does anyone know what CPU is in the DEC3920 / DEC3940 / DEC3960 series? why opnsense product page never state this is annoying. trying to figure out if it how well it would do wireguard and zenarmor, my guess is terrible, otherwise they would publish the specs
Hm. Either the Ryzen 8840U/8845HS or the Epyc 2435. I'd guess the latter, but it's just a guess. Not likely an Intel, as it'd be hard to find a model with 8 homogenous cores.
It has a stat for Wireguard: 4.23Gbps/410kpps, which seems low, but it's not something I've tested. "Threat Protection" is 7.93Gbps/700kpps. IPS software is not specified. Also not something I've tested.
What kind of performance are you looking for?
i don't think its any of those based on the TDPs. *shrug*. zenarmor being singled threaded, i just would like to support opnsense, but this sort of hiding of information is nonsense, if i am going to spent $2000+ on a router, i'd like to know the exact specs.
Quote from: dirtyfreebooter on March 15, 2026, 03:02:51 AMi don't think its any of those based on the TDPs.[...]
They look OK to me. ~30W idle, ~150W max. My own firewall is similar (a 7700X), but with a bit higher power limit.
But yes, if it was up to me I'd publish more data. And send one to ServeTheHome. The folks there really don't have a complete firewall test suite (I'd like to see e.g. session setup rate and time to set up, say, 1500 VPNs), but I don't know of any review site that does.
At some point, someone will buy it and probe it at bsd-hardware info. Give it some time :)
Quote from: dirtyfreebooter on March 15, 2026, 03:02:51 AMi don't think its any of those based on the TDPs. *shrug*. zenarmor being singled threaded, i just would like to support opnsense, but this sort of hiding of information is nonsense, if i am going to spent $2000+ on a router, i'd like to know the exact specs.
The TDPs are confusing and don't fit anything in the AMD lineup, if you assume that they choose a chip without a graphic card.
Since it's 8 cores, DDR5 RAM and probably 10GbE on the SOC you'd think an EPYC series one (but none fit the TDP). With GPU the AMD Ryzen™ AI Embedded P164 would be a possible candidate (TDP 28W, 8C/16T, DDR5 RAM) but total waste of a GPU and don't know if you can get it with 10GbE.
I'm pretty sure it's based on AMD, developing a custom board is hard for a small team and they would need support from the manufacturer and everything else is based on AMD.
Let's have a look at the numbers:
The DEC3800 is based on AMD EPYC 3101/3201 and DEC4200 on AMD EPYC 3251/3451.
DEC3800: Firewall Packets Per Second 1200Kpps 1450Kpps, Firewall Throughput 14.4Gbps 17.4Gpbs
DEC3900: Firewall Packets Per Second 3070Kpps, Firewall Throughput 36.8Gbps
DEC4200: Firewall Packets Per Second 1760Kpps 5000Kpps, Firewall Throughput 21.1Gbps 60Gbps
It's double the performance of the EPYC 3201 (DEC386X) and about 75% more performance then the EPYC 3251.
Send an inquiry to sales@, maybe? 🙂
Quote from: dirtyfreebooter on March 14, 2026, 02:06:18 AMtrying to figure out if it how well it would do wireguard and zenarmor
That information would not actually tell you how well Zenarmor or WG would run. From the general benchmark provided for that specific CPU its not clear. The only real way to know, is to actually implement those features on that device and test.
For overall performance, OPN has a sheet under the product that contains throughput and pps as well IPS throughout for Suricata. And those benches are the base you should consider when buying a HW.
Otherwise contact Sales as mention by Patrick they can provide more details.
Regards,
S.
Quote from: patient0 on March 15, 2026, 11:20:46 AMThe TDPs are confusing and don't fit anything in the AMD lineup[...]
How not? The Epyc 2435 has a TDP of 45/55W, and likely has an actual max consumption of 65-100W. That seems to fit into the ~150W (estimated from 100V @ 1.5A) max for the 39x0, given the lack of specific stats. I don't think it would be more, given the (apparent) design of the Netboard A20 g2; less is easily configured via BIOS.
Quote from: pfry on March 15, 2026, 02:28:25 PMHow not? The Epyc 2435 has a TDP of 45/55W
Because the website listed 28/30/36W as typical power consumption and for the other appliances the value mentioned is around the TDP value of used CPU.
# model, typical power, cpu TDP
DEC600, 13W, AMD GX-420MC TDP 17.5W
DEC700, 15W, AMD V1500B TDP 15W
DEC800, 40/45W, AMD EPYC 3101/3201 35/30W
DEC4200, 55/80W, AMD EPYC 3251/3451 55/80-100W
But your guess is as good as mine.
Quote from: Seimus on March 15, 2026, 02:01:18 PMThat information would not actually tell you how well Zenarmor or WG would run.
i'm sorry, this is so completely wrong. wtf are you even talking about. you can literally correlate passmark scores to both zenarmor and wireguard performance. zenarmor scales linearly with a single CPU and wireguard scales multi-core. i have tested 15+ different cpus over the past few years and you can get a pretty good idea of zenarmor's performance with single core performance scores and wireguard's with multicore performance scores.
sure after you factor in 1000s of connections, etc, that probably changes a bit, or past 10gbe speeds but for home or light duty business, its close enough.
someone is going to buy one eventually and post the info, why not just include it on the product page.
Quote from: dirtyfreebooter on March 15, 2026, 08:38:50 PMi'm sorry, this is so completely wrong.
I am speaking about the real performance. The real performance at the end depends at the final implementation and design of all features you will use on OPN. Thats why for example even using the HW sizing provided by the ZA on their pages is not accurate, as it account performance results only for a lab state test without any overhead.
The moment you add, features, implementations and services this performance metric shifts significantly.
Quote from: dirtyfreebooter on March 15, 2026, 08:38:50 PMsomeone is going to buy one eventually and post the info, why not just include it on the product page.
Than suggest it to them, but this results account only for a favorable test case scenario. The final performance result will depend on the final implementation individual to the user.
Regards,
S.
lol, i emailed sales for the cpu model, they just said AMD V3000. i've asked for the specific model and no reply yet. so its one of these: https://www.amd.com/en/products/embedded/ryzen/ryzen-v3000-series.html#specifications
A Zen 3+ (https://www.techpowerup.com/cpu-specs/ryzen-embedded-v3c48.c3319)? Not what I expected. What does Passmark (https://www.cpubenchmark.net/compare/5613vs5118/AMD-Ryzen-Embedded-V3C18I-vs-AMD-Ryzen-Embedded-V3C48) tell you about them?
there are only 3 of them on passmark, ok scores for sure. but performance per watt isn't the greatest. curious if the actual cpu model is the V3C48 or another...
(https://i.imgur.com/xx5csFj.png)
Sorry, I meant your estimate of Wireguard and ZenArmor performance from their Passmark scores vs. the Deciso marketing materials. How do they look to you? (Another comparison (https://www.cpubenchmark.net/compare/5613vs5118vs3480vs5369/AMD-Ryzen-Embedded-V3C18I-vs-AMD-Ryzen-Embedded-V3C48-vs-AMD-EPYC-3201-vs-AMD-EPYC-3451).)
my latest testbed i had protectli vp2440 with N150, so with zenarmor (also noting this is variable based on how many things are in the policy, etc), with wireguard, testing with iperf3, i could get 2.2 Gbps upload, 1.7 Gbps download with 4-8 parallel streams. watching top, every once in a while you can see one of the CPUs go down to 0% idle, which is zenarmor on one core and wireguard over all 4 cores, hit a state where one cpu is totally consumed.
if zenarmor would ever bring multi-core to home edition, n150, would be a great 2.5g option, with zenarmor and wireguard (for home users where the # of connections is relatively low).
so seems like anything around 10,000+ multicore score could do 4 Gbps wireguard (without zenarmor).
but again, just post the specs and cpu model. i wasn't really asking for ppl's opinions on my setup or their opinions on real world performance, etc. all of the numbers and testing are very install/situation dependent, so just knowing the model and benchmark scores, you can start there and do some napkin math to help you decide if its something you want to try or not. not saying passmark == exact real world performance.
why opnsense/deciso is so secretive about this is ridiculous, frankly.
I would prefer a formal (preferably third-party) review, too. (Although I'm not in the market for hardware - I prefer to assemble my own.) It may simply not be a priority for Deciso, and they may have the right of it - marketing is not one of my skills.
sales confirmed it the V3C18I cpu
Is this a new platform, meaning a desktop variant could be on the horizon?
Quote from: pfry on March 16, 2026, 11:22:07 PMI would prefer a formal (preferably third-party) review, too. (Although I'm not in the market for hardware - I prefer to assemble my own.) It may simply not be a priority for Deciso, and they may have the right of it - marketing is not one of my skills.
That would be my guess, too. Reviewers would focus on price:performance comparisons though I'm guessing Deciso isn't keen to do that. They are not really selling h/w kit but rather integrated firewall solutions; premium ones designed by them and certified for OPNsense. ;)