Bigfoot Networks’ new Killer 2100 is more than just an Ethernet card as it features a network-processing unit designed to reduce lag when on-line gaming. The manufacturer’s claim the benefit is improved performance in games with high network traffic: think MMOs in crowded cities or raids. To try out this card, I tested it in the …
You test it on one game? How about in a 32 man dustbowl server stage 2 1st gate in TF2, thats a good slowdown area. Or perhaps in *any* first person shooter perhaps?
Not read up yet on the card so this one is just a shot in the dark. But how much difference is a network card going to make when the connection to the outside world is slower than most network cards out there?
Yes there may be stack optimisations but ... ahh that's it I will read up on it.
Does the local network card really make that much difference to ping times? Surely most of the lag comes from routers and switches making decisions rather than from your network card?
Also, does this mean ping is not cumulative? Because if your ping time to your router is really in the order of 40ms then I'd be asking why...
(I am no networking or electronics expert so feel free to correct my logic)
Yes it Does! here comes the science...
Whilst there is a cost from any enroute router which cannot be avoided, the greatest cost is at the endpoints, especially where that endpoint is software processed by windows, and it gets even worse when windows is busy playing a game or doing its own thing!
Its not just the ping though, When you are in an area with a large amount of other players there is a far larger amount of data to come down the network pipes. not only that there is more processign to be done in the game itself, when one cpu is trying to process a game and encode/decode network traffic then of course it slows down.
Hardware network cards have been around for a long time but they have always been expensive, around the £80-£100 mark. This bigfoot card is just a hardware network encoder, (and some nice tweak software) that takes all the network processing away from the CPU, by processing the network encodig/decoding on the card, the CPU is free to play the game.. no matter how much data is coming and going through your rig.
Oh btw if you have a decent ISP and a good line request FastPath to be enabled it does make a hell of a difference to ping times.
surely fast paced FPS would be much more useful test than a MMORPG?
does bad company 2 play any better?
Same problem as the gaming mice review
Yes, a lot of people do play MMOs such as World of Warcraft but online gaming takes on a lot of other aspects as well. In games of Starcraft, Heroes of Newerth, Team Fortress 2, amongst others, a delay of 50ms can make much more of a difference than in an MMO where a delay of 200-400ms often does not make a significant impact on raiding or PvP. I should know, I usued to play over a 3G dongle.
In future reviews, can you please try gaming equipment out with other games besides Warcraft? Not trying to be harsh but a variety of different perspectives helps.
"a delay of 50ms can make much more of a difference than in an MMO where a delay of 200-400ms often does not make a significant impact on raiding or PvP"
I live with a very Hard-Core WoW raider. If he ever has a ping higher than 150ms to the Aman'Thul or Silverhand servers (From Australia) there is hell to pay. Even 10ms of additional latency is enough to wipe the raid due to dots, cool-down, etc being cast on the client but the server not registering it in time.
How you were able to get in a guild or pug a raid with a 200-400ms ping is beyond me!
*Beer to settle my nerves as raid time approaches, praying my connection holds up
I admit the guild I was in was very casual and not very advanced at the time I left (beginning of Uldar). My role was mainly as back up in a group of 22+ normal speed users but I wasnt exactly useless. Unless the game has changed a lot since I left (it's possible), 10ms wouldn't make a lot of difference given that there was a global cooldown of 0.5 seconds, making a timing difference of 10ms not overly relevant. It might be that in the truly obsessive ones, where they calculate attack speeds and all that crap down to the nanosecond, it might be important but I don't think the reviewer was talking about that sort of situation from the testing.
I feel dirty just discussing that stuff now. -.-
150ms from Australia?
Please please tell me how you get 150ms pings playing WoW from Australia!
With Wowtunnels or similar services the best anybody in my guild gets is about 200ms. On a good day. And never in ICC! Most of us struggle through the bleedin' Lich King fight with 400+ms. If we don't disconnect outright. We've now been trying for weeks.
I'm happy as a clam when I get a steady 250ms through one of the Wowtunnels servers.
Do you want to buy some magic beans?
I have some for sale....
Recommend you pick up a license for Sisoft Sandra, that way you can give us some informative comparison figures for network latency and throughput.
Good network hardware is worth the extra money. It's the reason why I spend £70 on a 3Com wireless network card instead of buying a cheap £15 wifi dongle. It's the reason why I spend £150 on a Cisco router instead of £30 on a D-Link router.
If the review of this Bigfoot NIC was more informative I'd be willing to consider buying one. I like removing bottlenecks in system performance, if the Bigfoot card can deliver better performance than the NICs integrated directly into nVidia 980 chipsets or those included in the AMD 800 series chipsets then consider me a customer in the making.
Consider firing off an email to the bigfoot guys and ask them. I would like to know whether there's vendor supported linux (and *BSD) drivers or whether they'd be willing to sponsor a dev for them. Those OSes already have a pretty good network stack but they also support TOE and so on. I'm sure shops running such an OS on their routers/firewalls would be interested in gear like this too. And don't forget to share. Here, or as a review submission to el reg, or something.
- How does it compare to NVidia's nForce solution, which also uses a co-processor and does stack optimization as well?
- What about external factors, i.e. shitty ISPs, or when paired with cheapo switches, no-brand routers, etc.?
I can well believe that a card like this makes the difference...
back in the pre-broadband days I had that BT home ISDN product I can't remember what it was called and it's too early to be googling) and I had to buy ISDN cards for my gaming PC's. I had two identical computers and initially bought some no-name ISDN card from (I think) Dabs.com.
I was initially satisfied with the connection to the likes of TFC, it being a vast improvement over POTS dial-up I was using before. As i started to get into clan matches and the like though, and playing on custom 32 player servers I was noticing lag spikes and such more often and a friend suggested an Eicon Diva ISDN card... at the time it was a good £70 which I thought was extortion, my cards having set me back £10 or so including delivery, but I had a job and I was *really* into my TFC so I thought what the hell and bit the bullet.... made a tremendous difference...
Year or two later I experienced the same thing when I went from a built in ethernet connection on the motherboard to a separate NIC in my new gaming PC.
I would definitely consider one of these or a similar product if I was actually using a gaming PC these days (as it is, I now do almost all my gaming via PS3 and XBOX 360, only playing WoW / BC on my late 2008 Macbook Pro - which incidentally cost more than any gaming PC I have ever bought in my life - and has better stability + maccy goodness :p)
I call bollocks!
The Killer NIC software that is running in the background would negate the improvement in offloading network processing onto the NIC.
So, we benefit by offloading the TCP stack, then bog the system down by running yet another program on the computer?
Most plug in NIC's these days feature the ability to offload some of the TCP processing onto the card, this is nothing new, indeed it has been a feature of all server oriented NIC's for many a year, or even decade.
To top it off, your test is based on a "finger in the water" method. You tested it on one day/time, then changed the card for the Killer card, and tried again. The situation was NOT exactly the same, you couldn't control the network congestion and routing once the packets leave your router, so its a purely speculative guess on the improvement at best.
So, lets call it the Killer Networks Placebo NIC shall we? It makes you feel better about being hopelessly outclassed by that guy with the big sword when you are grinding for XP.
Jobs because if he could have sold this as the iKiller, he could have made a bazillion dollars off of peoples gullibility on yet another device....
The network card can't make a difference of more than 1ms
Fibre maybe 12ms
Cable < 20m
ADSL2+ with IP not ATM backhaul and no interleaving < 40ms?
ADSL with ATM backhaul and interleaving high maybe 70ms
3G/HSDPA maybe 120ms to 1000ms
3G/HSPA+, iHSPA, maybe 80ms to 300ms
It can only make a difference on the top 3, which are already so low that it doesn't matter.
Fancy case with venting over the circuit board
Why cover the circuit board with that vented box? Wouldn't it be better to leave the board open?
My guess is that if you were to remove the cover you'd see an embarrassingly empty board with a couple of tiny chips on it. I suppose they need it to look cool in a gaming rig.
I call bullshit
45ms ping from London to the the WoW East coast servers is physically impossible. WoW servers are apparently in Boston which is 5300km from London - in a direct line, not the way the fibre goes.
Speed of light in monomode fibre is normally around :
c/1.5 = 200,000km/sec. That works out to 53ms as a THEORETICAL MINIMUM for the ping time.
The reality is of course that the fibre is probably 1000km longer (20% extra is fairly conservative) than that and there are multiple routers/switches which also add latency. If we assume this then we get a more believable figure of 70ms from London to Boston - even then I'd be willing to bet that's never achieved from a machine outside a London datacentre. It certainly wouldn't be achieved from a DSL connection or cable.
In short, the 45ms is bullshit. As is the traffic prioritisation nonsense - for that to work properly you need END TO END support. What you have there is a simple traffic capper, and there are loads of freeware apps that will do just the same.
Pretty much all the integrated NICs offload checksums etc from the main cpu, which was once the only conceivable reason for these "Killer NICs". Now, they have no purpose at all other than to seperate idiots from their cash.
I remember these cards when they first came out. Total waste of money then, total waste of money now.
Emperor's New Clothes Indeed
Highly skeptical of this product.
First of all, pottering about in WOW doesn't constitute a review. I get the impression that the reviewer is one of those people who's awareness of PC gaming doesn't extend beyond World of Wargrind.
Secondly, Stack optimisations, that's the only interesting part of this product and I bet it can be acheived by other means. The best way to reduce your latancy is and has always been.
1, Get a decent router.
2, Get decent internal phone cabling and ADSL filters.
3, Good internal networking.
Of course you lot probably know this, but the drooling, spotty, freaks that spend all day typing "lol - noob" into CS:S will (of course) lap it up.
It's all marketing, just like telling the same people that putting heatsinks on their memory will give them the "gaming edge".... lol, noobs.
I like the product, as I have liked Bigfoot's previous offerings, however, I have achieved similar gains in WoW with a tweak to the tcp parameters for the nic in my machine - adding registry key TCPAckFrequency with DWORD32 value 1 prevents windows from delayed responses to ACK requests by stopping windows from bundling packet acknowledgement replies to ACK requests from the server in lots of 2 or more.
I have reduced Dalaran latency from around 140-160ms to 40-60 ms (and up to 75ms outside the bank in the evening lol), a significant gain. Windows delays responses to ACK requests because the default settings are designed for office use in a medium or large organisation and are therefore tuned for network efficiency rather than low latency. I would think that all games that run under windows would benefit from this tweak, however I have not tried any other games to test that notion.
What the F*** is wrong with you people?
You have a dedicated HARDWARE card for Graphics processing and dont think twice about it yet your happy with all your network encoding/decoing being sent to your CPU for every single packet????
HARDWARE processing is very well known about, its only since manufactures started putting network sockets on motherboards (and those manufacures dont want the expense of extra chips so it goes to the CPU!) that the yuves forget that it has to be processed at all!
As has been mentioned above there have always been two types of network adapters whether its LAN or ISDN or even Modems, theres the cheap £10 variety and there s the expensive £70-£100 type. guess what the difference is? one goes to the CPU the other doesnt. what do you think the freeby on your motherboard is?
Go drink your own snake oil!!
Nothing - you're the clueless one
The thing is AC that hardware acceleration only makes sense when there's a substantial data throughput. There isn't ANY such requirement for gaming. Now if we were talking about a switch in a datacentre with a 10Gbps connection that is an ideal candidate for full h/w acceleration.
A NIC in a desktop PC sitting on the end of a poxy 10/20/50Mbps DSL/cable line isn't.
Anyway as already stated, most of the current integrated NICs do offload a lot of the processing.
This product is a waste of money. Always has been too.
Killer 2100 works
Hey, Johnny. If you and some of the other naysayers spent half a minute reading about what this product does or better yet, trying it, you'd see that the whole point to is to reduce latency. Online games only use a few Kbps but games are very sensitive to delays or lots of variability in packet timing.This has nothing to do with throughput. I see the same needs on the VoIP systems I work on in my job.The Killer uses the processor on the card to manage packet flow and minimize CPU interrupts. It also has a specialized SW that routes traffic around bottlenecks in the Windows networking stack. Sure, you can ping another PC on your network in less than a millisecond when your PC is idle, but what makes you think it still works that way when it is cranked up and running BFBC2 at high res? Maybe you arent as technical as you think. I got one of these cards a couple weeks ago after reading a solid technical review here: http://www.thinkcomputers.org/bigfoot-networks-killer-2100-gaming-network-card-review/ This card actually delivers. I play WOW and Battlefield Bad Company 2 and I have tested back and forth with my Intel NIC. I consistently get better ping with teh Killer and I can feel the difference when I play. Before posing as an expert, you might want to do a bit more homework.
Network Processes = Core 0
A standard, integrated NIC network is not really handling the processing of network data. Network traffic is run as a kernel level process, on core 0 of your CPU. A dedicated card on the other hand, that has a processor on it, will be able to do the processing and free up resources from the CPU, along with whatever other features it provides. So, unless you are using some crazy integrated NIC that I've never heard of.... That isn't really accurate.
Also, throughput is different than latency. Throughput is a measure of volume, not of speed. Although 'faster downloads' are usually seen as greater throughput (getting 1MB/s down vs getting 100KB/s down), that doesn't mean that packets are traveling any quicker.
Latency on the other hand is the measure of how fast a packet is going from one place to another. That's like comparing a sports car to a bus. The sports car gets to the destination faster, because it can go faster. That's what is important for real time applications where responsiveness is paramount (gaming, VOIP, Video Conferencing, etc...). So it isn't about the quantity of data, in this case.
RE: Network Processes = Core 0
Latency is all about how long it takes to successfully transmit a complete packet without errors, using speed and cars as an anology is massively misleading.
Thanks for your comment, can I both agree and diasagee with you, whilst you are correct that there is minimal processing to be done for the volume of network data, there is a requirement for immediate (lowest latency) processing af All the data, and whilst your PC is idle this is not a problem. BUT if you are gaming then your CPU is not idle, and packets are queued up for processing thereby increasing the latency. The whole point is that a CPU pushing 100% can induce vast amounts of network latency that are not apparant when the system is idle. So whilst your satement about volume is correct it is not the only reason to dedicate the task.
And yes, as you state "A lot of NIC's do offload a lot of the processing" BUT not all the processing, only the high end £60+ dedicated cards do ALL the work, and remeber that most people these days use freeby onboard lan which is cheap as chips and offloads next to nothing.
Buy an INTEL CT Gigabit NIC instead.
Bought mine for £20 and its doubled the network speeds over my gigabit network instead of the gigabit Realtek on the motherboard. Thats with jumbo frames switched on etc.
Pings have also dropped a fraction as well. Works real smooth.
I have noticed on several reviews for these Bigfoot cards several "usual suspects' commenting on how great it is. Some are from the Bigfoot company themselves. They dont cover their tracks that well. Maybe the next Bigfoot card should add in browsing anonymisation.
Except CPUs aren't single cores any more
Does the game you're playing eat all four cores of your high-end CPU in your gaming rig? The ping to your router should be around 1ms on well made ethernet cable. If I put artificial load on my CPU to max out 3 of the 4 cores that climbs to 2ms, on an integrated Broadcom chip. Max out all 4 cores up and that goes to 10ms, although that could be due to factors as well - just because the card has processed the TCP packet instead of offloading that duty to the CPU doesn't always mean the CPU can do anything with it at one point in time.
The thing that people don't realise is that even sub-1% packet loss does far more damage than 50ms of lag. I'd much sooner spend a fraction of that £70 on a few well-made ethernet cables than this funny little trinket.
Nice figures, about what I'd expect. loading up with Prime95 on all cores and set to highest priority and pinging is a good demonstration. 10ms is a BIG difference.
unfortunatly some games as you state are single threaded, so having a faster Core is better than multiple cores. So some gamers took the desision to run a Dual core OC'd higher than a quad (hence my wolfdale at 4.2GHz, I know its getting old!)
Anyway less cores mean less abilty to process, on one hand and on the other, multicore gaming means less ability to process again. always best to use a dedicated processor if it is crucial.
- Boffins attempt to prove the UNIVERSE IS JUST A HOLOGRAM
- China building SUPERSONIC SUBMARINE that travels in a BUBBLE
- Review Raspberry Pi B+: PHWOAR, get a load of those pins
- That 8TB Seagate MONSTER? It's HERE... (You'll have to squint, 'cos there are no specs)
- Review Reg man looks through a Glass, darkly: Google's toy ploy or killer tech specs?