Advice for a professional build [Web Dev] [Linux]

[locuro]

Hello, I would like your advice for a professional build: web dev docker under Linux, AI LLM & stable diffusion, a little data science & machine learning, what is needed in terms of efficient CPU and GPU consumption?

Ideally a low energy build, doing dev under Mac Mini M1 and Mac Book Air M1, thank you!

In terms of budget, I don't really have one but if we could stay around €1500, max €2000 (i'm based in France).

I'm looking for a good price/performance/energy ratio

I thought about i5 13400F, i5 13600K, i7 14700K.

And for IA, I think there is no match, I should go for team Nvidia, maybe a 4060?

CPU Software web development with docker, efficient TDP
Graphics Card no gaming, only hardware acceleration & IA, efficient TDP
RAM Software web development with docker, min 32gb to be comfortable
SSD's/HDD's NVME M2 SSD 1Tb minimum
Case Small or medium, I'm not very fan of watercooling but OK if necessary, minimum of RGB
PSU 80 Gold+
Monitor I have an Apple Studio Display (5K, thunderbolt only) and a LG Dual UP 28" Square Double QHD 2560 x 2880
Operating System Linux, modern distro
Peripherals Logitech vertical mice, apple bluetooth keyboard or wired ducky one mechanical keyboard
Cooling efficient, quiet, I'm not very fan of watercooling but OK if necessary, minimum of RGB

 

[Solaris17]

I think people drastically over estimate compute. Running hyper visors or containers actually dont take a lot of compute horsepower. Its all about RAM and the work load.

Web dev takes no workload. Its just a software stack at best. Static content most likely.

If its just a web front end for things like stable diffusion or just LLMs in general, its going to be about how you train it.

SD and other image generation arent going to use any compute at all, they are going to want to run on a GPU. As far as GPUs are concerned just like hypervisors RAM is king.

If you are doing ML and you are using tools that dont train on GPU then you will destroy your CPU.

So build for your workload. RAM and VRAM will be more important than any other type of compute. As far as your other compute goes that is going to come down to how you want to train or render.

 

[oxrufiioxo]

I haven't done it professionally but even my 4090 can run out of Vram in stable diffusion you might want to look at maybe a 4060ti 16GB just to be on the safe side. As far as the rest goes no idea hopefully some others will chime in.

 

[dgianstefani]

PCPartPicker Part List: https://fr.pcpartpicker.com/list/dhBrFs

CPU: Intel Core i7-14700F 2.1 GHz 20-Core Processor (€395.50 @ Amazon France)
CPU Cooler: Thermalright Peerless Assassin 120 Black 66.17 CFM CPU Cooler (€40.90 @ Amazon France)
Motherboard: Asus ROG STRIX B760-I GAMING WIFI Mini ITX LGA1700 Motherboard (€198.90 @ Infomax Paris)
Memory: Corsair Vengeance 64 GB (2 x 32 GB) DDR5-6400 CL32 Memory (€236.99 @ Amazon France)
Storage: Samsung 990 Pro 2 TB M.2-2280 PCIe 4.0 X4 NVME Solid State Drive (€179.99 @ Boulanger)
Video Card: PNY VERTO GeForce RTX 4060 Ti 16 GB Video Card (€491.05 @ Amazon France)
Case: Fractal Design Meshify 2 Nano Mini ITX Tower Case (€112.88 @ Alternate)
Power Supply: be quiet! Dark Power 13 850 W 80+ Titanium Certified Fully Modular ATX Power Supply (€252.60 @ Amazon France)
Total: €1908.81
Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2024-06-12 23:55 CEST+0200

 

[Selaya]

can the 14700(f) even run 6400 w/o downgearing?

 

[locuro]

thanks for your replies
and what do you think about the Intel Arc A770 16Gb, for AI, apparently, it does the job and have enough VRAM ?
I will wait for summer deals I guess

 

[AleXXX666]

Hello, I would like your advice for a professional build: web dev docker under Linux, AI LLM & stable diffusion, a little data science & machine learning, what is needed in terms of efficient CPU and GPU consumption?

Ideally a low energy build, doing dev under Mac Mini M1 and Mac Book Air M1, thank you!

In terms of budget, I don't really have one but if we could stay around €1500, max €2000 (i'm based in France).

I'm looking for a good price/performance/energy ratio

I thought about i5 13400F, i5 13600K, i7 14700K.

And for IA, I think there is no match, I should go for team Nvidia, maybe a 4060?

CPU Software web development with docker, efficient TDP
Graphics Card no gaming, only hardware acceleration & IA, efficient TDP
RAM Software web development with docker, min 32gb to be comfortable
SSD's/HDD's NVME M2 SSD 1Tb minimum
Case Small or medium, I'm not very fan of watercooling but OK if necessary, minimum of RGB
PSU 80 Gold+
Monitor I have an Apple Studio Display (5K, thunderbolt only) and a LG Dual UP 28" Square Double QHD 2560 x 2880
Operating System Linux, modern distro
Peripherals Logitech vertical mice, apple bluetooth keyboard or wired ducky one mechanical keyboard
Cooling efficient, quiet, I'm not very fan of watercooling but OK if necessary, minimum of RGB

my advice 13600K minimum or 13700K/14700K if you want. You can get "non-K" i7 for better price & energy.

 

[locuro]

Memory: Corsair Vengeance 64 GB (2 x 32 GB) DDR5-6400 CL32 Memory (€236.99 @ Amazon France)

Should I really buy some DDR5 ?
When you see the little performance gain in gaming, is it the same for software development and applications ?
Because DDR4 build is so much cheaper

 

[dgianstefani]

Should I really buy some DDR5 ?
When you see the little performance gain in gaming, is it the same for software development and applications ?
Because DDR4 build is so much cheaper

It's not that much cheaper and you squander DIMM density and performance.

Get a 5600 MT kit if you're concerned about cost.

Realistically your best bet is to wait a couple months for Arrow Lake.

 

[locuro]

linux web dev topachat (assembled)

CPU: Intel Core i5-13400F 2.5 GHz 10-Core Processor (€195.00 @ Amazon France)
CPU Cooler: ARCTIC Freezer 34 eSports DUO CPU Cooler (€53.67 @ Amazon France)
Thermal Compound: Thermal Grizzly Aeronaut 3.9 g Thermal Paste (€14.85 @ LDLC)
Motherboard: MSI B760M GAMING PLUS WIFI Micro ATX LGA1700 Motherboard (€165.49 @ Amazon France)
Memory: Corsair Vengeance 64 GB (2 x 32 GB) DDR5-5600 CL40 Memory (€205.99 @ Amazon France)
Storage: Corsair MP600 ELITE w/Heatsink 1 TB M.2-2280 PCIe 4.0 X4 NVME Solid State Drive (€101.99 @ Corsair)
Video Card: Gigabyte WINDFORCE OC GeForce RTX 4060 Ti 16 GB Video Card (€495.94 @ TopAchat)
Case: Asus Prime AP201 MicroATX Mini Tower Case (€82.65 @ Amazon France)
Power Supply: SeaSonic PRIME TX-750 750 W 80+ Titanium Certified Fully Modular ATX Power Supply (€225.31 @ Amazon France)
Case Fan: ARCTIC P12 PST 56.3 CFM 120 mm Fan (€9.99 @ Amazon France)
Case Fan: ARCTIC P12 PST 56.3 CFM 120 mm Fan (€9.99 @ Amazon France)
Case Fan: ARCTIC P12 PST 56.3 CFM 120 mm Fan (€9.99 @ Amazon France)
Case Fan: ARCTIC P12 PST 56.3 CFM 120 mm Fan (€9.99 @ Amazon France)
Case Fan: ARCTIC P12 PST 56.3 CFM 120 mm Fan (€9.99 @ Amazon France)
Case Fan: ARCTIC P12 PST 56.3 CFM 120 mm Fan (€9.99 @ Amazon France)

Total: €1600.83

Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2024-06-14 17:22 CEST+0200

1 675,45€ All taxes included
1 396,21€ Without taxes

I went for a white build, with the duo vents GPU for the 0db mode, and with the 13 400F because I don't do a lot of compilation, and it was cheaper.

 

[dgianstefani]

linux web dev topachat (assembled)

CPU: Intel Core i5-13400F 2.5 GHz 10-Core Processor (€195.00 @ Amazon France)
CPU Cooler: ARCTIC Freezer 34 eSports DUO CPU Cooler (€53.67 @ Amazon France)
Thermal Compound: Thermal Grizzly Aeronaut 3.9 g Thermal Paste (€14.85 @ LDLC)
Motherboard: MSI B760M GAMING PLUS WIFI Micro ATX LGA1700 Motherboard (€165.49 @ Amazon France)
Memory: Corsair Vengeance 64 GB (2 x 32 GB) DDR5-5600 CL40 Memory (€205.99 @ Amazon France)
Storage: Corsair MP600 ELITE w/Heatsink 1 TB M.2-2280 PCIe 4.0 X4 NVME Solid State Drive (€101.99 @ Corsair)
Video Card: Gigabyte WINDFORCE OC GeForce RTX 4060 Ti 16 GB Video Card (€495.94 @ TopAchat)
Case: Asus Prime AP201 MicroATX Mini Tower Case (€82.65 @ Amazon France)
Power Supply: SeaSonic PRIME TX-750 750 W 80+ Titanium Certified Fully Modular ATX Power Supply (€225.31 @ Amazon France)
Case Fan: ARCTIC P12 PST 56.3 CFM 120 mm Fan (€9.99 @ Amazon France)
Case Fan: ARCTIC P12 PST 56.3 CFM 120 mm Fan (€9.99 @ Amazon France)
Case Fan: ARCTIC P12 PST 56.3 CFM 120 mm Fan (€9.99 @ Amazon France)
Case Fan: ARCTIC P12 PST 56.3 CFM 120 mm Fan (€9.99 @ Amazon France)
Case Fan: ARCTIC P12 PST 56.3 CFM 120 mm Fan (€9.99 @ Amazon France)
Case Fan: ARCTIC P12 PST 56.3 CFM 120 mm Fan (€9.99 @ Amazon France)

Total: €1600.83

Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2024-06-14 17:22 CEST+0200

1 675,45€ All taxes included
1 396,21€ Without taxes

I went for a white build, with the duo vents GPU for the 0db mode, and with the 13 400F because I don't do a lot of compilation, and it was cheaper.

View attachment 351325

Oh well. You saved less than 250 euros, and for that, got:

Alder Lake Cores (an architecture from 2021) 6+4 (not Raptor Lake cores with 8+12), so, much less cache, lower clocks, half the cores, and less efficient.
An old model PSU that isn't ATX 3.0/3.1 ready, so will not natively support current and future GPUs.
A 1 TB SSD instead of a 2 TB SSD.
Wasted 15 euros on extra thermal paste for a 65 W CPU, what comes with the cooler is more than good enough, even the Intel stock cooler would have been good enough, and that's potentially 70 euros saved right there.
5600/40 RAM instead of 6400/32 RAM for a 13% savings (30 euros), almost 50% higher latency at 14.3 instead of 10 ns first word.
60 euros worth of cheap fans for a 65 W CPU and a 150 W GPU (completely unnecessary, the fans on the CPU cooler and GPU, plus the preinstalled case fan is more than enough).

I confess, I'm slightly confused why you asked for advice?

This is essentially an obsolete build from the get go.

And you've just spent almost 1700 euros on it.

I guess there's something to be said for a preassembled build if you're not familiar with building a PC, but it really is rather straightforward, and there's very easily accessible DIY guides, which you can follow to get up and running in less than two hours.

 

[locuro]

I asked advice especially for your last comment, because I'm new to PC build, and I would like to learn from experienced user.
The screenshot is just for having a price, I did not ordered it.
Regarding all the thread, it appears that the CPU is not so important, but you made your point regarding the generation and futureproof.
For the PSU, didn't know the ATX 3.0/3.1 point, thanks for the tip.
For the thermal paste, is it not the better to have a cooler CPU performance wise ? But got it, unnecessary.
The same for the case fans, what setup is recommendend ? Number, how much intake, exhaust ?
I tried to go for a RAM model labelled 'optimized for intel / amd expo' and with 5600 MT regarding one of your comment but I'm not aware on how to choose a good ram, I don't understand "almost 50% higher latency at 14.3 instead of 10 ns first word."

I will consider the cost of your build with topachat, but if I would go a little cheaper, what options should I consider ?
I read all your comments, thank you for your implication

As a professional, it is possible to depreciate a computer assembled by a professional in accounting terms, which is impossible to do for the purchase of computer components.
So yes, it's 72,99€ but it's financially interesting and I don't have to do cable management

 

[cvaldes]

All of Apple's M-series based desktop Macs destroy the competition on performance-per-watt metrics. Not sure why you're not considering them.

I bought a refurbished Mac mini M2 Pro at a slight discount directly from Apple last year. Since you are in France, I'm thinking that official Apple refurbs are occasionally available. I'm unclear why you aren't considering Macs.

As for the web development usage, I'm not sure how hardware requirements matter. People have been doing web dev on ultrabooks for well over a decade, regardless of the operating system.

That said, if your goal is to transition toward AI programming, you're probably better off putting most of your budget into a higher-specced 40-series GeForce GPU to have access to NVIDIA's programming ecosystem. Even Jensen thinks that computing systems are about 70% software/30% hardware. It's NVIDIA's vast array of tools that make it the dominant AI marketshare leader (about 98% in datacenters right now), not a bunch of synthetic benchmark results.

At least here in the USA, there are mom-and-pop PC stores that will assemble a computer based on your shopping list. I'm unfamiliar with French tax laws but that would probably qualify as a depreciable business asset purchase here in the USA. I'm guessing that most industrialized countries have some sort of native custom PC manufacturing businesses. I know Japan has tons of them.

 

[dgianstefani]

At least here in the USA, there are mom-and-pop PC stores that will assemble a computer based on your shopping list. I'm unfamiliar with French tax laws but that would probably qualify as a depreciable business asset purchase here in the USA. I'm guessing that most industrialized countries have some sort of native custom PC manufacturing businesses. I know Japan has tons of them.

Yes, literally just have a PC store order and build, they will likely charge ~50 euros for this.

Ensure you actually get every part and the specific parts you ordered though.

That said, if your goal is to transition toward AI programming, you're probably better off putting most of your budget into a higher-specced 40-series GeForce GPU to have access to NVIDIA's programming ecosystem. Even Jensen thinks that computing systems are about 70% software/30% hardware. It's NVIDIA's vast array of tools that make it the dominant AI marketshare leader (about 98% in datacenters right now), not a bunch of synthetic benchmark results.

This is true, but TBH for AI the 4060 Ti 16 GB is enough, the next worthwhile upgrade would be the 4070 Ti S or the 4090, neither of which are smart purchases because the 5090 is releasing this year. Hence, the 16 GB 4060 Ti now, and perhaps a 5080/90 later. For image generation for example, even the 8 GB 4060 Ti is more than twice as fast as a 7900XTX or the Arc A770.

I will consider the cost of your build with topachat, but if I would go a little cheaper, what options should I consider ?
I read all your comments, thank you for your implication

PCPartPicker Part List: https://fr.pcpartpicker.com/list/jT2BVW

CPU: Intel Core i7-14700K 3.4 GHz 20-Core Processor (€430.00 @ Amazon France)
CPU Cooler: Thermalright Peerless Assassin 120 Black 66.17 CFM CPU Cooler (€40.90 @ Amazon France)
Motherboard: Asus ROG STRIX B760-I GAMING WIFI Mini ITX LGA1700 Motherboard (€198.90 @ Infomax Paris)
Memory: Corsair Vengeance 64 GB (2 x 32 GB) DDR5-6400 CL32 Memory (€236.99 @ Amazon France)
Storage: Western Digital Black SN770 2 TB M.2-2280 PCIe 4.0 X4 NVME Solid State Drive (€137.90 @ Infomax Paris)
Video Card: PNY VERTO GeForce RTX 4060 Ti 16 GB Video Card (€491.05 @ Amazon France)
Case: Fractal Design Meshify 2 Nano Mini ITX Tower Case (€112.88 @ Alternate)
Power Supply: be quiet! Straight Power 12 850 W 80+ Platinum Certified Fully Modular ATX Power Supply (€160.66 @ Amazon France)
Total: €1809.28
Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2024-06-14 18:54 CEST+0200

So, the 14700K is actually cheaper than the non K version, since the price went up for the non K, and the integrated graphics are a good backup, plus have Intel Quicksync. The CPU is fine at stock, but if you want to boost efficiency you could limit it to 200 W in the BIOS and lose maybe ~5-10% peak MT performance, no ST performance, but shave 50-100 W off peak power draw. Bear in mind that the vast majority of the time you will be using the CPU at nowhere near it's full load potential, so the likelihood is even at stock it will be running at around 50-100 W most of the time.

This is using a cheaper PSU, I would still recommend the better PSU the Dark Power 13 in the other build, but this is the only way to reduce the cost without compromising performance. You don't need to worry about case fans as the case already comes with two fans, and the rest of the components have their own fans.

If you do want to buy case fans get some from our Editors Choice list.

TLDR, NV is king for AI.

 

[cvaldes]

This is true, but TBH for AI the 4060 Ti 16 GB is enough, the next worthwhile upgrade would be the 4070 Ti S or the 4090, neither of which are smart purchases because the 5090 is releasing this year. Hence, the 16 GB 4060 Ti now, and perhaps a 5080/90 later. For image generation for example, even the 8 GB 4060 Ti is more than twice as fast as a 7900XTX or the Arc A770.

One thing we know is that AI models require substantial memory which is why some speculate iPhone 15 (6GB of RAM) isn't getting new Apple Intelligence features but the iPhone 15 Pro (8GB of RAM) is.

So yes, 12GB or 16GB of VRAM on an NVIDIA graphics card is really what the OP wants, assuming that AI engineering is his future usage case (he is not clear about this). That's a big assumption because he blandly states that he's interested in web development and Linux which is not exactly demanding of high-performance hardware.

Running Linux certainly isn't a performance driven usage case. My Super Nintendo Classic runs Linux. I run Linux on a Raspberry Pi 4. And Linux runs pretty well on every consumer PC apart from Macs.

 

[locuro]

All of Apple's M-series based desktop Macs destroy the competition on performance-per-watt metrics. Not sure why you're not considering them.

I bought a refurbished Mac mini M2 Pro at a slight discount directly from Apple last year. Since you are in France, I'm thinking that official Apple refurbs are occasionally available. I'm unclear why you aren't considering Macs.

As for the web development usage, I'm not sure how hardware requirements matter. People have been doing web dev on ultrabooks for well over a decade, regardless of the operating system.

Apple's M-series are really great indeed but there are 3 main drawbacks:

1. you are stuck with the specs you bought, especially for the ram, can't upgrade the mac mini for example
2. ARM support in dependencies is OK but for legacy projects, like the actual one I'm working on, there are compilations error of binaries, you must emulate some images, so you have less performance
3. docker IO to files is less performant in mac than in linux, so operating a server and running tests is longer

I'm maintaining huge legacy code base, so you must up a few services : redis db, grafana, web server, pg db, kafka instance and run 3000 tests + 150 end to end integration tests with a browser.
So it's possible to dev on ultrabook, but on my MBA m1, the 3K tests are running in 7 min vs 4 min for a linux desktop build from a colleague.

Mac mini could be a great deal, not sure for Mac Studio, and MBP are way too expensive.

I'm working remotely at my office 99% of the time, going some days to my client's office, so the laptop form factor is not relevant performance and thermal wise.

Indeed, my goal is to transition toward AI programming because of the client and market requirements, so I would like to be AI setup compliant, and to run and experiment with some AI locally for side projects, training.

At least here in the USA, there are mom-and-pop PC stores that will assemble a computer based on your shopping list. I'm unfamiliar with French tax laws but that would probably qualify as a depreciable business asset purchase here in the USA. I'm guessing that most industrialized countries have some sort of native custom PC manufacturing businesses. I know Japan has tons of them.

Thanks for the tip, I'm not sure why I forgot this option

So, the 14700K is actually cheaper than the non K version, since the price went up for the non K, and the integrated graphics are a good backup, plus have Intel Quicksync. The CPU is fine at stock, but if you want to boost efficiency you could limit it to 200 W in the BIOS and lose maybe ~5-10% peak MT performance, no ST performance, but shave 50-100 W off peak power draw. Bear in mind that the vast majority of the time you will be using the CPU at nowhere near it's full load potential, so the likelihood is even at stock it will be running at around 50-100 W most of the time.

This is using a cheaper PSU, I would still recommend the better PSU the Dark Power 13 in the other build, but this is the only way to reduce the cost without compromising performance. You don't need to worry about case fans as the case already comes with two fans, and the rest of the components have their own fans.

thanks for your advice and this updated pc part list

One thing we know is that AI models require substantial memory which is why some speculate iPhone 15 (6GB of RAM) isn't getting new Apple Intelligence features but the iPhone 15 Pro (8GB of RAM) is.

Apple just launched Apple Intelligence, so I'm not sure of about the feature support for the desktop hardware.

So yes, 12GB or 16GB of VRAM on an NVIDIA graphics card is really what the OP wants, assuming that AI engineering is his future usage case (he is not clear about this). That's a big assumption because he blandly states that he's interested in web development and Linux which is not exactly demanding of high-performance hardware.

Running Linux certainly isn't a performance driven usage case. My Super Nintendo Classic runs Linux. I run Linux on a Raspberry Pi 4. And Linux runs pretty well on every consumer PC apart from Macs.

It's not just about running linux, software web development is my job, so the performance of running server instances, large tests suite is mandatory, and having not problem of hardware compatibility, ARM is good but not all my clients are operating up to date projects.

For AI, it's not yet mandatory for my day to day work, but it will be soon, I have to train myself

 

[AleXXX666]

One thing we know is that AI models require substantial memory which is why some speculate iPhone 15 (6GB of RAM) isn't getting new Apple Intelligence features but the iPhone 15 Pro (8GB of RAM) is.

So yes, 12GB or 16GB of VRAM on an NVIDIA graphics card is really what the OP wants, assuming that AI engineering is his future usage case (he is not clear about this). That's a big assumption because he blandly states that he's interested in web development and Linux which is not exactly demanding of high-performance hardware.

Running Linux certainly isn't a performance driven usage case. My Super Nintendo Classic runs Linux. I run Linux on a Raspberry Pi 4. And Linux runs pretty well on every consumer PC apart from Macs.

that iPhone part and Apple marketing

 

[locuro]

TLDR, NV is king for AI.

where did you find this stable diffusion relative benchmark ?
I can't find it on the site

 

[dgianstefani]

where did you find this stable diffusion relative benchmark ?
I can't find it on the site

XFX Radeon RX 7900 XTX Magnetic Air Review

The XFX Radeon RX 7900 XTX Magnetic Air comes with user-replaceable fans that pop out without any tools. It's a fantastic solution to clean your card and to simplify RMAs. The card achieved excellent performance results in our review, and the overclocking potential is highly impressive.

www.techpowerup.com

 

[pressing on]

Yes, literally just have a PC store order and build, they will likely charge ~50 euros for this.

I would highly recommend this. The LGA CPU socket that both Intel and AMD now use can be a weak point. A PC store builder can also install the latest BIOS and drivers, and configure the system for you.