morganb4 wrote:
Dont understand why this message is taking so long to get through to a LOT of people. A SSD will not noticeably improve your slider lag in LR.
How so? What role does disk I/O have after the image is loaded in RAM and the processor is doing all the heavy lifting from that point forward?
This thread is also instructive regarding the role of the processor: "...the new controls are very processor intensive."
Again, I'm sitting here with everything on a spinner and I can't detect any sort of lag no matter how I slam the sliders around. Maybe I'm one of the lucky ones. Once I've got my other drives installed and I'm using dual displays I'll report back. Should be able to get it done this coming weekend.
^^^^WHOOOOA
Stop, tell me whats in your box? I need to know now so I can go out and buy it! Please confiorm that with shadow and highlight sliders set, and with a high MP file, that you are getting instantaneous effect from noise sliders - go extreme here bang detail and luminance to 100%, got to 1:2 view in a defocused area and turn the detail down until the artefact/blockiness disappears.
Have car keys and wallet in hand....
(Also I think you may have misread my bit about SSD having litlle effect in dev module)...
morganb4 wrote:
^^^^WHOOOOA
Stop, tell me whats in your box? I need to know now so I can go out and buy it! Please confiorm that with shadow and highlight sliders set, and with a high MP file, that you are getting instantaneous effect from noise sliders - go extreme here bang detail and luminance to 100%, got to 1:2 view in a defocused area and turn the detail down until the artefact/blockiness disappears.
Have car keys and wallet in hand....
(Also I think you may have misread my bit about SSD having litlle effect in dev module)...
You are 100% correct on that last point; I had it exactly backwards.
As for how my new box stacks up performance-wise, you (and I) will have to wait until I'm done moving all my hardware over and I'm operating with a catalog of several thousand images instead of a few hundred. For now, the bare specs are 3770K @ 4.4GHz, Asus Z77 Pro, 250GB Velociraptor, 32GB RAM, GTX 650 2GB.
If you want to know more about how and why performance is killed, think in the terms of "stages".
The raw file is built in stages, I won't go through all the individual hard setpoints, but lets say that if an action is performed in stage 2 (data still in linear, form, first NR applied) then stages 3-6 have to be redone to present an image on your screen.
When you see a full preview on the screen in the "Develop" pane, there are five levels of cache (I think...) involved. That's the image, in five different stages of processing. If you change a control that acts on something that is applied in stage three, the stage three intermediate result cache is used to recalculate the preview, and only stages 4-5 have to be completely recalculated. Change something in stage 4, and the result is instantaneous, even if you're working on 59MP MF images.
Change something in stage 1 though (NR goes here...) and every other step the image has to go through has to be completely recalculated before you have an updated preview. This includes vignette and distortion, almost everything except CA control.
Luckily I don't use NR or any other of the stage 1 controls very often. NR off, and make sure you have updated lens profiles so that you don't have to change CA, vignette and distortion controls when you develop.
^ On my machine. I can alter anything in the BASIC module and not get a hit on the noise sliders. As soon as I touch shadow or highlight or clarity, even by as little as 1 point in either direction, the system turns to crap. .
There are people who do not experience this and people who do.
I dont use clarity very often, I brush may be about 15% of images. Most of my adjustments sit with tonal and colour adjustments and noise. Having a brush edit or two doesn't impact it, even if shadow/highlight has been adjusted in the brush. Its only when I use a h/h highlight adjustment in the basic section - boom
LR4 is close to unusably slow in the three systems I've used it on, all of which had i7 Sandy Bridge or Ivy Bridge processors (some OC'd) with top-tier SSDs. It's getting to the point where I dread opening up LR. And I've done all the performance tweaks documented online and it's still horrible.
Man, I'm counting my blessings. Mine runs like a dream.
One weird thing I found was that typing in numbers for settings slowed things down. This was mainly in the luminance section. After I created a preset to do it, it was no long an issue. Using the slider became a hog. With the preset, it's no longer an issue. I'd suggest that if you find yourself doing repetitive changes, turning them into presets and taking advantage of syncing, should I prove your issue slightly.
Outside of that one little thing I found, auto-masked brushing also slowed things down. I'm very attentive to doing my touch-up on the very last pass now.
Just a note... Windows 8 made a huge impact on speed for me. If you're still debating Win8, you might give it a shot. I'm still not thrilled (two weeks later) with other aspects of the OS (mainly the new browser, mail, etc.), but the system has easily increased speed around 20% for me.
snapsy wrote:
LR4 is close to unusably slow in the three systems I've used it on, all of which had i7 Sandy Bridge or Ivy Bridge processors (some OC'd) with top-tier SSDs. It's getting to the point where I dread opening up LR. And I've done all the performance tweaks documented online and it's still horrible.
Can you please tell me what the IB system was made up of so I dont make the same mistake?
That link shows how all threads of all cores (in a quad core CPU) are working hard during times of sluggish processing. I have the same exact observation. This demonstrates a bottleneck on CPU power and has nothing to do with RAM or SSD/HDD. Peak CPU load of 95-100% per CPU core means that the process will only speed up if you have either more cores or faster CPU with same number of cores. I guess my next computer build will be a 6-core CPU...
As for people having no problems with image rendering speed, it's all relative. I don't doubt your satisfaction with your rig, it's just that our expectations may differ. Even a turtle is fast compared to a snail.
Matthew Cham wrote:
That link shows how all threads of all cores (in a quad core CPU) are working hard during times of sluggish processing. I have the same exact observation. This demonstrates a bottleneck on CPU power and has nothing to do with RAM or SSD/HDD. Peak CPU load of 95-100% per CPU core means that the process will only speed up if you have either more cores or faster CPU with same number of cores. I guess my next computer build will be a 6-core CPU...
That's not necessarily the case with respect to RAM. For big data sets DRAM speed (particularly latency) can have a large effect of effective CPU throughput. A core stalled waiting for a DRAM fetch is calculated as having the same utilization as an unstalled core executing a data set that fits entirely in L1, at least while the workload is running.
snapsy wrote:
For big data sets DRAM speed (particularly latency) can have a large effect of effective CPU throughput. A core stalled waiting for a DRAM fetch is calculated as having the same utilization as an unstalled core executing a data set that fits entirely in L1, at least while the workload is running.
I agree with the CPU utilization, but I have yet to see a single benchmark showing how DRAM latency can noticeably improve Lightroom performance (or any content creation). Even comparing average DRAM to the fastest DRAM, the performance gain is 2% at best. Basically not noticeable IMHO.
Matthew Cham wrote:
I agree with the CPU utilization, but I have yet to see a single benchmark showing how DRAM latency can noticeably improve Lightroom performance (or any content creation). Even comparing average DRAM to the fastest DRAM, the performance gain is 2% at best. Basically not noticeable IMHO.
Agreed, the fault with Lightroom's slow performance lies with Lightroom. If Lightroom can't run acceptably fast with the amazing processing power available today then it's just a dog period.
It would seem prudent to not apply any NR during file import if there is to be a lot of work done in the Develop module. I'll have to change my workflow. I used to have NR applied automatically according to camera and ISO but if I defer that until later then I can go through and weed out the chaff without incurring all of the speed penalties, and later on apply the NR as a preset just to the better images that have survived the first pass culling and second pass editing. Overall this ought to be more efficient.
So long as image noise is not getting in the way of what I am trying to do it should be done after any other stuff.
Just opening an image in the develop module incurs some processing overhead and it is much greater if noise reduction is involved. Some data reading from the raw file may also be required and this can slow the process even more. By leaving out the NR steps for as long as possible the processing time is minimized. By having the raw files and caches on an SSD the data transfer time is reduced. The two benefits can be combined for very quick Lr response as far as showing an accurate image preview and histogram, although additional work goes on behind the scenes.
RAM latency may not have a significant influence but the amount of RAM can in some situations - often not directly but by virtue of allowing other applications such as Ps to have what they need and still leave "enough" for Lr, and also by allowing the operating system to cache data reads from the hard drives. So it may be that the CPUs are able to get their data quicker by not having to wait so long for file reads from HDDs.
NR is one of the very last things I do. I also do one and sync the rest based on ISO and similar image attributes. That helps avoid the pitfalls of the coding issues.
amonline wrote:
NR is one of the very last things I do. I also do one and sync the rest based on ISO and similar image attributes. That helps avoid the pitfalls of the coding issues.
Yeah but it's catch 22 as the noise sliders become unresponsive due to shadoe/highlight it clarity being applied.
