# 400 concurrent clients **URL:** \ **Category:** Ignition\ **Created:** [February 3, 2023, 7:27am UTC](https://forum.inductiveautomation.com/t/400-concurrent-clients/69390 "2023-02-03T07:27:59Z")\ **Posts on this page:** 7\ **Page:** 1
**Author:** ![grzegorz.kaczmarczyk](https://sea1.discourse-cdn.com/inductiveautomation/user_avatar/forum.inductiveautomation.com/grzegorz.kaczmarczyk/32/24167_2.png) [@grzegorz.kaczmarczyk](https://forum.inductiveautomation.com/u/grzegorz.kaczmarczyk)\ **Post date:** [February 3, 2023, 7:28am UTC](https://forum.inductiveautomation.com/t/400-concurrent-clients/69390/1 "2023-02-03T07:28:00Z")
Hi, one of our clients has a very small project (3 devices and about 1000 variables) but many clients that will connect to the visualization (up to 400 concurrent clients). I've read the server sizing and architecture guide, but what do you think? Is 16 Cores (4 GHz+), 32GB memory, SSD enough or can we take lower options, because of small tag number? ---
**Author:** ![pturmel](https://sea1.discourse-cdn.com/inductiveautomation/user_avatar/forum.inductiveautomation.com/pturmel/32/3295_2.png) [@pturmel](https://forum.inductiveautomation.com/u/pturmel)\ **Post date:** [February 3, 2023, 1:15pm UTC](https://forum.inductiveautomation.com/t/400-concurrent-clients/69390/2 "2023-02-03T13:15:26Z")
You don't say whether this is Vision or Perspective. That server would likely support that many clients in Vision. Based on what I've seen so far, unlikely to be enough for that many Perspective clients. (Like, not even close.) ---
**Author:** ![nminchin](https://sea1.discourse-cdn.com/inductiveautomation/user_avatar/forum.inductiveautomation.com/nminchin/32/6911_2.png) [@nminchin](https://forum.inductiveautomation.com/u/nminchin)\ **Post date:** [February 3, 2023, 2:08pm UTC](https://forum.inductiveautomation.com/t/400-concurrent-clients/69390/3 "2023-02-03T14:08:01Z")
We have a customer with 70 odd perspective clients, 90 vision clients running on two split gateways. 20gb front / 16 gb back, both with 6 cores. So I guess 36gb and 12 cores for a single server. They have 650k tags though.. So it's hard to compare apples ---
**Author:** ![dkhayes117](https://sea1.discourse-cdn.com/inductiveautomation/user_avatar/forum.inductiveautomation.com/dkhayes117/32/68344_2.png) [@dkhayes117](https://forum.inductiveautomation.com/u/dkhayes117)\ **Post date:** [February 3, 2023, 2:34pm UTC](https://forum.inductiveautomation.com/t/400-concurrent-clients/69390/4 "2023-02-03T14:34:40Z")
> [@Perspective versus Vision Load on Gateway](https://forum.inductiveautomation.com/t/perspective-versus-vision-load-on-gateway/26715/2): > > Vision over 100 clients will probably have more of a performance penalty on the gateway than Perspective, due to the less efficient polling architecture. > Perspective sessions, due to their backend component running on the gateway, do have a higher _initial_ penalty on the gateway (ie, 5 Vision clients are likely going to have less impact than 5 Perspective sessions) **but** Perspective sessions scale much, much better - the impact is basically linear. Hmm, I wonder how well this response from 2019 holds up. ---
**Author:** ![paul-griffith](https://sea1.discourse-cdn.com/inductiveautomation/user_avatar/forum.inductiveautomation.com/paul-griffith/32/37249_2.png) [@paul-griffith](https://forum.inductiveautomation.com/u/paul-griffith)\ **Post date:** [February 3, 2023, 4:16pm UTC](https://forum.inductiveautomation.com/t/400-concurrent-clients/69390/5 "2023-02-03T16:16:30Z")
Probably amended somewhat; as long as you take care to take advantage of Perspective's mechanisms to share repetitive query results to different sessions, you can trade memory for CPU to pretty good extent. That said, Perspective _does_ consume more resources than we'd like it to on the gateway, and we're always working to improve that. ---
**Author:** ![grzegorz.kaczmarczyk](https://sea1.discourse-cdn.com/inductiveautomation/user_avatar/forum.inductiveautomation.com/grzegorz.kaczmarczyk/32/24167_2.png) [@grzegorz.kaczmarczyk](https://forum.inductiveautomation.com/u/grzegorz.kaczmarczyk)\ **Post date:** [February 9, 2023, 9:27am UTC](https://forum.inductiveautomation.com/t/400-concurrent-clients/69390/6 "2023-02-09T09:27:41Z")
Thanks for your response. The project will be in Perspective and we are going to use Azure cloud. What about scalable resources? Have anyone tried it? Wouldn't it work for 400 concurrent clients? ---
**Author:** ![pturmel](https://sea1.discourse-cdn.com/inductiveautomation/user_avatar/forum.inductiveautomation.com/pturmel/32/3295_2.png) [@pturmel](https://forum.inductiveautomation.com/u/pturmel)\ **Post date:** [February 9, 2023, 1:12pm UTC](https://forum.inductiveautomation.com/t/400-concurrent-clients/69390/7 "2023-02-09T13:12:38Z")
> [@dkhayes117](#): > > Hmm, I wonder how well this response from 2019 holds up. > [@paul-griffith](#): > > Probably amended somewhat; Meh. I don't agree with that at all. Vision's polling is trivial compared to Perspective UI scripting and binding. Not to mention that Vision is taking delivery of dense binary representations of original java objects while every similiar item for Perspective has to be converted to very non-dense JSON.