Do I Need A Super Computer To Edit Photos?

By Wayne Kenward

Topaz has been using Ai for a while now and Adobe is now using Ai including Denoise in Lightroom Classic and Ai Generative Fill in Photoshop. Ai is now mainstream and here to stay.

You may have seen your old Windows and Mac OS systems struggling a little when you run things like Denoise in LRC or run Topaz, or have seen Photoshop Generative Fill takes a long time to provide you with options. If so, the first thing you think about is do you need an upgrade to your existing system or would a new system be a better option?

Buying the wrong system or upgrading the wrong components may not give you what you need. You really need to know what the different components of your computer are doing to work out what is required.

Overview of Computing Components. (Sorry if I’m teaching you to suck eggs)

The key components are:

• CPU
• NPU
• GPU
• RAM
• Storage
• BUS

CPU - Central Processing Unit is the primary "brain" of a computer. It is likely that any system purchased new from 2020 onwards will have a good enough CPU for most photo editing tasks, as long as you didn’t buy a budget system. One task that is currently CPU intensive is the LRC preview building, especially 1:1 previews.

NPU - Neural Processing Unit is a specialized microprocessor designed to accelerate artificial intelligence (AI) and machine learning (ML) tasks. Most systems purchased since 2025 will have an NPU. I haven’t seen Abode or Topaz utilise the NPU yet.

GPU - Graphics Processing Unit is a specialised circuit designed for rapid parallel processing, to accelerate 3D graphics & rendering. It is essential for gaming, AI training, video editing, and, more importantly for us, Topaz and Adobe will really hammer the GPU when using GenFill, denoise, sharpening and rescaling images.

RAM - Random Access Memory is a computer's short-term, volatile, high-speed memory used to temporarily store data, applications, and the operating system. Think of this as your desk, the more you have, the more files you can place on your desk. Every file, image and program you opens takes up RAM, even just starting you MAC or Windows system and not opening any programs or files, you will take up a lot of RAM. If you run out of RAM, the system will place the files back on your storage, which will really slow things down as RAM is much much faster than storage. 32GB should be the minimum RAM for any graphics system, ideally 64GB or more.

Storage – This is the non-volatile storage for your system, it is where your files, images, applications and operating system are stored before they are copied to RAM. The larger the storage, the more files you can save, the faster the storage, the quicker the files can be loaded into RAM. Think of storage as the filing cabinet. If you are a fast runner and you have a good filing cabinet where the draws don’t stick, you can get the files back to your desk faster.

Traditional hard disk drives (HDD) offer a relatively low costs storage option, with some very large disks, over 30TB in size. The trade off is they are relatively slow, a 7200rpm drive will transfer data at around 150-250MB/s. They also draw more power and make more noise. They are ideal for mass storage systems like NAS (see my other blogs).

Solid State Drives (SSDs) are modern, high-performance storage devices that use NAND flash memory to deliver significantly faster data transfer speeds, quicker boot times, and improved system responsiveness compared to traditional HDDs. Without moving parts, SSDs are quieter, more durable, and energy-efficient. They are more expensive and currently the maximum size is 8TB (some enterprise SATA SSDs are up to 32TB). There are two types:

SATA SSDs These use the same interface as HDDs, most as a 2.5” format. Older systems used SATA SSDs. The max theoretical speeds are 600MB/s.

M.2 SSDs These are the smaller SSDs that are used in most modern systems. They use the PCIe interface and provide the fastest transfers speeds. The latest is PCIe5.0x4 which has max theoretical speeds of 15,800MB/s.

You really need to be using a M.2 SSD for your operating system, applications and the Lightroom catalogue.

BUS – This is what connects all the parts of your system together, the CPU, NPU, GPU, RAM, Storage, networking, USB devices, printers, keyboards, mice etc are all connected by a BUS. The faster the BUS, the faster things will communicate. There is no point installing a really high speed M.2 PCIe5.0x4 SSD in a system with a PCIe3.0x4 BUS, even if the M.2 PCIe5.0x4 SSD supports 15,800MB/s, the maximum speed you will get is 3,500MB/s. The same for USB devices, plug a USB-4 external SSD into a USB3.0 port and the maximum speed you will get is 5Gbps. The maximum speed will always be down to the slowest part which will be your bottleneck. Even a cable can be a bottleneck, use a USB 3.0 cable on a USB 4.0 port and a USB 4.0 device and your speed will drop to USB 3.0 speeds, or even worse, you could end up with data corruption.

The Test – Lightroom Classic Denoise

I took a 45MP RAW file from a Canon EOS R5, the file size is 40,643KB.

This file was stored on the local storage in each system, opened in Lightroom Classic and then the LRC Denoise was run.

Take note of the GPU usage, RAM usage and CPU usage, ignore the NPU and networking, this test was not using these resources.

I’ve run tests on several systems from an old workstation from 2018 (with upgrades) to the latest Lenovo ThinkPad and a pro workstation with the latest NVIDIA RTX Pro 4500 Blackwell GPU costing over £2500. Dave Griffiths and Jonnathan Unwin have run the same tests on Apple MACs old and new.

All Windows tests were run on dual monitor systems. The video was recorded using OBS Studio, which does use the GPU, around 5-20% depending on the GPU. This does mean all results would be slightly faster if OBS Studio wasn’t running. As an example, the ThinkPad P1 Gen 8 with the NVIDIA RTX PRO™ 2000 Blackwell Laptop GPU 8GB took 22.02 seconds in the test with OBS running, without OBS running, the same task took 19.27 seconds.

Pay close attention to the Task Manager, in all cases, the RAM and CPU are very stable, but the GPU will hit close to 100% in all tests, proving the GPU is very important for the software we use.

Apple MacBook Pro 14” (Laptop)

Denoise Results 13.36 seconds FASTEST laptop with onboard GPU

• New November 2024
• Apple M4 Max 14-core CPU
• 64 GB unified memory
• 32-core GPU

Lenovo ThinkPad P1 Gen 8 (Laptop)

Denoise Results 22.02 seconds

New November 2025

• Intel® Core™ Ultra 9 285H vPro® Processor(Core™ Ultra 9 285H vPro®)
• Integrated Intel® AI Boost, up to 13 TOPS
• 64 GB LPDDR5X-8533
• NVIDIA RTX PRO™ 2000 Blackwell Laptop GPU 8GB
• 4TB SSD M.2 2280 PCIe 5.0x4 Performance NVMe Opal 2.0

Lenovo ThinkPad P1 Gen 8 (Laptop)

Denoise Results 1 minute & 30.23 seconds

• New November 2025
• Intel® Core™ Ultra 9 285H vPro® Processor (Core™ Ultra 9 285H vPro®)
• Integrated Intel® AI Boost, up to 13 TOPS
• 64 GB LPDDR5X-8533
• Intel® Arc™ 140T GPU
• 4TB SSD M.2 2280 PCIe 5.0x4 Performance NVMe Opal 2.0

The NVIDIA GPU was a minute faster than the Intel GPU on the same high end laptop.

Lenovo ThinkPad P1 Gen 8 (laptop with GPU enclosure)

Denoise Results 7.28 seconds

New November 2025

• Intel® Core™ Ultra 9 285H vPro® Processor(Core™ Ultra 9 285H vPro®)
• Integrated Intel® AI Boost, up to 13 TOPS
• 64 GB LPDDR5X-8533
• NVIDIA GeForce RTX4080 16GB GPU 8GB in a Razer Core X V2 Thunderbolt 5 GPU enclosure connected via Thunderbolt 5
• 4TB SSD M.2 2280 PCIe 5.0x4 Performance NVMe Opal 2.0

This shows how a PCIe4.0 card via Thunderbolt 5 can out-perform the latest PCIe5.0 laptop GPUs.

What is a GPU enclosure?

A GPU enclosure is device that allows laptop users to connect high powered PCIe GPU cards used in workstations to a laptop via Thunderbolt or USB-C.

This is a Razer Core X V2 Thunderbolt 5 enclosure fitted with a high spec NVIDIA GeForce RTX 4080 16GB PCIe4.0 GPU.

Lenovo ThinkStation P5 (Workstation Tower)

Denoise Results 6.15 seconds FASTEST tested

• New December 2023
• Intel® Xeon® W5-2455X Processor(Xeon® W5-2455X
• 256GB DDR5-4800 ECC
• NVIDIA RTX Pro 4500 BlackWell 32GB (released Q4 2025)
• 2TB SSD M.2 2280 PCIe 4.0x4 Performance NVMe Opal

Lenovo ThinkStation P5 (Workstation Tower)

Denoise Results 7.21 seconds

• New December 2023
• Intel® Xeon® W5-2455X Processor(Xeon® W5-2455X
• 256GB DDR5-4800 ECC
• NVIDIA RTX GeForce 4080 16Gb
• 2TB SSD M.2 2280 PCIe 4.0x4 Performance NVMe Opal

The new PCIe5.0 32GB RTX Pro 4500 Blackwell was only just over a second faster than the PCIe4.0 GeForce RTX 4080 16Gb card, and there is about a £1,200 price difference.

Legacy Kit Test

Lenovo ThinkPad X1 Nano Gen 2 (Laptop)

Denoise Results 13 minutes & 44.3 seconds

• New October 2022
• Intel® Core™ i7-1280P vPro® Processor(Core™ i7-1280P vPro®)
• 32GBLPDDR5-6400
• Intel® Iris® Xe Graphics
• 1TB SSD M.2 2242 PCIe 4.0x4 NVMe Opal 2.0

Lenovo ThinkPad X1 Nano Gen 2 (Laptop)

Denoise Results 15.44 seconds

• New October 2022
• Intel® Core™ i7-1280P vPro® Processor(Core™ i7-1280P vPro®)
• 32GBLPDDR5-6400
• NVIDIA GeForce RTX4080 16GB GPU 8GB in a Razer Core X V2 Thunderbolt 5 GPU enclosure connected via Thunderbolt 4
• 1TB SSD M.2 2242 PCIe 4.0x4 NVMe Opal 2.0

Adding a high end GPU via an enclosure and connecting via Thunderbolt 4 took the denoise time down from 13 minutes and 44.3 seconds to 15.44 seconds. An investment is a Thunderbolt GPU enclosure and a GPU is a good option for extending the life of an older or lower powered laptop.

Lenovo ThinkStation P520c (Workstation Tower)

Denoise Results 29.43 seconds

• New April 2018
• Intel® Xeon® W-2125 Processor(Xeon® W-2125)
• 64GB DDR4-2666 ECC
• NVIDIA GeForce RTX4060-O8G
• 2TB SSD M.2 2280 PCIe 4.0x4 NVMe (in a PCIe 3.0x4 slot)

A good test of a 8 year old workstation. Less than 30 seconds with a low end PCIe4.0 GPU with only 8GB, costing around £350. For comparison, I installed a NVIDIA GeForce GT730 2Gb GPU and it took a staggering 14 minutes & 51.38 seconds.

Apple MacBook Pro 16” (Laptop)

Denoise Results 1 minute & 73 seconds

New 2019

• Intel® Core i9 8-Core 2.4Ghz
• 32GB DDR4-7 DDR4
• Intel UHD Graphics 630 1536mb
• SSD PCIe 3.0x4 NVMe soldered

The Test – Topaz Photo Studio

Topaz having been utilising GPU for a long time now, even the basic Denoise, Sharpen or Upscaling tasks hammer the GPU. Recently Topaz have added new Ai based models that will either use the cloud or local GPU to render. In Upscale, the new Recover3 and Standard MAX have been introduced, and unlike the older Ai Models, they really do work the GPU very hard.

I ran two tested with the following systems.

Lenovo ThinkPad P1 Gen 8 (Laptop)

StandardMAX Results 6 minutes & 4.63 seconds

New November 2025

• Intel® Core™ Ultra 9 285H vPro® Processor(Core™ Ultra 9 285H vPro®)
• Integrated Intel® AI Boost, up to 13 TOPS
• 64 GB LPDDR5X-8533
• NVIDIA RTX PRO™ 2000 Blackwell Laptop GPU 8GB
• 4TB SSD M.2 2280 PCIe 5.0x4 Performance NVMe Opal 2.0

Lenovo ThinkPad P1 Gen 8 (Laptop)

StandardMAX Results 10 minutes & 46.13 seconds

New November 2025

• Intel® Core™ Ultra 9 285H vPro® Processor(Core™ Ultra 9 285H vPro®)
• Integrated Intel® AI Boost, up to 13 TOPS
• 64 GB LPDDR5X-8533
• Intel® Arc™ 140T GPU
• 4TB SSD M.2 2280 PCIe 5.0x4 Performance NVMe Opal 2.0

Lenovo ThinkPad P1 Gen 8 (Laptop)

StandardMAX Results 3 minutes & 08.19 seconds

New November 2025

• Intel® Core™ Ultra 9 285H vPro® Processor(Core™ Ultra 9 285H vPro®)
• Integrated Intel® AI Boost, up to 13 TOPS
• 64 GB LPDDR5X-8533
• NVIDIA GeForce RTX4080 16GB GPU 8GB in a Razer Core X V2 Thunderbolt 5 GPU enclosure
• 4TB SSD M.2 2280 PCIe 5.0x4 Performance NVMe Opal 2.0

Lenovo ThinkStation P5 (Workstation Tower)

StandardMAX Results 2 minutes 42.36 seconds

• New December 2023
• Intel® Xeon® W5-2455X Processor(Xeon® W5-2455X
• 256GB DDR5-4800 ECC
• NVIDIA RTX Pro 4500 BlackWell 32GB (released Q4 2025)
• 2TB SSD M.2 2280 PCIe 4.0x4 Performance NVMe Opal

Summary

You need a good GPU for photo editing, the onboard Intel GPUs on the main boards of most Windows PCs and laptops are not powerful enough. Yes they work, but much slower than the NVIDIA GPUs tested. I’d expect Radeon GPUs to perform like NVIDIA, but no Radeon GPUs or AMD CPUs were available to test.

32GB RAM is the minimum you should have. With only Lightroom and Topaz Photo open, most systems were using close to or over 32GB. 16GB does not cut it anymore. If buying new, if you can’t upgrade the RAM, buy a system with at least 64GB for future proofing.

The CPU isn’t really hit much, but there is one Lightroom Classic task on Windows which does hammer the CPU, that is preview building. Even with the GPU selected to do previews

I’ve found the CPU is still heavily utilised and the CPU will often max out at 100% which will cause your system to slow down. This is most prevalent building 1:1 previews on folders containing a lot of images. I’ve heard Adobe does this differently on the MacOS, so hopefully Adobe will improve this in future Windows versions of LRC. Overall, it seems Adobe works more efficiently on MacOS, which may be why Apple specs are often lower than the equivalent priced Windows system.

Task Manager while performing a 1:1 Preview of 10,000 45MP images.

It is clear the GPU is not working hard. Even though LRC is set to use the GPU for building previews, I never saw the GPU go over 40%. The RAM usage was high, 87GB used out of 256GB, but the CPU was constantly over 80% and aften at 100%, this caused the PC to become very sluggish. It’s clear Adobe have a lot of work to do with preview building on Windows.

The NPU was not utilised in any task.

The Future

We are going to see Ai used more and more in all forms of computing, we can’t get away from it. Like Topaz, I think more software companies will offer cloud-based rendering for those with lower spec systems, but this is likely to be subscription based and will be slower than local rendering due to the upload and download time, especially if you have a slow internet. I think future releases of software will take advantages of the modern CPUs, GPUs and NPUs, it seems that the software is lagging the new hardware. If this is the case, older lower spec systems will be obsolete. Is it a good time to buy? The only good time is when you need new technology. 2026 will see a continued increase in NAND (used in SSDs and CFexpress cards) and RAM, in some cases, prices may double due demands with Ai datacentres, prices could drop in 2027, but who knows.