KillerNic
- Thread starter mzkhadir
- Start date
[Reading between the lines] This is going to bring out a new generation of aimbots/wallhacks/general cheats. Running on a different abstraction layer than Windows means that PunkBuster won't be able to see the cheating. About the only way you'd be able to stop it is to blacklist it from the game, and even that wouldn't really work as you can always change the device name and even fudge the PCI vendor ID.
I'll have to see what it's actually capable of once it comes out, but if it really is a Linux computer on a PCI card, and you can talk to it as though it were a regular computer, I may very well buy one just to play around with.
I'll have to see what it's actually capable of once it comes out, but if it really is a Linux computer on a PCI card, and you can talk to it as though it were a regular computer, I may very well buy one just to play around with.
Caught this on Slashdot this morning. Aside from the technical question of whether network i/o is cpu-bound enough to benefit from a coprocessor solution, you have to laugh at the marketing spin. Look at their web site. It's like a horror movie trailer that got stuck on the first note of the theme song.
I'm looking over the white paper, but on the face of it I'm skeptical just because of the spin and price. Not only is it the most advanced NIC ever, but it runs Linux, has a dedicated USB port, and can offload file sharing programs, too!
I bet all the people who write radar programs will buy one.
Maybe all those packet operations they talk about can use a coprocessor. I'm not sure, because I don't think it's CPU bound in a typical game, and with the growth of parallelism this doesn't seem to me like the problem that needs solving. You'd probably benefit more from QOS on the network path to the game server. But even if that is a good application, what do you need all the other stuff for? If network I/O benefits from a coprocessor why open it up to user code that might impinge on its performance? Maybe they have enough horsepower on the card that it doesn't matter.
I would play around with one if I could get it for free, I guess. But not for $250+.
Update: some tech specs from their white paper.
Processor: 400 mhz "Network Processing Unit"
Memory: 64 megs DDR
Interface: 32-bit PCI
NIC: 1000m, 100m, 10m
Ports: USB 2.0 (dedicated to linux core)
O/S: Modified linux core w/proprietary "Flexible Network Architecture"
They show a test in the white paper that claims the card can poll for data at about 2.5 khz. vs. about 1.5 khz. for the Intel pro 1000. Another test claims throughput of 28 vs. 12 megs per second.
The tech founder is out of Intel's networking group. The other two were in finance at DELL, and sales at AMD.
I'm looking over the white paper, but on the face of it I'm skeptical just because of the spin and price. Not only is it the most advanced NIC ever, but it runs Linux, has a dedicated USB port, and can offload file sharing programs, too!
I bet all the people who write radar programs will buy one. Maybe all those packet operations they talk about can use a coprocessor. I'm not sure, because I don't think it's CPU bound in a typical game, and with the growth of parallelism this doesn't seem to me like the problem that needs solving. You'd probably benefit more from QOS on the network path to the game server. But even if that is a good application, what do you need all the other stuff for? If network I/O benefits from a coprocessor why open it up to user code that might impinge on its performance? Maybe they have enough horsepower on the card that it doesn't matter.
I would play around with one if I could get it for free, I guess. But not for $250+.
Update: some tech specs from their white paper.
Processor: 400 mhz "Network Processing Unit"
Memory: 64 megs DDR
Interface: 32-bit PCI
NIC: 1000m, 100m, 10m
Ports: USB 2.0 (dedicated to linux core)
O/S: Modified linux core w/proprietary "Flexible Network Architecture"
They show a test in the white paper that claims the card can poll for data at about 2.5 khz. vs. about 1.5 khz. for the Intel pro 1000. Another test claims throughput of 28 vs. 12 megs per second.
The tech founder is out of Intel's networking group. The other two were in finance at DELL, and sales at AMD.
My first thought was "Retarded".
Now that I've had time to think about it, I stand by my first thought.
Now that I've had time to think about it, I stand by my first thought.
What sort of nonsense is this? They actually report throughput in "MB", not MB/s. And what sort of crap could only do 28 MB/s? Even if we assume that they were inventing a new unqualified measure, MB, they claim > 2x performance over nVIDIA, which is nonsense.
My nVIDIA NIC here does > 97 MB/s UDP throughput in a test that I just ran using the same tool that they used -- PCATTCP. Maybe it can do better with more tweaks, etc. (1) It's a heck of lot better than the "28" they report. (2) There's no way that anything can be twice as fast, because there ain't that much headroom here folks. Can't do 194 MB/s over 1 GbE, regardless of marchitecture.
but....but....It's got "ultimate ping" technology.
That means it's elite and the "best". I'm only getting killed so often because of my network card. I'm uber.
<*snicker*>
That means it's elite and the "best". I'm only getting killed so often because of my network card. I'm uber.
<*snicker*>
Well... PPUs actually have some remote use for those crazy people. They just havn't designed it well and the games don't have enough support for it yet. This killerNIC is just bullshit on a golden platter.
As you can see above, I didn't get the Killer NIC's stated performance results. So I decided to follow up on it with them, on their web site:
http://www.endlagnow.org/ELNForums/Topic400-12-1.aspx
I asked for details on how they did their tests, and once I got it, I reproduced them almost as closely as I could. I cheated a bit in this manner -- I spent a bit of time tuning the NIC parameters so that they'd get better performance. Once I'd done that, I didn't feel a great need to report the untuned performance, and so reported the tuned performance. I think this is the fairer test -- after all, what do we know of the Killer NIC's internal tuning, and isn't it fair to raise the bar to the point that the other NICs are actually capable of performing?
Here are some sample results -- more details are in the linked thread.
nVIDIA nForce (Bigfoot) 9.59 MB/s
Marvell Yukon (Bigfoot) 11.93 MB/s
Intel Pro 1000 (Bigfoot) 12.06 MB/s
Killer (Bigfoot) 20.64 MB/s
nVIDIA nForce (Madwand) 77.1 MB/s
Marvell Yukon (Madwand) 80.25 MB/s
Intel Pro 1000 (Madwand) 53.14 MB/s
http://www.endlagnow.org/ELNForums/Topic400-12-1.aspx
I asked for details on how they did their tests, and once I got it, I reproduced them almost as closely as I could. I cheated a bit in this manner -- I spent a bit of time tuning the NIC parameters so that they'd get better performance. Once I'd done that, I didn't feel a great need to report the untuned performance, and so reported the tuned performance. I think this is the fairer test -- after all, what do we know of the Killer NIC's internal tuning, and isn't it fair to raise the bar to the point that the other NICs are actually capable of performing?
Here are some sample results -- more details are in the linked thread.
nVIDIA nForce (Bigfoot) 9.59 MB/s
Marvell Yukon (Bigfoot) 11.93 MB/s
Intel Pro 1000 (Bigfoot) 12.06 MB/s
Killer (Bigfoot) 20.64 MB/s
nVIDIA nForce (Madwand) 77.1 MB/s
Marvell Yukon (Madwand) 80.25 MB/s
Intel Pro 1000 (Madwand) 53.14 MB/s
I'm waiting for the threads to start up talking about how to overclock it, and how to mount an LED fan to the heat sink.
I guess nobody has told these Mensa Candidates that your average Cable/DSL modem runs at 10mbs.
I guess nobody has told these Mensa Candidates that your average Cable/DSL modem runs at 10mbs.
Madwand, I don't get what your results mean.
Are you saying that bigfoot people setup the test bench to deliberately slow down the standard nics?
Are you saying that bigfoot people setup the test bench to deliberately slow down the standard nics?
That thing is so scary. It looks like it could be used as a weapon. maybe they could place it in army workstations. if they run out of bullets they can just take the nic out and use it as a weapon.
i had assumed that you were only as fast as your slowest part. so how does this nic compensate for faulty network equipement in between your pc and game server? (IE: crappy dsl modem. or dslam.)
i had assumed that you were only as fast as your slowest part. so how does this nic compensate for faulty network equipement in between your pc and game server? (IE: crappy dsl modem. or dslam.)
I'm not exactly sure why Bigfoot's performance numbers were so low (when my results were so much higher), and neither are Bigfoot, which is why they're re-doing their tests.
Their last word on this was that they screwed it up in part by having the receiving side running 100 Mb/s. This is also incredible, because they reported performance around 160 Mb/s in one case -- which would be impossible with 100 Mb/s on one end. Their counter-point on this was that the switch might have buffered the data, and hence given an inflated result. Not sure I'd belive that either.. Bottom line is that if this was their test, then it's not a very good test, and they have to re-do their benchmarks.
I have no reason so far to believe that they faked their results or deliberately slowed down the other NICs -- they've been pretty up-front about their tests in my discussion with them. I believe that they have a systemic problem somewhere that slows down their performance. Perhaps it's their Dell switch. Perhaps it's a combination of effects such as a different (older) nVIDIA implementation + lack of NIC tweaking + their systemic problem.
However, intentionally or otherwise, they showed much worse performance from their competitors than can actually be achieved by them, so as usual we'll have to take such vendor-supplied comparisons with a grain of salt, and rely on independent validation, ideally incorporating tweaking as I did, especially when it can give significant performance improvement.
Their last word on this was that they screwed it up in part by having the receiving side running 100 Mb/s. This is also incredible, because they reported performance around 160 Mb/s in one case -- which would be impossible with 100 Mb/s on one end. Their counter-point on this was that the switch might have buffered the data, and hence given an inflated result. Not sure I'd belive that either.. Bottom line is that if this was their test, then it's not a very good test, and they have to re-do their benchmarks.
I have no reason so far to believe that they faked their results or deliberately slowed down the other NICs -- they've been pretty up-front about their tests in my discussion with them. I believe that they have a systemic problem somewhere that slows down their performance. Perhaps it's their Dell switch. Perhaps it's a combination of effects such as a different (older) nVIDIA implementation + lack of NIC tweaking + their systemic problem.
However, intentionally or otherwise, they showed much worse performance from their competitors than can actually be achieved by them, so as usual we'll have to take such vendor-supplied comparisons with a grain of salt, and rely on independent validation, ideally incorporating tweaking as I did, especially when it can give significant performance improvement.
I think I might have to punch someone in the face if they even suggested this crap to me. Why? Because when someone says, "The 'KillerNIC' is the ultimate gaming network card," I translate that into, "You look like a moron, and I think you're stupid enough to buy this snake oil product, and you're SO stupid I think you'll even pay $300 for it." That's grounds for a punch in the face. I can't belive anyone is even taking this thing seriously! I'll rig a machine in a blind test with a 10/100 card and another machine with the "KillerNIC", and I'm sure no one would be able to tell the differnce when gaming. I'm not even going to include a 1Gb NIC in the test. That's how worthless I think this product is. And this is on a LAN, which most people playing games don't play on. Most people play over the net which makes this product look even more rediculous.
Yeah, suggesting physical violence is the uber cool and highly intelligent. :disgust:
Would I recommend the Killer NIC? Well, I don't think so. I just have an interest in NIC performance/benchmarks at this time, and it's some additional fun. There's no need to get angry or hugely emotionally involved about any of this. It's fun and games isn't it?
Would I recommend the Killer NIC? Well, I don't think so
. I just have an interest in NIC performance/benchmarks at this time, and it's some additional fun. There's no need to get angry or hugely emotionally involved about any of this. It's fun and games isn't it?Originally posted by: n0cmonkey
As ridiculous as mispelling ridiculous?
God damnit dude, you had to point out my spelling mistakes, didn't you.
For the record, I wouldn't take offense to people recommending me the "KillerDictionary." Mad, I wasn't serious about punching someone in the face. I'd hate to be in front of a judge when he asks, "Why did you assault this person?" "Well your honor, he recommended the KillerNIC to me. He said it was the ultimate gaming NIC, which we all know is a lie. So I punched him."Originally posted by: Madwand1
Yeah, suggesting physical violence is the uber cool and highly intelligent. :disgust:
Would I recommend the Killer NIC? Well, I don't think so
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 24K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 19K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter