Bigfoot Netoworks' Killer NIC Sneak Peek

Well I'll admit to being intrigued. From a gaming perspective, the best use of the "Ping Spoofing" feature would be to help keep your very fast server from always appearing at the top of the list. If you happen to just be hosting your own, you may not want the attention that comes from an ultra low ping.
 
1) Offloading network traffic from your CPU is nice and all, but who's CPU bottlenecked these days?

2) As several already posted above, multiplayer games use UDP anyway. The TCP protocol requires syns/acks, which increases reliability by guaranteeing that packets will arrive in proper order by resending if necessary (thus "packet loss" doesn't really apply) but increases latency. Offloading UDP isn't really necessary as there's very little processing required to construct each packet and no syn/ack requirement.

3) Network prioritization, also known as packet shaping or quality of service (QOS) is great technology. It lets you mark bittorrent as low priority "bulk" and counterstrike as high priority. So you can leave bittorrent running full blast and play your game at normal latency. It's also included in a bunch of consumer-level "gaming" and VOIP routers and if you're savvy is available in a more sophisticated tweakable form through hacked firmware like ddwrt... which by the way is fantastic. Note that these routers run linux too, and they also have real CPUs and onboard RAM. There are entire linux distributions for these routers with their own package repositories.

4) If they include their own network stack, there will be security holes. Lots of security holes. Kyle, when you get one of these in for testing, do us a favor and run ancient attacks like winnuke, ping of death, etc, against it, see if they work. Frankly, I really think this is BS. There's no way they're writing their own network stack, they're going to use microsoft's standard tcp/ip.

5) Why would you want to run antivirus and bittorrent on an addon card? I'm running utorrent right now seeding 7 files totalling 15GB or so and my CPU utilization is anywhere between 0 and 2%... on a singlecore A64 at 2500Mhz. When I run an AV scan it peaks at 10%, antivirus is I/O (hard drive) bottlenecked, not CPU.

Like everybody, I'm really skeptical about this card. But hey, maybe they'll surprise us all.
 
schizo said:
1)

5) Why would you want to run antivirus and bittorrent on an addon card? I'm running utorrent right now seeding 7 files totalling 15GB or so and my CPU utilization is anywhere between 0 and 2%... on a singlecore A64 at 2500Mhz. When I run an AV scan it peaks at 10%, antivirus is I/O (hard drive) bottlenecked, not CPU.

Like everybody, I'm really skeptical about this card. But hey, maybe they'll surprise us all.
Click to expand...

Big thing in my mind with running like an av off it is you are running it on a different os. That means for something to take out your av protection and infect your machine(as many virus' are made to take out norton and other big names) the virus would need to get past both linux and windows protections. Makes the chances of it happening go down.
 
I'm a software developer and long time [H] reader, and I'm disappointed to see [H] giving these guys air time. Their marketing line and their "whitepaper" ( found here http://www.endlagnow.org/ELN/PDFs/KillerNic_LLR_White_Paper.pdf ) contains a number of statements which at best grossly mis-represent the possible benefits of their product.

They claim their product will lower your ping by bypassing the the windows network stack, and by having better stuff on the card. Now this is plausible, but how much can you gain ?

A simple way to find the absolute upper bound is to play your favorite MP game on a LAN. This shows you the total latency introduced by a round trip through the game client, your network stack, your NIC card, the physical media, and the NIC, network stack, and game software on the server.

In my experience, this total is somewhere from 0 (i.e < 1ms) to 10, with spikes to 20 at the absolute max. Now obviously at most something less than half of this can be attributed to your NIC and network stack, because the server does equivalent steps, and some time is consumed transmitting data over the physical medium. Assuming that Killernic can eliminate all of this (which is obviously impossible) that would be a 5-10ms gain. In practice even that is grossly optimistic, because in game ping calculations are affected by a lot of other stuff. Actual packet latency on a 100mbit LAN is typically < 1ms, and contrary to bigfoots claims in their whitepaper, UDP processing is not much more expensive than ICMP.

Bigfoot tries to imply that processing your game traffic takes substantial CPU power, but this is simply not true. Your average consumer router/firewall has an extremely weak CPU (say a 200mhz MIPS for example), and still doesn't add any appreciable latency. Processing the few megabits per second that home broadband offers just doesn't take much CPU, and games typically utilize far less than the full capacity of even a low end broadband connection. For example, a q3 engine game with the network settings maxed on a busy server sends 100 packets and 8kbytes sec up, and gets ~20 packets and 25kbytes/sec down. A 486 with ISA NICs can handle that level of traffic with no sweat.

I could go into their other claims, but I'll restrict myself to one part of their whitepaper struck me as particularly bogus:

On page 4 there are a number of diagrams claiming to illustrate how they improve FPS by bypassing the overhead of network calls. In bold red letters, it tells you how checking for data in a typical setup takes 10+ instructions, while the KillerNic way is 2. OK, 2 is less then 10+, good so far, but they neglect to mention that a typical game loop is measured in hundreds of thousands, if not millions of instructions, together with many megabytes of data access. There are plenty of similar howlers in the whitepaper, I encourage you to read it.

This isn't to say the product is complete useless (traffic shaping for example is a good thing, and I'm sure there are neato applications for embedded linux on a PCI card), but as far as basic things like ping and game performance, I am quite certain it cannot offer you noticeably better performance than the generic PCI NICs you can pick up on newegg for under $5.
 
good to see that there are some people on here who feel the same way about this NIC.

and a whitepaper which contains "etc, etc, etc" as an example of how COOL and amazing this product is gets my "it smells like cowpat" stamp of approval.

I rest my case ones I have seen 3 independent reviews cause money talks (and promotes).
 
I haven't even see them mention once what architecture the processor is using, or what kernel version. Next time you get a chance at that console please do a uname -a for us would ya?
 
Kyle said:
You do not need to sign your posts, we do that for you. See rule 25. -- Kyle
Click to expand...

Sorry Kyle, I have just always done that, pure habit.

(almost did it again)

I still am curious about the Killer vs Broadcom offloading nics.
 
Well I'm pretty much bored with all the TheoryCraft with regards to these things. Theories about why they can't work, a few people who think they can. I'm looking forward to the actual gameplay reviews from people with no vested interest in the things.

Also, some actual applications for the imbedded linux box. Surely the people at Bigfoot have already written some that we can download (even if unsupported)?
 
so i did a lot of reading and actually read some real whitepapers and it seem that you can indeed improve your latency ... seems like you need a lot of traffic and large packets which are not really ahh used in games. (by a lot of traffic I do mean full gigabit traffic with jumbo frames)

Very interesting read (to much math in the end):
http://nowlab.cse.ohio-state.edu/pu..._tr04_eudp.pdf#search="offloading nic udp/ip"

very interesting too (deal with UDP offloading):
http://www.hpcc.jp/pfldnet2006/slides/s3_03.pdf#search="TOE udp network "

http://download.qlogic.com/datasheet/16291/isp4010.pdf yeah offloading chips which do do UDP checksum offloading. There are loads of TCP only offloading NICs use google search for "TOE"

what i find interesting is that there is a lot of interest to offload udp and that all large network corp are working to fully offload udp and tcp. give or take 4 years time and we will have 100 dollar 10GB NIC's which offload both. Vista network stack is also rewritten to fully work with offloading in mind. Yeah for me.

(and I am convinced that BF2 will see a 10ms to 15ms improvement from a 50ms ping now.... but that has more to do with BF2 then a NIC)
 
JOESKURTU said:
Now if we could just get REAL broadband like europe / asia we would have really low ping times. :eek:
Click to expand...

Europe's broadband isn't that great. Most upload speeds in europe suck, and I am fairly certain Asian countries only have broadband in their urban environments; I doubt that mongels living in the Gobi Desert are sporting a fast connection, or any connection at all for that matter.


Well, back on topic... I think this card has some potential. From what I gathered, there is a 3-10% fps increase. 10% may sound small, but in some cases that extra 10% can create a noticeable difference in gameplay.

I am eager to see if this card lives up to all of the hype and works like how it is designed to.
 
I noticed a few other people mention the SNP for server 2003. The use of TCP offload in servers with the SNP is AMAZING for streaming video at least. Msoft did a demonstration at Winhec where they had two identical servers, both pretty strong 4-ways i believe. They were streaming video off the servers. The clients on the non-SNP server started getting choppy at 400 users, on the SNP server.... over 1200 users with no choppiness.

Offload, in certain applications has a HUGE benefit. I think this will be most useful in cases with a lot of TCP/UDP overhead. I would assume things like bittorrents could allow a lot more connections with an offload engine running. I also think a firewall running isolated from the OS would be nice.

I wonder if this card can do tcp offload, or just UDP? I would think TCP would be harder to offload, since it is more compex than UDP...
 
EODetroit said:
One thing I noticed was that your comment about PingThrottle seems to be in conflict with what's written in the Killer Nic white paper (which I just read).

You said:


But the white paper says:



So unless I'm mistaken, one of these is accurate and one isn't. In this case I hope HardOCP isn't accurate because cheating is a problem in WoW (my game of choice of late). I don't want the cheaters to have more tools with which to get away with it.
Click to expand...
It depends on the implementation and how the game measures latency. If the game's latency measurement uses ICMP ping's rather than add timestamp information with the game data, the network card could be programmed to delay incoming ping requests independently of the game data. If the game's latency measurement is piggybacked with the game data, cheating here would be a little more difficult.
 
JerRatt said:
Some of the examples posted here are great and right on the money. I don't care how much you repackage, slim down, bypass layers, or prioritize but when that network packet leaves your network and hits the first hop it ain't going anywhere any faster than the routers at each hop and bandwidth allow.
Click to expand...

see, that's the issue.
it's not a matter of hardware ping, it's a matter of software latency.
I.E. how long will it take to send a packet given that it has to be scheduled to go out among a bunch of other packets.

what is happening is that you have :
3 packets to send out...
packet 1) some browser crap (will take 10 ms to transfer)
packet 2) a torrent chunk (will take 15 ms to transfer)
packet 3) game data (will take however long to transfer)

SO, given those 3 packets to send, which do you send first?
If you want less latency, send packet 3 first, save 25ms.

it doesn't mean that your network is now 25 ms faster, it just means that other crap that would have taken 25ms to complete before your game packet has a chance to go out, is now put at the back of the line.

know what? you can even choose to send packet 3 COMPLETELY before releasing control for any other transfer (this would be a driver capability).

Same issue is on the inbound traffic.
If you know that game data is coming down on port 65500...
you have 15 open ports, each receiving data.
assume 5 of those ports have received a complete packet...
which of those packets do you want to issue first?
Some random ones, or the ones from port 65500?


essentially, if you prioritize the game packets, your games will have the same latency as if they were the only network app running on the system.

i repeat, it's SOFTWARE latency.
it's what gets done on your system before anything is sent out, and what gets done as stuff comes in.
it's a matter of 'paying attention to the important stuff first'.

comparing plain ping times is unappropriate for this, because ping will just tell you how long it took to get a reply. it won't tell you how long it took before your system bothered to send out the ping request.

-scheherazade
 
scheherazade said:
comparing plain ping times is unappropriate for this, because ping will just tell you how long it took to get a reply. it won't tell you how long it took before your system bothered to send out the ping request.

-scheherazade
Click to expand...
This is not correct.

Using linux 'ping' as an example, since the source is avialable:
Before sending the packet it calls gettimeofday, and records that value in the packet. It sends the packet, and when it gets the reply back, it compares the gettimeofday value stored in the packet with the current time. In other words, it includes all software and hardware latency, including getting from the network driver to a userland program.

Feel free to inspect ftp://ftp.inr.ac.ru/ip-routing/iputils-current.tar.gz to confirm this for yourself.

Bigfoot claims that UDP has a lot more overhead than ICMP, but you can verify this is false for yourself. linux traceroute sends UDP by default. It gets an ICMP TTL exceeded message back. Thus it should give you a good indication of the latency associated with sending UDP. Switching it to ICMP and comparing the result is left as an exercise to the reader :)
 
Heh, we all buy things we don't need so why should this be any different? I'm getting a core duo system this week....do I need it? Not by any means, but it's said that it can increase my performance two fold that until now, I hadn't missed it. In other words, there will be a market despite all the debating of functionality versus price and need.

Innovation is how we move forward so I submit to you--this has to...one way or another, push the boundaries of computer science in a positive direction. Case in point, we've already discussed new and exciting ways to use this card aside its original intent. It won’t be long until other companies market similar hardware thus driving prices down.

Naturally, a company with a concept nobody else is selling will price their product way above anything we would expect. But let's face it, and don't deny it, we are all curious about this new technology (or old just used differently) and what it can potentially do for gamers and non gamers--I know I am.
 
For a nic that offloads the complete network stack and is able to be used for more than one purpose just completely totally rocks! This thing may be able to be used as an operating system in itself, it's a great price just because it offloads anything...especially if it actually works. Go price an alacritech nic will ya.

About the offloading TCP vs UDP...I think you guys don't know for sure what your argument is and your just picking bones. Kyle they said it offloads the entire TCP/IP stack. Sounds pretty generic for anything over IP. For argument sake, I can't see how it would offload just TCP packets and NOT any other packet since ALL IP traffic gets routed through it and after all, Linux does transport all protocols. It would be stupid to just do TCP manipulation while claiming to be able to prioritize game type UDP packets.

Do you think the technology is viable? Well, it's from the guy who designed the Intel 960 so I would give it a little credibility at this point. (The I960 is the chip on uber expensive Intel 10/100 Pro NIC's of yesteryear, among other things)
 
Offloading UDP isn't desirable as the packets are very simply constructed, connectionless, and have no flow control to reduce errors. That is why games use it-- it's much lower latency.
 
Skrying said:
Hmm, a lot more here than I first though. I do not have the money for one, but I must say that I'm a lot more impressed than first thoughts had me thinking.

I actually like the idea of it running my bit torrent client, but really uTorrent only uses next to nothing right now so that really does not matter.
Click to expand...

I agree, for the price I may not get it anytime soon, but when I first heard of it I wrote it off as being shit, and mostly a marketing device, now now that some things have been clarified, this has made it onto my wish list.

I like the idea of being to run my bit-torrent and/or firewall/virus scanner on the nic itself. Sure I understand it might not protect my pc from virus's like a true virus scanner, but scaning incoming packets for virus's, and downloading torrents without affecting peformance, hell yeah! (I realise something like that wouldn't be a full solution as it can't really touch your system and etc, but its an extra layer of sercurity that is nice)

Chilly
 
I read through the article and skimmed through most of the posts in this thread looking for someone to bring up what I am about to bring up. I am in no way an expert at what i'm about to say, but i do have experience in it and would consider myself somewhat knowledgeable.

First of all, the NIC is a great ideal, and it will sell no doubt about that.(i didn't say how much, but some people believe whatever they read)

The problem I see, and did not see questioned or mentioned in the article or in this thread is below.

From reading the article they are allowing you on the PC to prioritize traffic, also known as QoS. Thats all great and dandy, but let me break it down. QoS is based on few set of pre-defined numbers set for different levels or priority of packets. Most switchs/routers use a standard FIFO method, first in first out. Which in Bigfoot networks desgin and usage of the "QoS" feature to get your "pings" out first will *work*. (due to the fact that the first thing out of your pc and into the ISP's switch/router would be your pings)

Now going one step deeper. If the ISP has QoS implemented into they're switch/router things change drastically. They are no longer using FIFO, per se.(FIFO actually happens at the port level. QoS happens before FIFO. So technically they are still using FIFO, but the QoS decides in which order they get to the port) They are prioritizing incoming traffic and giving it a pre-defined set of numbers as I mentioned above. Generally Voice is given a CoS value of 5 or DSCP value of 46, making it a "High priority packet" (or "express forwarding" depending on who you talk too). Voice generally gets its own queue which is considered the "priority queue" and sent from the switch/router first (highest priority). All other traffic, depending on how it is marked is separated into 3 other queues and using WRR (weighted round robin) is sent in order of importance. Now, Internet traffic (most any internet traffic, pings included) are considered to be best effort traffic. In this case they are given a CoS value of 0 or 1 and a DSCP value of 0 or 8. Which by most QoS standards is the lowest level of traffic and the least important. (meaning begin sent AFTER voice or other high priority packets)

What does all this mean?

If your ISP has QoS turned on, your packets, providing they are begin marked with a 0 or 1 will NOT be the first thing sent from your ISP's router/swtich. Meaning the Bigfoot networks Killer NIC's QoS is fairly null. Its like ordering a product from newegg and using the expedited delivery and getting it in the same time as normal delivery option. You are at the total mercy of UPS/Fedex.

With QoS they are tons of variables, like the ISP only trusting certain devices to mark the packets with a CoS or DSCP value. (Say they're edge device, in many cases the CMTS. (where the cable modem terminates at the ISP's office)) The only real way to know what your ISP is up to is by reading the SLA (service level agreement).

Hopefully i didn't' confuse anyone lol.

BTW, UDP is a faster protocol than TCP, however it has NO retransmits. Which in gaming is fine.


EDIT:After re-reading my post it is still techinical. In lemans terms, the Bigfoot NIC may get the pings out of your home network faster, but you are at the mercy of your ISP. Meaning, it *could* improve ping times. I'd put a strong emphasis on could.

Then again i'm sure they know more about this than I do, maybe i'm just babbling about why I don't think it will work quite as well as advertised.
 
//Disclaimer, If Bigfoot Networks takes my idea I would like a free card ;)

Jeeze, from the perspective of a CTO I can tell you that they are targeting this technology at the wrong market segment!

This card should be stripped of it's wierd-ass heat sink and be sold for the same price to businesses/corporations. They should also have a contest for applications to run on the system to help early adoption. Later on they could release a half-powered model to the gamers once the technology has matured a bit.

Linux enthusiasts and anyone in VOIP/etc would love one of these to off load processing pressure from their main system or possibly use it to help mitigate the effect of DDOS attacks that normally saturate the processors on normal machines. The possiblities are near endless, however they won't get the response they want from the gamer segment, there are hardly any coders in this crowd.
 
Traffic shaping actually makes a lot of sense for home broadband users, especially those with more than one computer. Unlike NICs, your DSL/Cable modem actually IS a bottleneck, and most consumer broadband equipment uses huge FIFOs with no prioritizeation at all. This means that as soon as you use your full bandwidth, interactivity goes in the toilet. I run a fairly complex tc setup on a DSL line I share with a number of heavy internet users, and it works great.

Note however that in this situation, it make far more sense to do it on your router, since you want to shape all the traffic going over your broadband connection, not traffic between the individual systems on your LAN. For most home users, a LAN is so fast that there is no reason to shape internal traffic. I know from experience that having a LAN user leech files from me doesn't have any noticeable effect on ping or FPS.

If you only have one computer, and you like to run torrents and game at the same time, than having shaping on your PC would make some sense. However, you don't really need special hardware to do this, and if you do want to buy hardware, a router capable of doing taffic shaping is probably cheaper than killernic.
 
Malk-a-mite said:
That would be the same mrp trashing everyone that doesn't think the product is better than sliced bread and has only ever made 3 posts and all of them in this thread?
http://hardforum.com/search.php?searchid=3906877
Actually I trashed posts that
A) where based on price
B) could have been answered if they had done a bit of reading
Not based on someone not thinking the product may not amount to much.

What does the amount of posts have to do with anything? I can easily point to someones Gawd status based on one liners.
Click to expand...
 
GodSpeed said:
Jeeze, from the perspective of a CTO I can tell you that they are targeting this technology at the wrong market segment!

This card should be stripped of it's wierd-ass heat sink and be sold for the same price to businesses/corporations. They should also have a contest for applications to run on the system to help early adoption. Later on they could release a half-powered model to the gamers once the technology has matured a bit.
Click to expand...

They SHOULD, but instead they choose to label their card with the "gaming" moniker and jump on the "gaming" hardware bandwagon solely for promotion purposes. It's been done before by other hardware companies that provide little to no benefits for gaming in their hardware despite labeling as such. Even large PC OEMs do it... You see the Dell ads lately that say the $399 special is GREAT FOR GAMING?!

The only reason this pisses me off so much isn't because they are doing the same thing as other companies (using gaming as an advertisement tool only), it's because they are doing it with a product and charging thousands of times the price of a comparable card. Now I can blow off the "gaming" mouse that is an exact copy of the business mouse line simply because they charge an extra 10% or so, not hundreds or thousands of percentages higher. Like I said before, this is a slap in the face to the gaming community.

Simple, price and performance aside, market your hardware to the segment your hardware would be best used in and NOT in the market where you can take advantage of wealthy gamers who have nothing to go off of but PR and advertising. It's disgraceful.
 
scheherazade said:
see, that's the issue.
it's not a matter of hardware ping, it's a matter of software latency.
I.E. how long will it take to send a packet given that it has to be scheduled to go out among a bunch of other packets.

what is happening is that you have :
3 packets to send out...
packet 1) some browser crap (will take 10 ms to transfer)
packet 2) a torrent chunk (will take 15 ms to transfer)
packet 3) game data (will take however long to transfer)

SO, given those 3 packets to send, which do you send first?
If you want less latency, send packet 3 first, save 25ms.

it doesn't mean that your network is now 25 ms faster, it just means that other crap that would have taken 25ms to complete before your game packet has a chance to go out, is now put at the back of the line.

know what? you can even choose to send packet 3 COMPLETELY before releasing control for any other transfer (this would be a driver capability).

Same issue is on the inbound traffic.
If you know that game data is coming down on port 65500...
you have 15 open ports, each receiving data.
assume 5 of those ports have received a complete packet...
which of those packets do you want to issue first?
Some random ones, or the ones from port 65500?


essentially, if you prioritize the game packets, your games will have the same latency as if they were the only network app running on the system.

i repeat, it's SOFTWARE latency.
it's what gets done on your system before anything is sent out, and what gets done as stuff comes in.
it's a matter of 'paying attention to the important stuff first'.

comparing plain ping times is unappropriate for this, because ping will just tell you how long it took to get a reply. it won't tell you how long it took before your system bothered to send out the ping request.

-scheherazade
Click to expand...

Fair enough but, and this is a big but, do I need a $280 card to do this or just a new network stack? This just sounds to me like what it should be doing than some offload card.

Also, TCP/UDP offloading is great in server environments but give me a break, the UDP load (and this is calculating checksums) for the average game will be freakin miniscule.

I give this one huge MEH! This is coming from a old school Juniper/Cisco head/Server Admin (AIX/Sun)
 
schizo said:
Offloading UDP isn't desirable as the packets are very simply constructed, connectionless, and have no flow control to reduce errors. That is why games use it-- it's much lower latency.
Click to expand...


But...in order to manage packets from apps, you pretty much would have to.

In other words, everyone is talking speculation at this point, so instead of slamming it, the tech might actually be done right in this.

At least wait for the real review before you pass judgement.
 
WesM63 said:
I read through the article and skimmed through most of the posts in this thread looking for someone to bring up what I am about to bring up. I am in no way an expert at what i'm about to say, but i do have experience in it and would consider myself somewhat knowledgeable.

First of all, the NIC is a great ideal, and it will sell no doubt about that.(i didn't say how much, but some people believe whatever they read)

The problem I see, and did not see questioned or mentioned in the article or in this thread is below.

From reading the article they are allowing you on the PC to prioritize traffic, also known as QoS. Thats all great and dandy, but let me break it down. QoS is based on few set of pre-defined numbers set for different levels or priority of packets. Most switchs/routers use a standard FIFO method, first in first out. Which in Bigfoot networks desgin and usage of the "QoS" feature to get your "pings" out first will *work*. (due to the fact that the first thing out of your pc and into the ISP's switch/router would be your pings)

Now going one step deeper. If the ISP has QoS implemented into they're switch/router things change drastically. They are no longer using FIFO, per se.(FIFO actually happens at the port level. QoS happens before FIFO. So technically they are still using FIFO, but the QoS decides in which order they get to the port) They are prioritizing incoming traffic and giving it a pre-defined set of numbers as I mentioned above. Generally Voice is given a CoS value of 5 or DSCP value of 46, making it a "High priority packet" (or "express forwarding" depending on who you talk too). Voice generally gets its own queue which is considered the "priority queue" and sent from the switch/router first (highest priority). All other traffic, depending on how it is marked is separated into 3 other queues and using WRR (weighted round robin) is sent in order of importance. Now, Internet traffic (most any internet traffic, pings included) are considered to be best effort traffic. In this case they are given a CoS value of 0 or 1 and a DSCP value of 0 or 8. Which by most QoS standards is the lowest level of traffic and the least important. (meaning begin sent AFTER voice or other high priority packets)

What does all this mean?

If your ISP has QoS turned on, your packets, providing they are begin marked with a 0 or 1 will NOT be the first thing sent from your ISP's router/swtich. Meaning the Bigfoot networks Killer NIC's QoS is fairly null. Its like ordering a product from newegg and using the expedited delivery and getting it in the same time as normal delivery option. You are at the total mercy of UPS/Fedex.

With QoS they are tons of variables, like the ISP only trusting certain devices to mark the packets with a CoS or DSCP value. (Say they're edge device, in many cases the CMTS. (where the cable modem terminates at the ISP's office)) The only real way to know what your ISP is up to is by reading the SLA (service level agreement).

Hopefully i didn't' confuse anyone lol.

BTW, UDP is a faster protocol than TCP, however it has NO retransmits. Which in gaming is fine.


EDIT:After re-reading my post it is still techinical. In lemans terms, the Bigfoot NIC may get the pings out of your home network faster, but you are at the mercy of your ISP. Meaning, it *could* improve ping times. I'd put a strong emphasis on could.

Then again i'm sure they know more about this than I do, maybe i'm just babbling about why I don't think it will work quite as well as advertised.
Click to expand...


QoS is basically reserved bandwidth for, um... 'background' traffic, but if the PC does not enable QoS, it will not effect bandwidth from the router it connects through.

We dont know if the nic enables QoS or not.
 
JerRatt said:
The only reason this pisses me off so much isn't because they are doing the same thing as other companies (using gaming as an advertisement tool only), it's because they are doing it with a product and charging thousands of times the price of a comparable card.
Click to expand...


Can you please find a comparable product for a thousand time less money? I want to buy that version please.

If you can't find something comparable, then STFU.

I suspect we may not hear back from you until you find your comparable product...so I wont be holding my breath.
 
FWIW, IGN is reviewing this thing...
Based upon some rough benchmarking, without the Killer NIC our test PC (AMD FX 62 2600, ATI Radeon X1900 XTX, 2 GB RAM) runs F.E.A.R. at 1024x768x32 with details maxed and 2xAA-2xAF at an average of 74 fps though two 15 minute sessions. Our ping to server ranged from 40-80ms, and our average worked out to 67ms over the two sessions.

After installing the Killer NIC we got back on the same F.E.A.R. server and played on the same map. The improvement with the Killer NIC was noticeable and impressive. Pings to server dramatically improved, generally ranging from 17-50ms, averaging 45ms over another two 15 minute sessions. FPS also got a nice bump, jumping our average from 74 to 86 fps.
Click to expand...
So a 20ms ping increase and 12fps framerate increase.
Worth $300? Probably not...but hey, at least it seems to be doing what they say it does.
(You do have to take into consideration the source in this case, though I'd think it'd be rather tough to botch a benchmark, but I've been wrong before).
 
Wow, I would have bet good money this product was BS. If those results are correct, all I can say is I'm really impressed.
 
BBA said:
QoS is basically reserved bandwidth for, um... 'background' traffic, but if the PC does not enable QoS, it will not effect bandwidth from the router it connects through.

We dont know if the nic enables QoS or not.
Click to expand...

Its not really "reserved bandwidth". QoS prioritizes traffic for specific services that require certain amounts of bandwidth or response time. It doesn't reserve bandwidth (per se) and it doesn't increase bandwidth. It simply re-organizes the packets into which ones are consider more important and sends them first. (Makes the current available bandwidth more efficient)

What i am saying is, even if the NIC does enable some form of QoS it may not help do to the fact on what the ISP is doing. Same as if the NIC does not enable QoS, it depends on what the ISP is doing.

The NIC could enable QoS or some form of packet priorization which makes sure your Pings and other game specific packets get sent first, however your ISP may have QoS turned on. In which case, the ISP may declare these packets best effort and throw them into a queue with other Internet traffic. Or the ISP may not have QoS turned on which would be back to using a standard FIFO queue, which may help your pings and game specific packets, if the NIC does do some sort of QoS or network priorization.
 
Sigh. I like it. I like it a lot - time to buy that ExATA board I was looking at... Too many good ideas for add-in cards these days!

What they need to do with the KNIC though (IMO) is give it a removeable ethernet chipset - so that when 10gbps comes along mainstream you could just upgrade the chipset and leave the proc and everything else there...

Or ya could just buy the next one with 128mb of ram and a 1ghz proc...

Oh btw - that USB port? Anyone see a matrix orbital/GLCD display being attached on there REAL soon? :)
 
BBA said:
Can you please find a comparable product for a thousand time less money? I want to buy that version please.

If you can't find something comparable, then STFU.

I suspect we may not hear back from you until you find your comparable product...so I wont be holding my breath.
Click to expand...

How about onboard network cards that are basically free, jackass. This isn't 2000 where onboard NICs were crap and had no real features besides I/O. Look at all the bundled and integrated technology on the new nvidia NICs. Hell, I've seen some pretty decent onboard NICs that take such a miniscule hit on performance that it's only registerable by the hundredths of a factor in synthetic benchmarks,

If you wanna talk exact comparables, then you can still find offloading NICs with plenty of packet shaping and prioritizing technology in the market from Intel as well as many other NIC developers. Maybe not thousands of times less money than the onboard NICs, but no where NEAR $280. So how about you, STFU! It'd be better than bashing someone's post mostly with derogatory remarks and the only real content being a question that is easily answered.
 
Well...your completely wrong.


Did it even occur to you that the guy making this NIC actually made the Intel NIC's?

But, last time I priced an offloading gig-E NIC it was just a tad pricier than this one and did not have a unix console you could access.


I'm sorry I hurt your feelings, but that doesnt change the fact. You should watch your mouth...err fingers, around here as well. You failed to provide anything reasonably close, you failed to heed Kyles suggestions and your basically making an idiot of yourself.
 
BBA said:
Well...your completely wrong.
Click to expand...

No, he's not. He's just stating that perhaps 280 is a bit OTT for what is, at the end of the day, an NIC. That's his opinion and he does make some good points - integrated NICs are far beyond what they used to be.

That won't stop me buying one - because of the other things it can do (firewall/TS sounds very nifty), which is essentially the point kyle makes in the article.

BBA said:
Did it even occur to you that the guy making this NIC actually made the Intel NIC's?
Click to expand...

And the Price of Rice In China is?

The guys who worked on the 7900GTX also worked on the 7300. Does that make it a good product automatically? erk hell no :) Like everything else - don't buy it because your favourite company made it. Or EEEng for that matter.

BBA said:
But, last time I priced an offloading gig-E NIC it was just a tad pricier than this one and did not have a unix console you could access.
Click to expand...

Depending on what you mean by "offload" yes and no, as for the second bit hell yeah :) . This is certainly the first bit of consumer kit I've seen with stack bypass functionality - extra functionality over and above an NIC is what makes the value of this product.

It's just going to be a bit of a "wait and see" until it's there... Or a "dive-in headlong and develop" :)

BBA said:
I'm sorry I hurt your feelings, but that doesnt change the fact. You should watch your mouth...err fingers, around here as well. You failed to provide anything reasonably close, you failed to heed Kyles suggestions and your basically making an idiot of yourself.
Click to expand...

Dude that's really just trolling...
 
Instead of whining about what another TOE offloader will cost, why don't we just look it up and compare prices and then ask Kyle to compare these NICs to the Killer?

So far what I've dug up is theres TOE in the Broadcom BCM5708C or BCM5708S chipsets, though I havn't yet found a card that you can buy standalone that has this chipset.

Paydirt: HP sells one with the 5708S chipset called the NC373F. It costs... $700!
http://h30094.www3.hp.com/product.a...21&pagemode=ca&jumpid=in_3924_cote_394793-B21

PS, from the TOE Wiki (http://en.wikipedia.org/wiki/TCP_Offload_Engine):
Gigabit TCP communication using software processing alone is enough to fully load a 2.4 GHz Pentium 4 processor, resulting in little or no processing resources left for the applications to run on the system.

Somebody else want to pick up the leg work and find some other TOE cards?

PSS: It appears that the e1000 (The super Intel card) doesn't support full TOE, but is "good enough" and weighs in around $150-200 as the Intel Pro/1000 MT model.\

PSSS: Alacritech (never heard of them even) appears to offer a PCI solution which costs ~$800. Model #1000X1 http://www.alacritech.com/html/1000x1_accelerator.shtml

Seriously, how come nobody even BOTHERED to look?

PSSSS: Perfisans offers one but I can't find a price. Model# ENA-5031 http://www.perfisans.com/products/EthernetAdapters/ENA5031.aspx
 
You must log in or register to reply here.
Back
Top