Printable View
you can use the money saved to buy 3 video cards that you couldnt get to work on the server board in the first place.I mean since we are talking about ultimate gaming now who will use a server board for a gaming rig anyways?Quote:
Originally Posted by eXa
Ohh and those chips are duel cores and a 2600k is a quad with hyperthreading.I would wait for the next gen sandys to come out to build a new gaming rig,but bang for the buck favors sandy b hands down.
you guys do know that you can just put 1 in a regular x58 board right ;) I keep hearing pair in sr2 but 1 in an R3E will work.
Though I do agree that sandy is better in every aspect, you could bin 10 of them and still cost less.
Its a dual core, and there is plenty of games that like more than 2 cores. And the argument was for a "ultimate gaming rig".
Thank you. I just skimmed through it, will read later, it certainly seems interesting.Quote:
Originally Posted by dartaz
I don't care how easy it is, my computer does many things, including processing of data I care about. That's why OC is not an option. I stopped OCing when I realized that stability tests recommended at the time (24h of super pi) were useless - they could show that a computer is unstable, but because they stress a very limited subset of CPU functions, they are a very slight suggestion that it may be stable. Designing a good test is well beyond my capabilities and I doubt anybody capable decided it's worth the time.Quote:
Originally Posted by eXa
And back to somewhat more general case, I know 1 person that runs OCed computer. Most runs laptops, totally stock, some desktops and all but 1 are stock. I'd classify ~50% of my friends who I talk about computers with as gamers. I don't claim the group is representative, but I'll repeat that XS ain't representative either. I haven't seen any research on the topic, so it's moot.
If we're talking about the best machine money can buy for some workload and sb. says to ignore the more expensive option, because a cheaper one is better, it's clear to me that disregarding a cheaper one because is cheaper makes no sense. Disregarding the more expensive one doesn't make it either. So that's how we ended up with this comparison.Quote:
Originally Posted by Hornet331
Quote:
Originally Posted by Skratch
these are dual cores with HT to be accurateQuote:
Originally Posted by eXa
Intel seems to think alot of people overclock and is the whole reason of putting out the K series sandy bs that come with an unlocked multy and push overclocking to gamers.
super pi is good for getting your ram stable,there are other programs that stress the cpu more than anything you could ever think of doing.
No. I don't have knowledge specific to CPU testing, but I know some things about testing in general. And the most basic rule is that if you don't test something, it may be broken.Quote:
Originally Posted by Skratch
I'll just talk about spi, for the rest the same applies.
First, I guess you mean the original Super pi, single threaded. It doesn't check what happens when all cores are loaded. Also, do you tie it manually to each of cores or just verify one of them?
If you meant some multithreaded mods then you obviously miss turbo.
But that's just the start. What exactly does spi test and what it doesn't? I'll make guesses, may be very inaccurate but it doesn't matter - what does is that some important parts are totally unstressed.
http://i1.sinaimg.cn/IT/cr/2010/0205/3031036536.gif
Does it stress:
-memory controller? yes
-queue? no
-L3? Yes
-core? Yes
-QPI? No.
-rest of uncore? ??
Now we should dig into individual core:
http://upload.wikimedia.org/wikipedi...halem_arch.svg
What parts of it are stressed? Here for most blocks I could make only 100% random guesses, but SSE seems untested and I highly doubt that spi stresses both integer and FP - most likely one of them is idle or almost idle.
You can dig deeper, obviously. And there's a point when it makes sense to stop. Where is it? For me definitely at least the 2 mentioned diagrams would have to be covered. And not with guessed but with verified data. Possibly you could assembly a good test from a set of available banchmarks. But as I said, I doubt anybody did it, now even more because you didn't mention it.
EDIT: no, I wouldn't check if the 2 diagrams are covered. I would talk with a pro, cause I sure miss a lot.
Another thing is environment. What is environment temperature when you test? Do you make sure it will never get hotter?
Furthermore how do you account for CPU wear? Normally it's negligible, but how is it with OCed systems? I don't know. But I know that when Appro evaluated the decision for their OCed servers, they decided it's best for them to stay entirely on the safe side and give just 1 month warranty.
Do you leave a safe margin and periodically check that it's still there?
So that's why OC is not worth it for me. I could OC my GPU because it doesn't handle anything important. But I don't care about how the games look, so it's a waste of time.
I will post a picture on how I test my oc,I run super pi 32m(that stresses your ram big time) and run intel cpu burn test,an intel cpu streesing program that intel uses to strees there chips and I run linx all at the same time.I am running a closed air cooled 24/7 setup and I have not had one blue screen or problem with my overclock.The only thing holding me back is the fear of 1.55 volts and heat that is keeping me from pushing my chip to well over 5ghz.I bench it just for fun at 5.2-5.3 but stressing the chip brings massive heat with my air cooled heat sink.
I have pushed the cpu to almost thermal shutdown and this stress test I do will put more stress on it than any normal day to day programs.
and yes everything is monitored.
I overclock my gpu because I encode hd blurays and use the power of the cuda cores to encode about 70% faster.
I had to run linx on and off because it was maxing the cpu thernal threshold.Without linx it was hitting 82c high on the cores and after 10 seconds of linx it hit well over 95 and when I would se 98 I would kill it and restart when they cooled down.
http://a7.sphotos.ak.fbcdn.net/hphot..._5346251_n.jpg
There are many advantages to this chip:Quote:
Originally Posted by Skratch
1) Memory capacity. One can easily run 192GB at full speed with 12x 16GB DIMMs at 1333MHz with these.
2) The ability to run in pretty much any server board. One could buy these, underclock them and have a very cool running server. With the dual QPI links, these chips can be also be run in a dual IOH motherboard providing incredible I/O capability and many, many PCI lanes.
3) Two of these chip will still allow you to run 8 threads...still very decent multithreading ability at a very high clock speed.
4) Total reliability at said clock speed....they're Xeons....enough said...;)
Now you guys have gone besides the point. The discussion here was about the "ultimate gaming machine"
"I dont OC because i dont know how to check if my pc is stable" (24h superpi, wtf?), "i do other important work", 192GB ram etc.... Moot point! The argument was for GAMING, and SB is better no doubt about it.
@M^2: Do you really belive a 4.5ghz SB with close to stock volt really deteriorate that much faster than @ stock clock? (you know, you dont HAVE to push it to max multi with 1.5v etc,)
@ Radaja: Yes, and 2600K also have HT, tho probably not very helpful for gaming anyways.
HT can work well in some games to,
but normally the games are not using efficiently more than 3 "cores" anyway, so the best way to see some gains is in the dual core with HT like this (and not in a quad core with HT):
http://img.inpai.com.cn/article/2011...7c2b409fe1.png
http://www.inpai.com.cn/doc/hard/144022_9.htm
look at the 530 gain, quite nice some 40% increase with HT?
Yeah, but that is only dual cores. 2 X5698 or 2600K is 4 cores/8 threads wich wont give an increase like that.
Edit: Still very impressive increase from HT :)