to 5 tflops of 470.

 

I guess I will keep the 7850 at my workstation for now.

 

(Yeah 1080ti is a beast but I can't reach this price tag, also since my monitor is 1080p and I mostly play older games anything above rx580 is overkill for me, thanks for the recommendation however)

[verysame]

[...]

 

([].. also since my monitor is 1080p and I mostly play older games anything above rx580 is overkill for me, thanks for the recommendation however)

 

Yup, it makes sense.

I mentioned the 1080ti only because where I work we use Octane and those cards are the best solution for our needs, but for instance, the 980 is just as great and if it comes to games I'm sure there are many other alternatives for less money which can still provide great performances.

 

I think you're right: for now, you can keep the 7850 and, if and when, Affinity will be optimized for GPU then you can always decide to upgrade ;)

[R C-R]

At some point, I believe I even read a post by someone here about how GPUs are currently not being used by Affinity.

They are used, but it is more complex than that. See for example this discussion for some of the details & considerations.

 

So, since the CPU is used for most of the heavy lifting, systems with fast, multicore CPU's will perform best with Affinity, even with a relatively weak GPU.

 

Of course, it is more complicated than that. The cooling system must be adequate to keep the CPU from throttling back during sustained operations. Single core vs multicore performance or the details of the cache & other architectural features might make a substantial difference for some operations but not others. Affinity is more memory efficient than most other graphic apps, so large amounts of RAM may make less difference than one might expect, even for giga-pixel images, but the details of the VM implementation & the storage system interface (like SATA vs. PCI Express or SSD vs. mechanical or hybrid drives) could also be a factor.

[verysame]

From that thread it seems the GPU is used "for drawing to the screen", which sounds very minimal, and confirms my feeling that the GPU doesn't make any difference and explains the same conclusion nitro912gr came up with.

 

By the way, I'm not saying GPU is the way to go (although I love having that option, it sped up my work tremendously), but this doesn't quite sound right either either:

 

 

in order to offer features with significant algorithmic complexity (inpainting, performant blurs, etc.), the GPU is not an option

 

I can get a depth of field, or glare in Octane in real time, in fact, I use it for final renders (hd, 4k, 8k, doesn't matter).

 

And in which way Affinity is more memory efficient than other graphics apps?

[R C-R]

Just for reference, 8K UHD has a resolution of 7680 × 4320 px (33.2 megapixels). That is a lot of pixels, but consider that in photo oriented graphic apps like PS or AP, it is not unusual to work with composites of many layers of high resolution images, often 35 MP  or more for high end DSLRs. PS can handle up to 300,000 x 300,000 px image files. I don't know what the limit is for AP.

 

MattP alluded to this in his comment about "sane" video editing pixel size limits vs. the potentially much higher limits for editing gigapixel photo documents.

 

Memory efficiency determines how often an app must resort to paging to/from VM, using dedicated scratch disks, or whatever when everything won't fit into real memory (along with whatever 'wired' memory system level & other processes need). When everything won't fit in real memory, performance suffers. As mentioned here, PS requires a lot of tweaking to get good performance with high resolution files, in part because it uses complex, multi-level image caching to speed up redraws, which is not very memory efficient.

 

As some of you may know, the Affinity line grew out of a research project done at Serif to see if high performance image processing could be done on memory restricted systems. (They used older iPads for this.) I don't know what it is like for Windows but for Macs, the result is I can load up several memory hungry apps, open multiple large image files in AP, & according to Activity Monitor AP never uses much memory, particularly in comparison to what PSE used before I finally banished it from my system. Even on my iMac with only 8 GB of RAM, all the near realtime previews & very high pan & zoom rates never bog down. In comparison, PSE was a slug, used ridiculous amounts of memory, & was almost unusable when I needed to keep other apps open.

 

Maybe PS is better about this than PSE, but I only had PSE to compare AP with.

[paristo]

Well, something needs to be done for that as how well the AP works in iPad... And a GPU on desktops has very many times more processing power than any tablet on market. 

 

When raw conversion is done via GPU, it becomes almost instant. It is scary how now you open a 16Mpix JPEG file it can take many seconds to open, while any viewer gets it right away open. 

 

But give it a time.... Once the OpenCL is starting to be more widespread it is more useful to everyone.

This is just needed as so much heavy processing is done via CPU <-> GPU where heavy lifting is thrown to GPU to speed processing. 

[verysame]

Just for reference, 8K UHD has a resolution of 7680 × 4320 px (33.2 megapixels). That is a lot of pixels, but consider that in photo oriented graphic apps like PS or AP, it is not unusual to work with composites of many layers of high resolution images, often 35 MP  or more for high end DSLRs. PS can handle up to 300,000 x 300,000 px image files. I don't know what the limit is for AP.

 

MattP alluded to this in his comment about "sane" video editing pixel size limits vs. the potentially much higher limits for editing gigapixel photo documents.

 

Memory efficiency determines how often an app must resort to paging to/from VM, using dedicated scratch disks, or whatever when everything won't fit into real memory (along with whatever 'wired' memory system level & other processes need). When everything won't fit in real memory, performance suffers. As mentioned here, PS requires a lot of tweaking to get good performance with high resolution files, in part because it uses complex, multi-level image caching to speed up redraws, which is not very memory efficient.

 

As some of you may know, the Affinity line grew out of a research project done at Serif to see if high performance image processing could be done on memory restricted systems. (They used older iPads for this.) I don't know what it is like for Windows but for Macs, the result is I can load up several memory hungry apps, open multiple large image files in AP, & according to Activity Monitor AP never uses much memory, particularly in comparison to what PSE used before I finally banished it from my system. Even on my iMac with only 8 GB of RAM, all the near realtime previews & very high pan & zoom rates never bog down. In comparison, PSE was a slug, used ridiculous amounts of memory, & was almost unusable when I needed to keep other apps open.

 

Maybe PS is better about this than PSE, but I only had PSE to compare AP with.

 

35MP is nothing, your math doesn't seem right to me. A 24Mp pixel camera usually delivers approximately 24Mb raw compressed files (or twice uncompressed). If, as you say, many layers are added, 35Mp is way too little.

My daily workflow consists of images that go from a minimum of 500Mb to 1Gb and more.

Also, I never had an issue with PS in that regards. It gets slower when I'm working with many Smart Objects, other that that it works smooth.

 

The Adobe page you linked is just an help in case the end user experiences performance issues. I (and so my colleagues) never had to do any special tweaking to get PS running fine. Actually, if anything, I usually tend to increase the undo steps to at least 100. Other few things are just common sense: scratch disk

 

I'm coming to the conclusion that the fact that Affinity started on a Mac and with the respective restricted hardware and driver options, is not necessarily a good thing, as the drawbacks on the Windows counter part, at least when it comes to deal with big files, is more than obvious at this point. Yes, it definitely gets more stable as the amount of variables to consider are less, but also the scenario where the program has to run is more limited in terms of how much the application can be pushed.

[R C-R]

@verysame,

 

Regarding memory efficiency, it is not just how many MP each image layer needs but also how much memory overhead is required to do anything with it. For PS that includes loaded presets, cached images, & everything else it wants to keep in memory to optimize performance, which is quite a lot. There is no simple math for this, which is why the quoted PS article goes into such great detail about the various factors that can affect it.

 

As far as how much of it is "common sense," I think we will have to agree to disagree about that. I do not know of any other app that goes to such length to try to explain why the app needs its own dedicated scratch disk(s), why defragmenting them frequently is important, why for best practice it is important to put them on the fastest available bus, how driver conflicts can crash the app, or any of the other arcane "under the hood" stuff most users do not know about or want to bother with is required to get the best performance out of the app.

 

If this has never been an issue for you or your colleagues, great, but there is a reason articles like this one exist, why most third party sites recommend at least 16 GB of installed memory as a realistic minimum to run PS, or all the other tips like closing down other apps when using PS can be found all over the web. Everyone I know who uses PS more than occasionally for lightweight processing of small image files does one of two things: either they upgrade to powerful systems with gobs of memory or whine about how slow it is.

 

And to clarify any misconceptions might have, you said, Affinity did not "start on a Mac." This is not correct. As I said, it started as a research project that was run on an iPad, but that was simply because it ran iOS & thus at the time was a device that made relatively small amounts of memory available to apps. That research was applied to the design of the Affinity line, but from the beginning the core code was written to be platform independent, provide the maximum amount of feature parity possible across all platforms, & to support the universal native file format that allows any Affinity document to be opened in any Affinity app, regardless of platform.

 

This is all the kind of stuff that in practical terms can only be done with a "clean slate" approach, using entirely new code & abandoning the restrictions inherent in legacy apps that were first designed long before hardware & OS's had evolved to their present state. To its credit, Adobe has been trying to address some of that, but it still has not been able to implement anything like a universal file format all of its apps can open without conversion or compromise. Adobe's apps still rely heavily on private API's, which bloat the apps, are often far less efficient than what native OS API's provide, sometimes break certain features or cause other stability problems for OS upgrades -- which surprise, surprise, has become a marketing point for the "you-don't-own-much" subscription model that Adobe touts as the best & sometimes only way to keep using its products when users do upgrade their systems.

 

Sorry for the long rant, but I believe "the math" should be added up based on much more than what platform or app anyone prefers or dislikes, nor limited solely to a subset of the user base that may not be representative of all users. I am not saying my math is correct -- I am not sure about that -- only that there is more to it worth considering before jumping to any conclusions about what it tells us.

[verysame]

@verysame,

 

Regarding memory efficiency, it is not just how many MP each image layer needs but also how much memory overhead is required to do anything with it. For PS that includes loaded presets, cached images, & everything else it wants to keep in memory to optimize performance, which is quite a lot. There is no simple math for this, which is why the quoted PS article goes into such great detail about the various factors that can affect it.

 

 

That sounds a little trivial: what would be the purpose of opening an image with lot of layers and do nothing with it? When I say that's my workflow, I mean it. I work with those big images, meaning I do all the usual possible stuff: color correction, color grading, painting, photo manipulation, 3d post-process and so on so forth. So, yes of course is "not just how many MP", but that's pretty obvious.

 

 

As far as how much of it is "common sense," I think we will have to agree to disagree about that. I do not know of any other app that goes to such length to try to explain why the app needs its own dedicated scratch disk(s), why defragmenting them frequently is important, why for best practice it is important to put them on the fastest available bus, how driver conflicts can crash the app, or any of the other arcane "under the hood" stuff most users do not know about or want to bother with is required to get the best performance out of the app.

 

Like I said, never had a problem, never had to deal with those tweaking. Not me, not my colleagues. Not in 20 years. Not in all the companies I worked for. The work I did it's always been heavy on performance and speed, it would be really surprising to find out that my entire experience has been a very lucky and unique situation where PS worked fine for me and for all the people I worked with.

But, then again, it's also worth noting that the heavy work it's most of the time being done on PC/Windows, that has a better performance/stability than Mac when it comes to Adobe (at least, this was true until CS6, with CC the story changed).

 

 

Sorry for the long rant, but I believe "the math" should be added up based on much more than what platform or app anyone prefers or dislikes, nor limited solely to a subset of the user base that may not be representative of all users. I am not saying my math is correct -- I am not sure about that -- only that there is more to it worth considering before jumping to any conclusions about what it tells us.

 

I guess there's some assumption here. What I prefer is not the point, in fact I've been using PC and Mac for my entire career.

I enjoyed both systems for different reasons which I avoid to mention at this point as this is not my thread on why I like PC or Mac.

 

Anyway, the subset of the user base in this circumstance is more than enough for me to gather my conclusions as it is something based on my direct experience and thus is something tangible to me. Since the industry I work for is quite big, when you hear similar opinions from other professionals, and lately the same people who have been Mac users for years, saying that the advantage in terms of speed switching from Mac to Windows is a fact for them, this can only confirm my opinions in that regard. Of course, at the end of the day, they are just my opinions :)

[R C-R]

That sounds a little trivial: what would be the purpose of opening an image with lot of layers and do nothing with it?

I think you missed the point. Earlier you asked how Affinity is more memory efficient than other graphics apps. To answer that, I compared it to PS. Since memory efficiency includes the overhead of everything an app loads into memory to work on images, there is nothing trivial about considering everything PS tries to keep in memory to optimize performance, including presets, image caches, or anything it moves onto scratch disks to make more room in system memory.

 

Like I said, never had a problem, never had to deal with those tweaking. Not me, not my colleagues. Not in 20 years. Not in all the companies I worked for.

 

I am confused. Are you saying you & your colleagues don't configure & use scratch disks with PS, or do not use PC's with many times the minimum requirement of 8 GB of installed RAM to reduce the need for that?

[verysame]

I'm saying we never had to tweak anything.

 

As for the GPU, have a look here:

https://forum.affinity.serif.com/index.php?/topic/42454-gpu-is-not-feasible-they-said/