|
When using this resource, please read FragKrag's opening post. The Tech Support forum regulars have helped create countless of desktop systems without any compensation. The least you can do is provide all of the information required for them to help you properly. |
On March 20 2013 12:02 Myrmidon wrote:Maybe my CPU-Z is old, but since when do they show 5 digits for the bus speed (edit: checked some more images, and happens sometimes I guess), and since when do people actually land on exactly __00.00 MHz for clock speed? Show nested quote +On March 20 2013 11:22 Craton wrote:R.I.P. one of my Acer x223W's.  Flickering backlight + annoying noise + slight smell of burning when close to vent. Time to buy something better than an Acer x223W??
I have a U2713HM. You don't need good monitors when they're just used for browsing or irc.
I have two x223W's and even if it was my oldest that failed, it wasn't very old at all (Dec. 15 2010 / Sept. 20 2011).
My v223W from July 2007 is still going OK (strong would be an overstatement), though.
[Edit]
O snap, it's still under a warranty of some kind.
|
No, i dont know where that came from. Maybe Radeon cards have better driver support for sc2 so you can reach required GPU performance at lower price brackets - it just meant being CPU capped with a 7770 instead of needing a better GPU for his test. The slightly higher result for the 7970 would not be reflected in minimums and i think is just a weird outlying result that is probably wrong.
Maybe I'm thinking of another bench. I do recall benches showing that sc2 was much more nvidia favored though. But the 7xxx series was far from released back when that was done, I think 7xxx vs 6xx changed the dynamic a bit between the two as nvidia neglected mid-range and radeon (who also neglected midrange imo) really caught up for top tiers cards.
Nobody wants to run AA with sc2. The ingame method is awful, costs almost nothing to run but is a post processing effect that blurs textures and health/energy bars among a few other things - it's not at all an improvement of image quality - Out of game forced AA? Well good luck with that, you blur the hell out of the UI and minimap as well as everything else. I was unable to find a usable method.
You know there was a tomshardware article showing before/afters of all the forced nvidia settings in sc2, stuff like AA, ambient occlusion, and a lot of it makes a pretty big difference (as big as insanely intensive, minor artificial improvements can be). Ambient occlusion really stuck out as having a big impact.
There will be moments of the game that are more CPU inteisve, and moments that are more GPU intensive. It's not always just "well sc2 is bottlenecked by the cpu so your good to go with a weak gpu". A gpu can increase or decrease fps too.
But that's a i5-3570k in the screenshot, no HT possible. You can also see it's 4 threads in the status bar. I tried setting the same settings for memory in LinX and get something like 125 at 4.6 GHz, finishing a few seconds faster with a run.
It's still interesting: the CPU works correctly with low voltage but does not perform like you'd expect for 5.2 GHz. Perhaps it has something that keeps it slow and stable. But that would be weird, it has the exact same stepping, revision, etc. as my CPU.
Gflops are useless, they are not a benchmark at all. A lot of stuff affects your gflop score a lot more than your cpu overclock, too. Like his ram is a lot slower than most people's ram who are doing 5+ ghz overclocks.
At 5ghz, HT on, and 2400mhz CL8 RAM my gflops is 115.
But what actually would cause a performance difference like that, to get a significantly higher clock but clearly worse benchmark time?
Gflops is not a benchmark!
Actually, what it looks like are those dips below 100% CPU utilization on the graph might be causing lower performance, but why are those happening? I wouldn't think that a power or temp (CPU or VRM) limit would be hit, considering the voltage and how it's running on a Sabertooth... It's either something like the above, something I didn't think about, or a 'shop of the CPU clock on the CPU-Z window.
it's because linx is a terrible measure of both stability and benching (it's decent for temp testing, that's it) and when it loads up the next test of the ~7 it does, your system and cpu load drops (and temps) drops momentarily.
If you're running the same code on the same processor with same processor and RAM speeds, you should get the same results.
It's that gflops is affected more by nonsensical stuff, like having lower scores when certain components are faster and higher scores by slowing certain things down.
I can also confirm that the crazy i5's that can do over 5ghz on around 1.2v is true though.
|
Oh okay thanks for corrections. I never really paid attention to dumb FLOPs measurements. I guess that these CPUs are so complicated that there can be a lot of unexpected interactions with respect to the timing of cache hits and misses, memory access, branch prediction, etc.
Though sadly enough, FLOPs are primarily how even supercomptuers are ranked...
It's good enough for order-of-magnitude comparisons, with an asterisk maybe.
|
I would think it's fine for supercomputers given that's pretty much all they're there for.
|
Oh okay thanks for corrections. I never really paid attention to dumb FLOPs measurements. I guess that these CPUs are so complicated that there can be a lot of unexpected interactions with respect to the timing of cache hits and misses, memory access, branch prediction, etc.
Though sadly enough, FLOPs are primarily how even supercomptuers are ranked...
Gigaflops was the benchmark for CPU power like 15 years ago. Now with computers with more than one core!!!, instruction set optimization (brand new AVX!), and the fact it relies only on floating points (hence why HT hurts the score), gigaflops became outdated a while ago.
Supercomputers aren't exactly benched on gigaflops, they are generally calculated, theoretical numbers, that the system will not even come close to in the real world.
|
United Kingdom20323 Posts
It's not always just "well sc2 is bottlenecked by the cpu so your good to go with a weak gpu". A gpu can increase or decrease fps too.
Not your minimum's once you have an adequate one for your CPU, as shown by 7770 and 7950 having identical FPS at 1920x1080 max settings with nothing forced out of game, even in a situation where FPS was more than double worst case 1v1 minimums which would differentiate the GPU's a lot more
I can also confirm that the crazy i5's that can do over 5ghz on around 1.2v is true though
Its not just i5's.
|
On March 20 2013 13:14 Craton wrote:
I would think it's fine for supercomputers given that's pretty much all they're there for.
Some people are running a lot of integer code on those, not floating-point. And if they're running floating-point, it's probably more like double-precision. (okay, double-precision floating-point instructions / second is often just pretty much half of single-precision floating-point instructions / second, but not always)
Also, on a lot of workloads it's actually not really the CPU compute power that's the bottleneck, but all the network and backbone interconnects, data access, etc. It's fine to have a bunch of cores, but what if cores on one mobo on one rack needs to access data in memory in another rack? When you benchmark, these issues might not cause bottlenecks and you can keep most of the cores loaded most of the time, but they could potentially be significant under real usage.
And if you use FLOPs as a comparison, you end up with GPU-heavy clusters having a big advantage over x86-based clusters because they get a lot higher FLOPS / $ and FLOPS / W and FLOPS / rack (physical space). It's not like you can just run generic x86 code on GPUs, at least not yet. Hence the need for an asterisk IMHO, and more.
On March 20 2013 13:21 Belial88 wrote:Show nested quote +Oh okay thanks for corrections. I never really paid attention to dumb FLOPs measurements. I guess that these CPUs are so complicated that there can be a lot of unexpected interactions with respect to the timing of cache hits and misses, memory access, branch prediction, etc.
Though sadly enough, FLOPs are primarily how even supercomptuers are ranked... Gigaflops was the benchmark for CPU power like 15 years ago. Now with computers with more than one core!!!, instruction set optimization (brand new AVX!), and the fact it relies only on floating points (hence why HT hurts the score), gigaflops became outdated a while ago. Supercomputers aren't exactly benched on gigaflops, they are generally calculated, theoretical numbers, that the system will not even come close to in the real world.
Refer to this often-cited list:
http://www.top500.org/lists/2012/11/
and whoamg it's Titan on top. errrr... Titan the supercomputer, running lots of K20X based on those GK110 chips Nvidia now sells in graphics cards called Titan.
edit: trust me, they actually benchmark these things for those figures...when the one I use is taken down every once in a while for a week or two for repairs / upgrades / benching, it's not just to piss me off
|
|
|
Friend's working on a build. Has optical drive/monitor/KB/mouse. Don't think he needs a HDD/OS. $600 budget. Plays things like SC2, Skyrim, LoL. Will be buying everything from Newegg (not all items on list have store set correctly). Probably will do a medium (simple) OC. I'd probably have him swap RAM to http://www.newegg.com/Product/Product.aspx?Item=N82E16820231445 or something similar rather than a 9-9-9-24-2N.
http://pcpartpicker.com/p/KVWf (He picked most of these)
GPU not included in this budget. Leaning toward a second-hand 5850.
Thoughts?
|
Some user comments for the Z77A-G41 claim 4.5 GHz OC on i5-3570k (stability tested? who knows the voltage or how good that chip used is), but I really wouldn't try skimping that hard, seriously. If you're going to use low-grade MOSFETs, at least put heatsinks on them. x_x
If you can stretch a little extra money, get Rosewill Capstone 450 instead of Corsair CX. It's a leap from budget-tier through decent-tier all the way to high-tier for say $5-15 depending on how you count it
He really wants that Phantom 410?
|
Yeah it seems like it. I linked him a bunch of cheaper alternatives with similar features / size and he kept going back to the Phantom.
Sounds like you're iffy on the mobo? What would you recommend then?
|
Oh yeah and RAM timings really don't seem to do that much for most applications on modern CPU architectures like Ivy Bridge, even less than the RAM clock speed. I wouldn't really pay extra there and pay less on CPU cooling and the motherboard.
If you're serious about going cheap and really not pushing the OC much at all, I'd say AsRock Z75 Pro3 for $80 is probably better. Actually, there's a decent price gap between that and something significantly better, so it might be the right price point to slot in at. It's cheap for a reason though. Maybe somebody else has a better idea. btw the difference between Z75 and Z77 is not supporting Intel's SSD caching technology that people don't actually use.
|
Unless you can find the MSI Z77A-G41 for under $20 like I did, do not buy it. It's a terrible motherboard. Good luck hitting even 4.5ghz on a board with VRMs that bad, you'll motherboard will limit your overclock more than your CPU or heatsink. Unless you are familiar with working with crap motherboards and have a digital multimeter and discrete thermal diodes and some thermal tape and are comfortable attaching sawed on heatsinks like I am, to monitor the VRM and voltage, you should probably stay away from the g41. Get this motherboard instead, GA-Z77X-D3H:
http://www.newegg.com/Product/Product.aspx?Item=N82E16813128546R
SLI capable, about 10 times the build quality (doesn't feel like a piece of cardboard, you can just look at the g41 and see how bare it is, because the sound chip has less capacitors and controllers on it, the pci express lanes have less capacitors and chips on it, the usb has less chips on it, the sata ports have less chips on it, the vdimm has much worse voltage regulation. etc, etc, etc), an extremely capable VRM that'll last a long time and allow for extreme overclocks, a board that won't blow out on you because it's so bad, better circuitry protections... So worth the extra $10 (this d3h is actually on a really good sale right now).
oh it's open box, well if you want it brand new it's only $20 more. I would recommend the open box given newegg's 30 day return policy is quite quick, and the fact that gigabyte honors warranty even on used products.
Rosewill capstone at $60+ :O
I'm not sure on the quality of it myrm, how good is it? You can get the XFX Pro 450w and pc power and cooling 400w semi-modular, both seasonics with chemicons for only $49 (i just bought a used xfx pro for $45 but v1, not v2, and used for a few hours and unregistered so it'll still be covered under warranty). cx430 at $19 really is a good deal, if you are skimping that badly I think the cx430 for $19 (as opposed to $60 for a capstone or even $50 for xfx/pc p&p) is the right place to cut your budget. And coming from someone who's had 2 CX's blow out lol.
Does guy have access to microcenter? Phantom 410 is only $60 there i think, insanely cheap. Even at $99 it's a great case, imo the best case for $100. Neweg has it on sale at $79 which is a great price for it still. If you don't live by microcenter, you can actually buy the Phantom 410 for $60 at best buy because they price match microcenter (there have been reports of managers being jerks if no microcenter is close by, but mostly success stories, just call ahead).
Poor choice in RAM, just get the cheapest 2x2GB of RAM (gaming, general usage will never use more than 4gb of ram, and you clearly are on budget, although there's really no reason to get 8gb of ram, you just won't ever use more than 4 for gaming/general usage no matter how many tabs you got). The ram you picked is overpriced and slow, I'm pretty sure they are just some very low binned hynix (that or micron d9's). Right now prices are a bit jacked on newegg, I think you should wait a few days or consider buying your RAM elsewhere for the moment (im sure tommorow newegg will have the right price on it somewhere though, but definitely shouldn't pay more than $35 for 2x4gb or $25 for 2x2gb of ram if you are getting slow junk ram).
Don't get the hyper 212 evo, it's a rip-off. Get the 212+ instead. If you want additional cooling power, you can buy high end thermal paste under $5 and a 2nd fan for $5, and it'll outperform the 212 EVO by more than the evo outperforms the 212+ at stock. Also the fan isn't nearly as loud as the evo's (or as powerful, both are terrible fans really but evo is cranked up even more to compensate for a cooler that isn't really any better by itself). I'd really recommend against either hyper 212s for overclocking an ivy bridge but it'll do fine for a budget build, you can always get a better cooler later given how cheap they are anyways.
You should definitely get a solid state drive. I'd recommend the 128GB Samsung 830 at $85 on amazon oh newegg, right. well, http://www.newegg.com/Product/Product.aspx?Item=20-171-646&IsVirtualParent=1 $85 for a sandisk 128gb, what it lacks in speed, it makes up for in reliability, i use these all the time. Samsung 830 is a lot faster but if you say newegg only, you the boss. The extreme version is faster for only $10 more but not necessary (not as familiar with the extreme).
There's also this one for the same price: http://www.newegg.com/Product/Product.aspx?Item=N82E16820329003 Hynix. Seems to be a faster drive, same sandforce controller, but I don't know much about this ssd and couldnt find anything on it.
Second hand 5850, way to go! Smart shopper!
|
|
|
k, well i mean i wrote everything under the assumption of no microcenter. You are still near best buy and even if you aren't newegg still has that phantom at a great price.
|
On March 20 2013 16:03 Belial88 wrote:
Rosewill capstone at $60+ :O
I'm not sure on the quality of it myrm, how good is it? You can get the XFX Pro 450w and pc power and cooling 400w semi-modular, both seasonics with chemicons for only $49 (i just bought a used xfx pro for $45 but v1, not v2, and used for a few hours and unregistered so it'll still be covered under warranty). cx430 at $19 really is a good deal, if you are skimping that badly I think the cx430 for $19 (as opposed to $60 for a capstone or even $50 for xfx/pc p&p) is the right place to cut your budget. And coming from someone who's had 2 CX's blow out lol.
You see NZXT Hale90 selling for $100+? Kingwin Lazer Gold selling for $100+? There are other examples, including Rosewill's own Lightning and Sentey Golden Steel. Capstone is the same design, same internals, but it costs way less due to newegg house brand pricing magic.
If you want to take a Corsair-centric view of the world, it's say between modern TX and HX (maybe more like HX) in quality. It's way more advanced than Seasonic S12II, which is what the XFX Pro and PCP&P Silencer MKIII (lower-wattage) use, similar to Seasonic G. It's obviously more efficient than Seasonic S12II, quieter with 140mm ball-bearing fan at lower speeds, full NCCs if you're brand-conscious on capacitors, 5 years warranty.
edit: wait a sec I once compiled a list of non-cherry-picked English language technical reviews for Super Flower Golden Green power supplies. Let me find it, just for informative purposes for future reference (I know you probably believe me already lol, no need for me to make a point)
+ Show Spoiler [list] +
If you're counting MIR than CX430 for $19 is definitely a great value though. I never count MIR pricing towards value for price/performance, even though usually you can eventually get the credit. You could argue otherwise.
|
If you're counting MIR than CX430 for $19 is definitely a great value though. I never count MIR pricing towards value for price/performance, even though usually you can eventually get the credit. You could argue otherwise.
Hm i think it depends on the company, and corsair is known for the best support. I got all 4 of my rebates back within 2 weeks from them. I think it's definitely worth something, whether you think it's worth full value or not, and I mean saying the cx430 is $38 instead of $39 is a decent deal for a psu that's definitely not in the same level of quality as the capstone, but it's sufficient and shouldn't cause any trouble with such a build, or even much higher end ones.
psu's aren't so much my expertise, seems like the motherboard is a bigger factor and when the psu you are getting is worth more than the mobo :X
Interesting what you say about the capstone. Didn't know that. Just bought an xfx but $64, bleh, i'm not really comfortable with that much in a psu for single-gpu build. I kinda feel that's a more of a luxury, definitely a great place to put a bit of money if you got it, but sounds like this particular build is trying to shave all the corners and I think the cx430 is a good place to shave a good $20+ compared to the capstone. Whereas he really can't shave money anywhere else in his build (oh wait his ram, lol, yea if he can get 2x2 than he could fit that psu in no problem).
|
Mobo is a much bigger deal for usability, reliability to some extent, if you're making use of the extra features (which most people asking, will not), but I figure people can use good power supplies for ~7 years or longer. Though with respect to reliability, something that's open box may be more likely to have issues...
You might have to (or want to) replace a CX or something like that, at least once in that kind of time period. Also, the ~5% efficiency difference is small and often overstated, but it'll add up to something by several years, closing the price gap. Also, CX has high incidence of coil whine it seems? But if you're counting savings over time, it makes sense to count rebate savings too, unless you're assuming laziness.
Looks like from here on out, you need to toss a mobo pretty much every time you get a new CPU, so that doesn't really have the same staying power.
edit: doesn't seem like corners are being that shaved if going with Phantom 410? depends on perspective.
|
|
|
for CPU
the FX-8350 price for the performance is so enticing. 8 cores 8 threads for $190.
Would some of you choose intel's cores that perform similarly for a price premium (pay more)?
If so, why? which ones?
|
|
|
|
|
|
EPT
