Updating the TQ cluster is overdone I think - what are the next planes from CCPs to upgrade? A Hardware Upgrade would enable bigger fights

EVE Technology and Research Center Player Features & Ideas
1

Looking how much IPC from CPUs has improved it might be smart to support bigger battles and reduce TiDi to upgrade the cluster.

CCP is using CPUs from 2014.

IPC has improved much over the last 7 years.

Looking at AMD Epyc Rome or even Milan. Intel hasn’t caught up but Comet Lake or Cascade Lake are sold quite as Intel isn’t catching up with AMD currently. Still the IPC of those latest Intel CPUs would be a huge uplift.

Additonally going for NVMe Storage for the databases and shifting from SSD tiered SAS 10k rpm disk might help as well.

Looking at the current TQ, you will notice it’s quite old:

Looking at the bechmarks for example for the Epyc 7742 in pyton benchmark: https://www.servethehome.com/amd-epyc-7742-benchmarks-and-review-simply-peerless/2/

As there are no compare benchmark to the CPUs from TQ in pyhton I will compare the 7-zip:

So for the current it would be about 45.000 in decompression speed and about 38.000 in compression speed.

Looking at the Epyc 7742 it would be a bit above 300.000 in decompression speed and about 260.000 in compression…

There we can see how much IPC has improved.

Even comparing it to the currently availbe Intel Xeon SP with much worser IPC. Taking here the Dual Platinum 8280 , which delivers about 250.000 in decompression and about 210.000 in compression.

As far as I am aware IPC matters most for a Node in EVE. As far as I rember talking with devs back in 2018 at fanfest.

So going up the scale by about 6.6 with Epyc Rome or just 4.6 with current Intel, would allow at least much bigger fights. As the CPU speed and the database intructions are limiting TiDi and making much bigger fights possible.

If looking at the last upgrade and the possiblility to even double the numbers in system, just with a little IPC jump from “only” year 2011 CPUs (https://ark.intel.com/content/www/us/en/ark/products/52578/intel-xeon-processor-x5687-12m-cache-3-60-ghz-6-40-gt-s-intel-qpi.html) to year 2014 CPUs it was possible to go from 3k pilots to 6k pilots fights.

Now imaging what going to 2020/2021 CPUs on the servers would mean.

Going from 6k pilot fights to maybe 30 or 40k fights in a single system. Even if it only would tripple or just double the numbers that would be insane. Yes it would be delaying a problem and gaining time for code improvements by CCPs.

Yes I do know that a node is a single threaded application. Which makes IPC even more important to improve it. The IPC gaings are so huge (3 to 6 times - depending on CPU) that it would leave to an uplift of 2 to 4 times, if we compare the uplift with a little IPC gaing from TQ2 to TQ3 with that huge IPC buff TQ would get from upgrading to current hardware. I think its fair to say that battles north of 10k players, even of 20k might not be an issue anymore.

Addionally CCP is running vmware with spectre and meltdown infected CPUs. If they implemented the mitigation for this it would last in about 20% loss for the current and old hardware. Another reason to upgrade.

The only drawback and limiting factor would be the Storage and Database backend, which would have needed to be upgraded to NVMe and much faster stroage.

Anyone a different idea or heard a thing if CCP will upgrade?

Reddit post: https://www.reddit.com/r/Eve/comments/kqake4/updating_the_tq_cluster_is_overdone_i_think_what/

Edit as someone mentioned on Reddit: As someone mentioned below, it might even be enough to just buy some 2U server with the best IPC money could buy right now to have a few super nodes.

That would be the AMD Epyc 7F32…

The cost would be something like 10 to 15k including vmware license…

If for whatever CCP can only go Intel it would be the Intel Xeon Gold 6250 as the Xeon Platium 8256 isn’t in stock at most SIs.

That CPU is about 1k more expensive…

So all in all to create a few brand new super nodes there would be only 1 server needed with one cpu… thats freaking cheap!

2

No amount of upgraded hardware will fix anything until you can run 20k players without tidi in one system. Since this is not possible and/or CCP not rich enough to be able to afford such level of raw tech power, other things are needed to reign in on these issues. Whatever CCP does to increase the server power only leads to us stuffing more people in the system, which leads to the same problems but more people can enjoy this garbage.

The load needs to be spread out. Keepstar battles like YZ9, M2-, M-O and so on ought to be abolished. No one needs them, no one enjoys these hideously bad experiences. Instead, spread the battle to several systems to nodes that you need to capture and keep for a set amount of time. Fights over these needs can break out between smaller fleets, more people need to get involved to create enough fleets to cover these nodes and the geography of the constellation needs to be used to your advantage.
You’d still have the same number of people in the fight overall but since you have spread them to several nodes, they can actually enjoy their time and actually have the opportunity to make a difference instead of just sitting for hours in a jump tunnel or on a titan.

You could still have big capital slug fests on these nodes. However, even if such a slug fest erupts you would still have other nodes that you can capture or defend with the rest of your people, which in turn reduces the number of people in the slug fest and thus makes that smaller slug fest more enjoyable for the involved lunatics, too.

3

Computing power is not a issue. And that dev blog is quite old and outdated. Throwing more hardware at it is not solution. If it would change anything there would be new and more hardware years ago. CCP was never saving on servers. It’s about software or simple things like inability to use multithreading in current simulation software.

You did your math but forgot to do your homework. There is quite few informations about upgrades since that dev blog from 2015. Here on forum.


Is cpu you need to look at. Since best nodes use this ones.
4

While not addressing any of the technical debates on the subject, I’d like to iterate that if in a game where you’re supposed to be immersed into the greatness of space, half the players feel the recurring urge to be in the same spot, something might have gone wrong with the game play design.

5

The bigger the battles, the more players are going to want to be part of it. Better hardware will just be outpaced again. Busting the battles into smaller fleets might help but I have no clue as to how CCP could successfully pull that off.

All I can suggest is keep posting ideas.

6

IMO: bigger hardware will not solve it, force spread out will either not work and/or kill the enjoyment. Wars live from big decisive battles where everybody and their mothers want to be there.

The only solution options I see is a) make things concurrent, b) reduce simulation resolution. Option a) is super hard and many failed, so I propose CCP goes b)

7

A couple of years ago there was a collaboration tech demo with Hadeon:

Does anyone know if there was any movement here, is seems like moving away from the current simulation back-end is key?

8

This would require rebuilding EvE backend from scratch, no way this is happening except in Hilmar’s dreams. At least not with this game and this generation of developers.

9

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.