P67, X58, And NF200: The Best Platform For CrossFire And SLI

[Crashman]

[citation][nom]ortoklaz[/nom]I wish this article was published earlier in order to stop all the trash talking and the PCIe dilemma..great job Crashman !@banthracis..good point (games vs threads ..simple test open/stretch Task Manager when playing BFBC2 ) Long live P67 !Thanks again Crash[/citation]We do this kind of coverage about once a year, but people forget and changes in technology require constant reevaluation. Oh, and BTW, you're welcome!

 

[Mousemonkey]

Oh certainly, one wouldn't choose X58 for a new build; but there's no need to replace an existing X58 system.

Oh for sure dude, but I need to replace a dying 775 Quad core rig and this article has cost me £400 and counting! :lol:

 

[Crashman]

Oh for sure dude, but I need to replace a dying 775 Quad core rig and this article has cost me £400 and counting! :lol:

You're the kind reader I had in mind! Well, except for the "moderator" thing, but I try not to be judgmental

 

[Mousemonkey]

You're the kind reader I had in mind! Well, except for the "moderator" thing, but I try not to be judgmental

After having spent another £400 on graphic cards some ten minutes ago I'm kinda hoping that your next article is going to be on flower arranging or embroidery or anything that doesn't make me want to get my wallet out! :lol:

 

[Crashman]

After having spent another £400 on graphic cards some ten minutes ago I'm kinda hoping that your next article is going to be on flower arranging or embroidery or anything that doesn't make me want to get my wallet out! :lol:

Since I encode video as often as I game, I'm waiting for Z68 before I upgrade.

 

[Mousemonkey]

I don't really have the option to wait as I've been putting this off for a few weeks now but my P35 board has pretty much given up the ghost now what with the SATA port headers falling off the board an all.

 

[eshwar_andhavarapu]

So, in short, single monitor gaming upto dual sli/crossfire has negligible benefits with a NF200 chip right? Based on newegg, the NF200 is a $60 premium which i'd rather put into a graphics card/RAM ! Especially because there are plenty of good stock LGA1155 motherboards around!

 

[Hard_Rain]

[citation][nom]Crashman[/nom]Tell it to The Judge. Really, I've been trying to get us all on triple-displays for a long time. It's really a cost-benefit analysis for a site that bought a bunch of 2560x1600 displays around two years ago.[/citation]

One way to determine if triple displays makes sense for your research is to analyze your readership. The assumption is that Tom's Hardware (TH) is a site for gamers and enthusiasts. However, I am sure that TH gets readers from a broad spectrum of the higher-end PC market. For instance, I have never played a computer game, yet just purchased a high-end X58 system for trading the financial markets. Hence, I need support for multiple threads running on multiple (3-12) monitors. While my needs are somewhat different than a gamer, the fact remains that TH publishes some of the best computer hardware reviews in the world. Keep up the great work!

 

[Why_Me]

Just say no to the over priced boards with the NF 200 bridge is what I got out of that article. Cheapest NF 200 board in the states is the Asus Revolution @ $260 USD. meh...way over priced as I can see by that article. Good job on the benchies

 

[JohnnyLucky]

Nice article. I don't plan to upgrade this year. I am perfectly happy with my current rig.

 

[Marcus52]

I wouldn't call x58 dead for gaming because it performs "only" at 98%, on AVERAGE, as good as P67 in these tests.

If you are going to draw conclusions about the CPUs and the platforms they run on, wouldn't it make more sense to run games that are CPU intensive? Really, making flat statements about hardware performance in ALL games is wrong as soon as the words leave your mouth. Not ALL games have the same hardware requirements and respond to the same improvements.

The top 3DMark 11 scores are still held by x58 based systems.

Now, I'm not saying a person should necessarily go out and build a new rig based on x58 - but if you have x58 already there's no point in throwing it away because P67 is "better". Frankly, while the current Sandy Bridges are good, I'm waiting for the next enthusiast level improvement for the next build I make for myself. Anything I build for someone else will be on Sandy Bridge, especially if the budget is tight. However, it's not a huge improvement over the previous Core architectures, it's a minor one, at best, and the best argument for it over CPUs using socket 1366 is the price.

 

[Guest]

LGA 1155 dominates 1366. Keep in mind this review doesn't take into account a really hefty overclock on the 1155 processor - with the right board and a great heat sink or water you'll be up to 4.5-5ghz - or the fact that you'll be able to prolong the lifespan of your 1155 board by upgrading to a better processor.

 

[ubercake]

One thing I noticed by reading this article over again...

I have an X58 system (that I don't overclock) which regularly exceeds 750W while gaming with 2 GTX 580s. If I added another 250W power hog of a 580 (actually rated at 244W), my system would easily exceed the power output of the 1000W power supply used in the testing.

How is a 1000W power supply sufficient for the 3-GTX 580 portion of the test especially considering the processor was OC'd to 4.0 which could demand anywhere between 50 and 100W more from the system than a non-OC'd system? How was there enough power for a 3rd GTX 580 in this test? We they underpowered?

 

[banthracis]

Only way you'll get 750W+ power usage w/ 2 580's is in furmark.
Second, consumer level power meters measure power draw from the outlet, not actual power usage. Hence, 777W power draw under furmark for 2 GTX 580s
http://www.anandtech.com/show/4012/nvidias-geforce-gtx-580-the-sli-update

with 85% efficiency, nets you 660.5W of actual power usage.

Throw in another 250W for a third 580 and you'll see a meter increase of probably 292W, but actual power usage will be about 900W.

Hence, a 1000W PSU is perfectly fine. In fact, good OEMs will underrate their PSU's to a degree and most will run fine even at 110% load for extended periods.

 

[ubercake]

[citation][nom]banthracis[/nom]Only way you'll get 750W+ power usage w/ 2 580's is in furmark. Second, consumer level power meters measure power draw from the outlet, not actual power usage. Hence, 777W power draw under furmark for 2 GTX 580shttp://www.anandtech.com/show/4012 [...] sli-updatewith 85% efficiency, nets you 660.5W of actual power usage. Throw in another 250W for a third 580 and you'll see a meter increase of probably 292W, but actual power usage will be about 900W. Hence, a 1000W PSU is perfectly fine. In fact, good OEMs will underrate their PSU's to a degree and most will run fine even at 110% load for extended periods.[/citation]
So you're saying the power draw as measured by an appliance load tester with only the PC plugged into it will measure the draw of everything plugged into the outlet? I get in the 750Ws playing Medal of Honor.

I have a Corsair TX950 on this system.

Is this something you know about through experience or is it just something you've read about?

 

[ubercake]

[citation][nom]ubercake[/nom]So you're saying the power draw as measured by an appliance load tester with only the PC plugged into it will measure the draw of everything plugged into the outlet? I get in the 750Ws playing Medal of Honor. I have a Corsair TX950 on this system. Is this something you know about through experience or is it just something you've read about?[/citation]
Another thing... the appliance load tester reads '0' when nothing's plugged directly into it, yet other things are turned on and plugged into other receptacles on the same outlet. It's not measuring the power draw from any other devices.

 

[banthracis]

No, appliance tester measure the PSU's draw from the socket, not what a PSU uses. PSU's draw more power than they use, that's what efficiency is.
a 85% efficient that draws 100W from the socket is actually only using 85W.

This is stuff covered in physics 101...

 

[ubercake]

[citation][nom]banthracis[/nom]No, appliance tester measure the PSU's draw from the socket, not what a PSU uses. PSU's draw more power than they use, that's what efficiency is. a 85% efficient that draws 100W from the socket is actually only using 85W. This is stuff covered in physics 101...[/citation]
So... Is a 1000W PSU rated to power 1000W worth of equipment? And the actual draw from the socket at max would actually be (1000W/x=85/100) 1176.47W?

Or are you saying a 1000W power supply can only draw a total of 1000W from the wall and, due to 85% efficiency, only supply components with 850W?

Which is correct?

 

[banthracis]

[citation][nom]ubercake[/nom]So... Is a 1000W PSU rated to power 1000W worth of equipment? And the actual draw from the socket at max would actually be (1000W/x=85/100) 1176.47W? Or are you saying a 1000W power supply can only draw a total of 1000W from the wall and, due to 85% efficiency, only supply components with 850W? Which is correct?[/citation]

Um, power draw from wall has nothing to do with what wattage a PSU is rated to hold.

1 is a measure of how much load a device can handle, the other is a measure of efficiency at a given time, two completely different things.

Your question is basically ridiculous.

It's like being told a plane is rated to reach 1000 mph.
Then you ask, so if my plane can reach 1000 mph, does it mean it can reach 1000mph or does it mean that it's fuel efficiency at 1000 mph is 2 gallons per mile?

My response:
Um, yes... a 1000mph rated plane will fly at 1000 mph. WTH does it's fuel efficiency have to do with its max rated flying speed?

Or just think of it this way.
A 100w PSU with 85% efficiency will draw 118W at the socket at full load.
A 82.6w PSU with 70$ efficiency will also draw 118w at the socket at full load.

why on earth do I think power draw at socket has anything to do with a PSU's rated wattage?

 

[ubercake]

[citation][nom]banthracis[/nom]Um, power draw from wall has nothing to do with what wattage a PSU is rated to hold. 1 is a measure of how much load a device can handle, the other is a measure of efficiency at a given time, two completely different things.Your question is basically ridiculous. It's like being told a plane is rated to reach 1000 mph.Then you ask, so if my plane can reach 1000 mph, does it mean it can reach 1000mph or does it mean that it's fuel efficiency at 1000 mph is 2 gallons per mile?My response:Um, yes... a 1000mph rated plane will fly at 1000 mph. WTH does it's fuel efficiency have to do with its max rated flying speed?Or just think of it this way.A 100w PSU with 85% efficiency will draw 118W at the socket at full load. A 82.6w PSU with 70$ efficiency will also draw 118w at the socket at full load. why on earth do I think power draw at socket has anything to do with a PSU's rated wattage?[/citation]
@Banthracis - Though your approach isn't all that pleasant, I have to say thanks because I learned something valuable through the information you've provided. I must have not been paying attention during that part of my physics 101 class (or I've just forgotten). I did find a good power test that illustrates the draw versus the power provided to the components through my own psu:

http://www.jonnyguru.com/modules.php?name=NDReviews&op=Story2&reid=170

It's totally consistent with what you're saying. Thanks again.

 

[banthracis]

I don't mind taking the time to teach people engineering ubercake, and I'm happy you were able to learn something new.

I also don't normally take a condescending tone, but your responses when I attempted to correct your misconceptions, and inform you about AI and threading earlier weren't exactly endearing.

However, for your more mature attitude this time, thanks appreciated and apologies for the tone.

 

[Guest]

I have an idea about multithreading AI
you are a unit in the real world and you have a behavioural program full of if clauses
So if you are a soldier you take commands from the sergeant and your sergeant takes commands from the lieutenant and so on
But if you are wounded in a battle you ask for help or you retreat from battle
Similarly,if you have a supreme commander in a battle game,that commander assigns commands to his lieutenants who are randomly selected from the produced units and so on..
When a ranking unit dies,another one is picked up and promoted by the main AI
so every unit has a simple behavioural mechanism:if I'm healthy ,I obey commands,if I'm not healthy I retreat an'd give commands to my units to regroup

 

[banthracis]

I have an idea about multithreading AI
you are a unit in the real world and you have a behavioural program full of if clauses...
When a ranking unit dies,another one is picked up and promoted by the main AI
so every unit has a simple behavioural mechanism:if I'm healthy ,I obey commands,if I'm not healthy I retreat an'd give commands to my units to regroup

That is basically the definition of a behavior tree, and already done.

Doesn't solve the issue of a single main AI though. Something has to track all the individual units running around and their current status. Something has to track each and every bullet fired to determine where it goes and what it hits. Something has to then check each item in the behavior tree to see if it's doable, IE tell individual AI units that it's been hit. These somethings have to be in constant communication. Remember, in real world physics and cause and effect determine what happens as a result of an action. In a game world all of it must be calculated.

To put the complexity into perspective, in a single millisecond, a human folds more proteins than the entire folding at home project have folded, ever. In fact, it would take all the computers in the entire world a couple thousand years to fold as many proteins as a human does in a millisecond. That's the sort of complexity your comparatively puny little CPU is trying to approximate.

Further AI is more than just bots. It's animation decisions, sound playing, physics tracking, pathfinding, etc.

AI isn't programmed to treat each bot separately, it's inefficient. It's much more efficient to have 1 animation protocol for ex, and have each bot "ask" that protocol for what it needs to do at a given time, given current conditions. Since each AI bot doesn't need to consult this protocol nonstop (a 5 second animation is a very long time in processor terms), it's stupid and wasteful to run a copy of this protocol for each AI bot running around. In fact, this "bot" doesn't even exist as a separate entity with it's own isolated thinking. It's merely an object whose current status is determined by a lot of central protocols.

I could keep going, but suffice to say that AI is a LOT more complicated than you imagine, and indeed, the basis behind which we program AI is nothing like how individual humans think. If you are interested, take some courses in computer science. Trust me, most of the idea's you guys think up have been considered and tried at some point.

Video games are just a tiny application and not even a big player in AI research. Many of the top minds in the nation are recruited to just do AI research. Why? Wall street. Firms will pay millions to get the top physicists, mathematicians and computer scientists, to make their trading programs AI better.

 

[ubercake]

[citation][nom]banthracis[/nom]I don't mind taking the time to teach people engineering ubercake, and I'm happy you were able to learn something new. I also don't normally take a condescending tone, but your responses when I attempted to correct your misconceptions, and inform you about AI and threading earlier weren't exactly endearing.However, for your more mature attitude this time, thanks appreciated and apologies for the tone.[/citation]
Thanks again for the info.

 

[davewolfgang]

Video games are just a tiny application and not even a big player in AI research. Many of the top minds in the nation are recruited to just do AI research. Why? Wall street. Firms will pay millions to get the top physicists, mathematicians and computer scientists, to make their trading programs AI better.

I've worked at a worldwide commodity firm for the last 7+ year - and the tech here is as close to the best/latest as I've seen for any other company I've worked for.

They throw out 6x and 8x (23" widescreens) display setups regularly, and multi-CPU run 12 disply setups for the traders and researchers here like candy at Easter. Even the Assistants (they don't call them Secretaries anymore...) have dual monitor setups.

We actually have here in the building a predictive weather system that's bigger than the National Hurricane Centers, and more accurate too.

So a lot of these people not only need the multi-monitor support, but also multi-core/threaded CPU's to power the programs that throw all that stuff up on those multi-monitors. I still rarely see any advertising for the "business class" video cards that are used here, and their setups for actual "work" - it's all about the gaming. But then even the multi-monitor set ups aren't really using the graphics for high speed "movement" and "action" but just to power the screens - when they are maxing out a 4-CPU/24-core server at 90%+ on all 24 cores.