Need your help on this next one...

[JN]

Hey guys,

One of the next items up on my review table is a killer of a network card

After thinking about how to benchmark this card *accurately* for several days, I'm still pondering over if there are any better ways, so I thought I'd ask you lot how you'd like to see it done:

1. Install some games, play on some public servers, make note of high/low/avg ping and FPS.

Likely to give the most inaccurate results as you have no control over the number of players on the server you are playing on, what is happening in the game or how stable the connection between you and the random server is.

2. Set up several private game servers on one of the OC3D servers and record my ping and fps on them.

Likely to give much more accurate results as I will be the only person on the server, the server will be hosted on IDNet (1 hop to server) and I will be able to reenact my gameplay without any problems.

3. Set up several private game servers on a PC hooked up to my home network.

Likely to give most accurate results as there is no need to take into account internet traffic etc. However, most ping results are likely to be <1ms which means it will be impossible to compare the cards.

4. Other. Do you have any better ideas?

 

[NickS]

I don't think there are any better alternatives, XMS. Seems like #2 might be the best, but you're right. Local traffic could cause an issue.

Maybe take 3 measurements at 6AM before work, 6PM after work and 10PM for some late nite gaming. Average the 3 measurements at the 3 times, and you'll get 3 relitively accurate numbers. If you want, average those too, to get one averaged ping for "round the clock" gaming.

I'm guessing you'll ping the best @ 6AM, worst @ 6PM and in-between 6AM and 6PM @ 10PM.

 

[Luffy]

How about using Nicks times of day but uploading/downloading a large file. Maybe if this was carried out over 5 days with an average then being taken you would get some decent comparison data?

Just a thought really

 

[JN]

How about using Nicks times of day but uploading/downloading a large file. Maybe if this was carried out over 5 days with an average then being taken you would get some decent comparison data?

Just a thought really

That would be the easy way, but unfortunately the card prioritizes UDP traffic which is used mostly by games etc :s

 

[equk]

If you're reviewing the Killer network card, you could try a few apps on it based on its open source thing (as that seems to be the only gain from it, altho you can get a router for £8 to do the same)

As long as you end by saying it's a waste of money tbh. The tests I've seen so far make it look pants in thru-put etc. Also make sure you test it on a core2duo and not the pentium3 they seem to have referenced it on

Altho most of what I think is in the thread on the specialtech part of the forum.

 

[JN]

As long as you end by saying it's a waste of money tbh. The tests I've seen so far make it look pants in thru-put etc. Also make sure you test it on a core2duo and not the pentium3 they seem to have referenced it on

Yeah it's gonna be tested on a C2D. As for the conclusion..we'll see how things pan out during the testing

 

[equk]

Yeah it's gonna be tested on a C2D. As for the conclusion..we'll see how things pan out during the testing

cool

I was just annoyed at how they advertised it as being a gamers card, claiming it made 10+fps difference .

What system gets 68fps in CS seemed a bit fixed.

Will be interesting to see how the tests come out tho. I was only thinking a few nights ago about whether you were going to review it or not as I remember the specialtech thread stating you would.

One test you could do is put 2 of the cards in identicle systems and connect them via a crossover cable so they are direct. Then do the same with a onboard NIC to compare I'd guess they could argue about throughput of hubs/switches etc

 

[nunzio]

how about setting up smokeping on the machine, we use it at work to get the ping times of our servers to our customers. Might make from some good results if you leave it running for a few hours, and try some gaming/voip/p2p stuff while you are at it. Issue i think is that smokeping runs on unix systems, not windows. Dunno if the killer works with that.. aint really checked it out too much

 

[JN]

One test you could do is put 2 of the cards in identicle systems and connect them via a crossover cable so they are direct. Then do the same with a onboard NIC to compare I'd guess they could argue about throughput of hubs/switches etc

Yeah I was considering this - the thing is tho the ping will probably be too low to measure. I know when I used to host a CSS server on a box at home my ping was always a constant 5ms and never budged even with a big standard onboard nic :\

 

[Archmage]

IIRC, you can use Ethereal, http://www.ethereal.com/ to time, log, graph or whatever.

I use this at work when analysing TCP packet info in industrial control applications, like Allen Bradley EthernetIP or Siemens ProfiNet.

If nothing else, you can compare performance with another known card.

 

[old.gregg]

Can you run two nearly identical PCs sharing the same connection?

You need 2 PCs being tested simultaneously on the same connection to observe how variances effect killer and non-killer.

Check the stats and perception of play on both PCs pre and post killer NIC.

I think the perception is going to be as important as stats - might be good to have several people casually try out the experience on both PCs to see what they feel.

 

[glocktodahead]

w00t I was looking for a review of this thing lol. Thanks

 

[Dav0s]

sorry not read the thread so dont know if it has been mentioned...

ping -t some game servers, or even some web servers and measure the average ping. im sure it cant be that much different to gaming, and it will bemore accurate as there are no factors such as number of players, lag etc

 

[NickS]

ping newegg, lol

 

[ionicle]

try doing somthing, on the internet, and moniter CPU usage

then do the same thing, in the exact same scenario, without the card, and moniter CPU usage, it suposadly free's up CPU cycles, so that is worth a test methinks