Should I upgrade my personal PC

[Blues]

I have a desktop PC with an AMD Ryzen 5 3600, 16 GB DDR4, & an AMD RX 580 8GB on a B450 board it also currently runs the OS off a 500GB 2.5" SSD that I do intend to upgrade to a 1TB NVMe drive. I am debating upgrading the CPU and RAM I know I can upgrade to any AM4 CPU but thinking the AMD Ryzen 7 5700X is probably the best upgrade point for me but wondering if it is really worth it. The RAM I am less convinced I should but I like the idea of 32GB of RAM and also have though about the GPU but really with a board that only supports PCIe gen 3 I'm on one of the best cards for that generation so the improvements from newer GPUs will be more limited and many newer are PCIe x8 cards which w/o being on Gen 4 or newer would possibly end up as a downgrade making a GPU more cost prohibitive to me.

This is my personal PC used for ripping media and converting into compressed formats for my Plex server on my NAS as well as gaming but most of my games are not that new as I don't mind many to be worthwhile that are newer. My newest tiles are probably Pillars of Eternity 2 and The Witcher 3. So my main question is really if a GPU upgrade from a Ryzen 5 3600 to a Ryzen 7 5700X is worth while.

 

[GTP]

If you’re already thinking about putting money into a CPU upgrade, it might be worth stepping back and asking whether staying on AM4 actually gives you the long-term value you want.

The 5700X is a fine chip, and on AM4 it's the sensible endpoint. But you’re basically polishing off a platform that’s already had its last major release. For workloads like ripping, transcoding, and general compression/encoding, core performance and cache behaviour matter - a lot! You’ll get a bump with the 5700X, sure - but compared to what’s available on AM5 now, the jump is very modest to really nothing.

If you’re open to a more forward-looking upgrade, the 9800X3D is in a completely different league. It carries the same strengths the X3D chips are known for- huge cache, excellent gaming behaviour, and surprisingly strong throughput in mixed workloads. Pairing it with 32GB of DDR5-6000 (which is the current sweet spot for AM5 memory controllers) gives you a setup that won’t bottleneck you for years.

While you’re at it, moving to a 2TB NVMe also makes sense. Once you start doing regular conversions, rip/encode workflows, and general storage shuffling, the space disappears - fast! Gen4 NVMe drives are cheap and fast now, so you’d see a very noticeable improvement over a 2.5" SATA SSD.

A good AM5 motherboard is the last piece of the puzzle. The better boards give you rock-solid VRMs, far more usable PCIe lanes, and decent QoL features - PCIe 5 support, more M.2 slots, better BIOS recovery, faster USB, etc. You don’t need to go crazy, but AM5 definitely rewards buying a feature-rich board rather than the bare minimum.

Bottom line:
If you want the cheapest path to “good enough,” the 5700X on AM4 is that.
If you want an upgrade that actually feels like a modern generational leap - in both gaming and your media workflow - the 9800X3D + DDR5 + a solid AM5 board is the way to go, and it sets you up for far longer.

Both routes work. One is meh... The Other is just… the kind of upgrade you only have to think about once.

 

[Rosco]

if you are upgrading might as well get something with am5

 

[Blues]

Main reason AM5 is not an option is cost/budget for me at least at this time. @GTP the talk about the boost 5700X would give for the encoding is a large reason as for 2TB NVMe my media is only on my drive during the rip and conversion so little to none of it stays on disk so I don't see a need it all gets moved to my NAS quickly and any files being held over I have a 2 TB internal drive dedicated to just general storage so the NVMe will be for applications and "temp" files.

 

[timeshifter]

AMD Ryzen 5 3600 vs Ryzen 7 5700X [cpubenchmark.net] by PassMark Software

Seems like its always a slippery slope. Tempted to do one small upgrade, but why not add more memory or better storage, etc. In the end I usually hold off upgrading until it's a whole new platform and start over.

 

[sapphirescales]

Microcenter has a bundle deal, a Ryzen 5 7600X with a new motherboard for $219. You can't beat that. The only downside is DDR5 memory prices at the moment. I wouldn't bother staying on AM4 with that bundle price. Just be aware that the 7600X is a HOT running CPU. It's DESIGNED to run at 95 degrees! Even a tower cooler can't keep it below that. AMD says it's fine to run it at that temp, but time will tell whether that's true or not. Personally I power limit the CPU to 65w in the BIOS and then it runs at normal temps with a cheap $20 tower cooler.

 

[Metanis]

I have a desktop PC with an AMD Ryzen 5 3600, 16 GB DDR4, & an AMD RX 580 8GB on a B450 board it also currently runs the OS off a 500GB 2.5" SSD that I do intend to upgrade to a 1TB NVMe drive.

That's a very capable system.

Biggest bang for the buck would be the PCI Gen 4 NVMe. It would feel like a whole new machine! Just make sure you buy one with some cache memory! (Yes I'm aware your MB only supports PCE 3, but you'll be able to use the 4 version as a secondary drive on a new PC build.)

Upgrading your RAM would be the next biggest upgrade, but only if you are actually using more than 12GB on a regular basis. And you didn't mention memory speeds, but an upgrade from 2133/2400 to 3200/3600 would feel very nice and help keep that CPU pipeline full. The problem here is that memory upgrade is throwaway because your next system won't use it.

I don't think the CPU upgrade is worth the money long term. There's just not enough improvement since your old one already has the big L1/L2/L3 cache memory sizes. Save it for your next PC.

All that said... this new Unified Memory Architecture tech is very exciting if you ever plan on using locally hosted LLMs. That's what I'm saving my pennies for currently. I'm liking the new AMD AI Max systems which can be had for less than $2500 fully populated with 128GB of UMA RAM. That sounds like a lot but back around 1990 I paid over $2500 for a 12MHz 286 system...

 

[sapphirescales]

I'm liking the new AMD AI Max systems which can be had for less than $2500 fully populated with 128GB of UMA RAM.

Ironically the Mac Studio is a much better bang for your buck. I never thought Apple would actually offer value but here we are. The memory bandwidth on the M4 Max is well worth the extra $1,500 you pay over the AMD chip. Apple actually has a true unified memory architecture and you can use almost all of that 128GB instead of having to manually allocate how much goes to the GPU in the BIOS. If you need more than 1TB the value proposition gets worse thanks to Apple's insane greed when it comes to upgrade pricing but hell, if you can swing the M3 Ultra Mac Studio with 256GB of RAM, it's less than $5,000 if you buy it at Microcenter if you can get an additional 5% off by using your Microcenter card. If I were doing local LLM stuff, that's what I would buy. But personally, I'd rather just pay $20 or $30 a month for a subscription. The pricing on web-based AI stuff is below the actual cost in most cases.

 

[YeOldeStonecat]

Could always look at Task manager and see if you're pushing the limits for RAM, disk I/O, CPU ute, etc.
Instead of wondering, look for factual bottlenecks....

 

[Blues]

current RAM is 2x 8GB 3200 iirc CL16 and I probably average 50-70% on RAM but haven't watched it during video encodings.

NVMe is on the list yeah I know I can use PCIe 4 and up it will be what is best for budget on that one but good point on Gen4 or better drives being good on future systems

Thanks for the input Ill focus on the NVMe if I can find it for good price the RAM and just go from there.

 

[britechguy]

It's DESIGNED to run at 95 degrees!

Which is not really surprising or shocking. I have seen more hand-wringing over "running hot" that, when you check out the manufacturer's specs for what's "within normal limits" is nowhere near to Tmax or over. I also seldom see AMD processors, which do run hotter than Intel, come close to staying at their designed Tmax when even decently cooled under typical stressful operating condition. The temperature you quote is Tmax for the Ryzen 5 7600X, so I figure real-world max will be about 90, with an occasional spike, sometimes very briefly above Tmax.

The i5-1240P in my laptop has a Tmax of 100° C. I think the highest I've ever managed to get it was in the 80s, somewhere, but since I've never been a temperature nanny because I've never had any computer fry itself, and particularly not in the self-throttling and thermal monitoring age, I just don't care.

 

[GTP]

CPU Temperature Reporting on AMD Ryzen (Zen 4 / Zen 5 / X3D) - What’s actually real.

AMD processors expose several different temperature sensors, but only one represents the true silicon temperature of the CPU cores:

1. Tdie (also shown as TSI0_TEMP) is the real CPU temperature​

This is the internal junction temperature measured directly on the silicon.
It’s the value the CPU uses for thermal management, boosting, and throttling.

2. Tctl - control temperature (often shown by monitoring software)​

Tctl is not a separate physical sensor.
It is Tdie plus an offset when AMD decides an offset is required.
AMD adds a +10 deg C offset (and sometimes more up to + 27 deg C depending on model and cooling profile) on certain CPUs to ramp up fans and AIO pumps sooner. This was introduced to improve cooling response, not because the CPU is actually hotter.
X3D chips and some non-X3D parts use these offsets.

3. Software confusion (CPUTIN, SYSTIN, etc.)​

Many programs - especially generic board-level monitors like CPUTIN - read whatever register the motherboard exposes first. These values are not reliable CPU temperatures.
They often report:
Tctl instead of Tdie
A board sensor unrelated to the CPU
Static or “spiky” values due to polling timing
This is why you see “fake spikes” to 95–97 deg C when the CPU isn’t actually sitting there.

My personal Ryzen 7 9800X3D as example​

Zen 5 X3D:
Normal idle/light load Tdie in the mid-50s is realistic
Micro-spikes to the 90s are real, but only last a few milliseconds
These chips boost extremely aggressively. Small spikes to the mid-90s are expected and are still within AMD spec.
What’s fake is when software only shows the instantaneous top edge of the spike without showing the real sustained temperature.

Verify the true temperature on Linux​

Running: watch -n1 sensors will reveal much information about temperatures and voltages.


ChatGPT helped me tidy this up so it made sense!

 

[britechguy]

This is why you see “fake spikes” to 95–97 deg C when the CPU isn’t actually sitting there.

I find the phrase "fake spikes" to be misleading here, because spikes, by definition, are brief and transient. If a temperature sits at any given level then it is, by definition, not a spike. (And I know you didn't write it, but it's a bad phrasing because spike and sitting there are mutually exclusive.)

I've seen plenty of processors spike, very briefly, above even their supposed Tmax, but never stay there. I've seldom seen any sustained temperature that is not at least 5 degrees below that value, and often 10 degrees below that value, under pretty darned heavy load.

That's why I call it "hand-wringing" when someone sees a temperature spike and seems to believe that the end of the processor world is near. It's not, and in the age of self-throttling processors that can and will shut themselves down, entirely, once the danger zone has been entered and seems to be staying there, I don't worry about it. The few times I have seen repeated shutdowns, and rapid ones, is when thermal paste has gone bad, and I can count fewer than 5 incidents since the 1980s. I've definitely seen computers whose cases look like they've been building up their dust bunnies into dust yetis for years never miss a beat.

The days when processors even could "burn themselves out" are long gone. They'll shut themselves down before that can happen.

 

[GTP]

I find the phrase "fake spikes" to be misleading here, because spikes, by definition, are brief and transient. If a temperature sits at any given level then it is, by definition, not a spike. (And I know you didn't write it, but it's a bad phrasing because spike and sitting there are mutually exclusive.)

I've seen plenty of processors spike, very briefly, above even their supposed Tmax, but never stay there. I've seldom seen any sustained temperature that is not at least 5 degrees below that value, and often 10 degrees below that value, under pretty darned heavy load.

That's why I call it "hand-wringing" when someone sees a temperature spike and seems to believe that the end of the processor world is near. It's not, and in the age of self-throttling processors that can and will shut themselves down, entirely, once the danger zone has been entered and seems to be staying there, I don't worry about it. The few times I have seen repeated shutdowns, and rapid ones, is when thermal paste has gone bad, and I can count fewer than 5 incidents since the 1980s. I've definitely seen computers whose cases look like they've been building up their dust bunnies into dust yetis for years never miss a beat.

The days when processors even could "burn themselves out" are long gone. They'll shut themselves down before that can happen.

I wrote most of it - just had the AI make it sound sensible.
I've been dealing with this very issue/nonissue for a few weeks now that we've got our summer.

Yes, "fake spikes" wasn’t the best term - what I meant was that some readings show exaggerated or misleading sharp rises depending on which sensor the software is pulling from.

I have several monitoring Conkys that report temps from all over the PC, including the CPU.
If I use tctl I get relatively low temps being reported. Same with CPUTIN and SYSTIN.

Example: ${execi 10 sensors | grep 'tctl:' | awk '{print$2]'}
or ${execi 10 sensors | grep 'TSI0_TEMP:' | awk '{print$2]'}

I get totally different readings.
I see it all the time on my game PC, where the temp will suddenly peak at 96 deg C the in the same second fall back to ~55 deg C.