Ryzen 3000 series for FreeNAS?

Would getting a Ryzen 3900 (12C/24T) or 3950 (16C/32T) for FreeNAS be too much? or should I stick with 3700 (8C/16T)? Primarily using the server for video editors (8 seats max). I’m leaning towards the latest Ryzen becoz it can handle 128GB RAM. Any thoughts?

8 seat video feeder i would go with 3900 as a good starting point .but would highly recommend thread ripper over ryzen. cant beat its ecc and memory capacity for the price.

1 Like

Thanks. Is TR 1950X already good or should I go with TR2950X?

Btw, I think Wendell’s coverage of the Ryzen event at E3, he mentioned that we can use upto 128GB RAM on series 3000. Also, I thought we can use ECC ram on Ryzen. Just trying to balance cost vs performance.

Or would it make sense to go the Intel Xeon route?


ecc depends on board vendors. and that 128 gig ram kit is not out in the wild yet. let alone a 128 gig ecc kit with 32 gig sticks. TR2950x is a solid option but a TR1950x on sale is hard to turn down.


Noted. will look into it. thanks man

Would going Intel Xeon E3-1245v6 with either Supermicro X11SSH-CTF or Xeon E5-2609v3 with ASRock EPC612D4U-2T8R be better? I think it falls into the same budget as Ryzen build.

i get why you want FreeNAS and all of that ram
but do you need all those cores?

1 Like

Hence why I’m asking for advice. I read in multiple forums it’s basically one core per station then max out the RAM so that the ARC can handle the transfer 10Gbe to multiple stations. I think 10Gbe maxes out at 6 stations if we zpool of raidz2. Please help clarify if I misunderstood

And i’m trying to help :slight_smile: it is why i’m asking. Idk if i can answer everything but i can help with the questions and others can help answer it.

So so fare you have 6 editors working at once.
You need a 10Gb link
You are aiming for raidz2

I asume we are talking spinning rust? (3.5inch mechanical drives),
You mentioned the Arc, Are you editing large video files?
If so you could benefit from a l2Arc. (separate read cache drive).

How do you intend to connect to these shares? Can i assume from the one core pr station that you are thinking samba shares?
Are your editors on windows?

Yes, aiming to do spinning rust for now. If I have to go to SSD, that will be later.
Yes, editing large video files.
Noted on l2arc but what capacity? 256GB or higher?
Mac and PC clients connecting NFS to FreeNAS. If there is limited permissions, I might give them access through SMB only.

Idk about NFS but the thing with samba, unless something has changed in the last 2 years , (My freeBSD knowledge is limited) Samba is a single core application.
And thus it wont scale with many cores and you wont get 1 user pr core.

However it does not mean you can not get very good speeds, Rather i’m saying you might be happier with fewer cores if that means you get higher clock speeds.
But i think we need someone else to tell us what speed is enough. As i lack the experience.

However i think i need to stop here s my knowledge with scaling a l2Arc is limited, but i can say that generally one looks for high iops rather then size. And you need alot of ram to have a usecase for it, but that is already your intention.

i hope someone else picks it up and i can leave with you this video about iso files, your video files should behave the same.

PS: You might want to turn this into a build log and tag it accordingly, As you are asking for hardware guidance. And with that tag the right people will jump on this.

1 Like

Thanks man.

If i was building a FreeNAS box for video editing storage, i’d be looking at EPYC. 8-12 cores would be fine…


PCIe lanes for SSD storage adapters (via PCIe slots as required) and multiple 10 GbE interfaces.

Ryzen just doesn’t have the lanes IMHO - you might be able to get away with it in the short term, but once you fill those m.2 slots and/or want to go 2x (or more) 10GbE network for redundancy and/or speed purposes you’re screwed.

If it was just basic file/print i’d say Ryzen would be fine, but you specifically mentioned video editing - so don’t fuck about, do it properly, and skip the desktop platform… you’ll be wanting lots of high speed network and SSD, and for the quantity you’d be looking at for video editing Ryzen just doesn’t have the capcacity.


Also consider threadripper, but IMHO you can probably build/buy a low end EPYC for similar cost and although it clocks lower it has a lot more capacity for high speed IO.


Thanks Thro. Unfortunately EPYC CPUs and Motherboards are not as easily accessible in Indonesia. As it is, a Supermicro motherboard that sells for $450 is selling for 3x the price here. I’ll have to go the TR route. Thanks for the input.

Oh yeah, I see above you mention going for rust on the back end (with SSD cache in front of it)…

Would still suggest Threadripper for the lanes because:

  • SSD caching
  • SSD cost falling rapidly, you may be able to go SSD only (or at least SSD as working space entirely, and rust for archive) inside of 3 years (but rust for now)
  • 10 GbE networking. Single 10 gig networking will only get you 1 gigabyte per sec, split 8 ways that’s 125MB/sec or about the streaming speed of a hard drive. So if your 8 users are going to be using it at the same time heavily, you’ll need multiple 10 gig… especially if you’re planning for their workstations to have 10 gig cards in them at some point.


8-12 cores for threadripper will be fine IMHO. I’d go for the baby threadripper, but you DO want to get the X399 platform for the PCIe lanes.


Yep, I think I can fit the rig with 3x Intel 540/550-T2s so that the clients can connect directly to NAS. SSDs will depend on the project load. Film jobs eat a lot of storage so I might still need spinning rusts.

1 Like

TL:DR I agree IO is king for a NAS or almost any workload at the end of the day.

Yeah, though I’m happy that the Zen2 is coming out in less in a month, I was triggered a bit by the 4 lanes between the CPU and the chipset. Certainly it’s better with PCIe 4.0… but not really if you’re hanging 4.0 devices on the PCIe slots provided by the chipset. I wish both Intel and Amd would learn their lesson about creating bottlenecks like this, but it’s par for the course on anything that doesn’t approach the highend desktop or workstation products. Cost cutting and such. And to be fair I know it’s not new to the X570 chipset. Especially since wendell pointed out that they’re just reusing the IO chiplet from the ryzen.

At least they don’t separate chipsets into tiers by limiting the PCIe lanes available by chipset… but then AMD has had only one socket and chipset to rule them all of late so… there’s that.

Has AMD made any mention of a TR refresh? I was a little worried considering they’ve incorporated some TR features into the 3000’s.

AM4 or socket 1151 is built to a consumer pricepoint. I don’t think there’s anything for them to “learn”. They know what they’re doing with it perfectly well.

The typical AM4 or 1151 user is running a GPU and… nothing else in their PCIe slots. Sometimes, not even a GPU.

16 lanes and a few for the chipset is PLENTY for that purpose.

I wish the consumer platform had more lanes too, but for the 99.9% of people who buy into it, even the provided lanes are overkill. If you’re an enthusiast and need the additional lanes, that is what HEDT is for.

Yeah, it sucks that it is expensive, and you or I may not like it, but…

So you’re saying it makes sense to offer 4+ times the PCIe lanes worth of PCIe slots on the chipset vs the link to the CPU because that market segment doesn’t care about having room to grow? Then why bother supporting so many lanes on the chipset and having it a prominent feature listed in the marketing? Oh, so what you’re saying is that they’ve learned to be disingenuous at best? You claim the average consumer doesn’t need them but that certainly doesn’t stop them from selling a paper tiger. They do it because it’s an excuse to sell premium boards at higher margins due to relatively weak demand compared to the past. It has nothing to do with consumer’s needs… tbh has it ever been in the consumer segment? /shrug. I just get frustrated with the games they play even if I understand it.

I’m saying that people are wanting Ryzen to do things that AMD have a specific HEDT product for.

As i said above, the VAST MAJORITY of users of the consumer platform will only ever install 1 card in their system’s life, and it will be a discrete GPU.

The chipset lanes are plenty for the slots for the other peripherals a typical end user of Ryzen may need - maybe a serial IO card, a second network adapter, etc.

You can like it or you can buy something else, this is how the market works - in the market segment socket AM4 plays in, the platform is more than competitive. So is X399.

If you want to harp on about gimped products, intel are who you should be crying about. Even if you buy into HEDT with intel (X299), you’re getting gimped (even worse than X99) on their HEDT platform unless you buy a heap of cores you don’t need and/or probably can’t use in 70-80% of desktop applications. At least with AMD it is clear - you get 64 lanes with Threadripper, whatever spec CPU you put in the board. If it fits the socket, you get all the lanes…

As to an “Excuse to sell high end boards”… the flip-side is that if every consumer ryzen board was wired for 2 or more times the lanes from the CPU, you’d be looking at much more expensive consumer boards. For people that won’t use the extra lanes in 99% of cases.

How is that better for the 99% of people who just want a system to slap a GPU in?