Is there anything out there yet that beats an overclocked 2500k?

[The Sauce]

I last built 4 years ago and have been running my 2500k @ 4.7-4.8 ever since. I am wondering if there is even anything out there yet that would beat it by more than 10% at base clock. MY main use is gaming so hyperthreading not really a big deal.

I already upgraded my video once on this system and I am considering just upgrading the gpu a second time. Seems crazy to me but it just seems that cpus are not really improving much in processing power over the past several years. I was planning on upgrading with Skylake, but now I'm not so sure.

 

[Majcric]

Probably not with the GPU in your sig. But yes Games are starting to utilize more than 4 cores.

I'm either going to do Skylake or seriously thinking about x99 platform.

 

[Yuriman]

Stock 3570K is ~10% faster when both are at stock.

4790K has maybe a 10-15% IPC advantage, and a 20% clockspeed advantage (approximately), which results in it being in the range of close to 40% faster at stock in single-threaded stuff, and the better part of twice as fast in programs that can fully utilize 8 threads, and some games are starting to. Of course, 4.7-4.8 is a very healthy overclock for Haswell, and most chips don't actually get that high without a lot of voltage. Haswell at ~4.4 should be around as fast, maybe a bit faster than your Sandy (when using only 4 threads), while consuming a lot less power.

Broadwell has an IPC advantage on Haswell, and Skylake has an IPC advantage on Broadwell. If I were to make an educated guess, I'm thinking Skylake will also hit around 4.4 with reasonable voltage, and will be around as fast as a ~5.2ghz Sandy Bridge chip, while consuming half or less power.

 

[toyota]

so you upgraded to the 670 or from the 670?

since your main focus is gaming then you really need a faster gpu if what you have now is the 670. you are at 1440 so you must not be playing demanding games if that 670 is still getting the job done.

and hyperthreading gets put to use in some games especially parts of Crysis 3 where I see nearly 90% cpu usage at times on my oced 4770k. of course I have a gpu thats nearly three times faster than the 670 though so I am not gpu limited.

so bottom line is upgrade your gpu if you want more performance then look towards a Skylake i7 to fully push it and cover your cpu needs for the next few years.

 

[DigDog]

nah man, you dont need an upgrade yet. wait for the next release.

 

[The Sauce]

I upgraded TO the 670 - about 2-3 years ago. That's overclocked quite a bit as well. Yeah, I game casually, but don't really beat the crap out of it. I may just go for the GPU upgrade...again.

Of course how long can this poor 2500k stand to be run at that overclock anyway? I've never clocked this high for this long - 4 years is unbelievable. I'm just waiting for the day of no-POST and possibly smoke coming from the damn thing. Even the mobo running that long clocked that high is amazing. I run pretty much 24/7.

 

[frozentundra123456]

4790k overclocks to around 4.5 I think, so maybe 5% slower than your 2500k, but ipc advantage is around 15%, so single threaded maybe 10% faster. Hyperthreading is starting to show 15 to 20+ percent gain in a few new games, so overclocked 4790k could be up to 30% faster depending on the game. I still question some of the results showing this much gain from hyperthreading though, and in some games, it does not help much at all. So basically, I dont think it is worth it for you to upgrade to another quad core. Now a haswell E if you can get a good overclock could be worth it in some newer games and supposedly in future games as DX12 makes use of more cores.

 

[toyota]

I upgraded TO the 670 - about 2-3 years ago. That's overclocked quite a bit as well. Yeah, I game casually, but don't really beat the crap out of it. I may just go for the GPU upgrade...again.

Of course how long can this poor 2500k stand to be run at that overclock anyway? I've never clocked this high for this long - 4 years is unbelievable. I'm just waiting for the day of no-POST and possibly smoke coming from the damn thing. Even the mobo running that long clocked that high is amazing. I run pretty much 24/7.

well why not knock it down a 100 to 200 mhz? you will never notice the difference anyway with a 670 at 2560x1440.

 

[jpiniero]

Of course how long can this poor 2500k stand to be run at that overclock anyway? I've never clocked this high for this long - 4 years is unbelievable. I'm just waiting for the day of no-POST and possibly smoke coming from the damn thing. Even the mobo running that long clocked that high is amazing. I run pretty much 24/7.

Unless you turned off EIST, it wouldn't be running that high 24/7.

 

[IntelUser2000]

Hyperthreading is starting to show 15 to 20+ percent gain in a few new games, so overclocked 4790k could be up to 30% faster depending on the game. I still question some of the results showing this much gain from hyperthreading though, and in some games, it does not help much at all.

Well, for one thing, Hyperthreading does wonders for dual cores.

I have seen benchmark of one of GTA versions, and it seemed well threaded, since it took advantage of every threads until 8 of them in the quad core chips.

Then you notice that in actual gameplay having too little threads make it really unplayable, or unusually choppy. Or that for the graphics quality, it doesn't get too many frames.

Perhaps Hyperthreading makes crap code do better?

 

[BonzaiDuck]

I upgraded TO the 670 - about 2-3 years ago. That's overclocked quite a bit as well. Yeah, I game casually, but don't really beat the crap out of it. I may just go for the GPU upgrade...again.

Of course how long can this poor 2500k stand to be run at that overclock anyway? I've never clocked this high for this long - 4 years is unbelievable. I'm just waiting for the day of no-POST and possibly smoke coming from the damn thing. Even the mobo running that long clocked that high is amazing. I run pretty much 24/7.

You may wonder, you may have reservations and concerns, but I don't.

It all depends on how you overclocked it, your voltage settings and whether or not you implemented the power-saving features.

I've found that to run either of my SB-K's at 4.7 with an offset V of +0.005V and "Extra Voiltage for Turbo" of 0.012V. LLC is set to third-highest setting or "High" -- leaving maybe 20 to 30 mV of droop. During severe stress tests, the drooped voltage is ~ 1.35V; in mild stress tests it may bounce between 1.36 to 1.37V. The voltage at unloaded turbo speed may show up between 1.38 to 1.39V. These voltages are right at the limit of the last published Nehalem "safe" spec for a 32nm processor, and SB-K is a 32nm processor.

Some folks -- our illustrious IDontCare among them -- think the processor is safe for higher voltages. Others may push their settings higher than 1.4+V. Whatever I might think, I don't know for sure. I simply chose not to push VCORE and VID (1.4 to 1.41V) any higher.

My 2600K has been running 24/7 overclocked for more than 4 years. Nothing has changed; it still shows itself stable for a 50-iteration affinitized LinX run with max. problem size and max. memory. And by "24/7" I mean that the system never sleeps or hibernates; I may reboot it once a week; the only times it has been shut down have been for dust-control, routine maintenance, and casual modding projects.

I actually expect these systems to have double their current lifespan. As for the motherboards -- I've got the original P8Z68-V Pro and the Pro/Gen3 like yours -- they could give out before the processors, which is not an uncommon thing. If your duty-cycle and phase-power settings are "Optimized" and "thermal balance," it will mean less stress on the board. I had to choose "current balance" over thermal. Also, it may be more stressful to set "PLL Overvoltage" to "Enabled." But I've heard nothing of anyone whose motherboard has gone south, and some of those folks use the "Extreme" duty and phase-power settings as well as PLL-Overvoltage.

They're hard to give up. I almost hesitate on going forward this year or next to build a new machine: it might not turn out as "perfect in every way" as these.

 

[moonbogg]

I turn all those motherboard power features to normal or off. My chip still works fine at 4.6 without any of that stuff. I think all those settings do is create more heat and run more power through the board without helping a damn thing. I just bumped the Vcore a little and that's it.
And yeah, Sandy Bridge won't die. They are as fast or nearly as fast as everything out there besides the new 8 core monster. I remember reading the 2600K review on Guru3D and he said it was the fastest CPU available and it was going to stay that way. Boy he didn't know just how right he was.

 

[frozentundra123456]

Well, for one thing, Hyperthreading does wonders for dual cores.

I have seen benchmark of one of GTA versions, and it seemed well threaded, since it took advantage of every threads until 8 of them in the quad core chips.

Then you notice that in actual gameplay having too little threads make it really unplayable, or unusually choppy. Or that for the graphics quality, it doesn't get too many frames.

Perhaps Hyperthreading makes crap code do better?

I agree for dual cores but I was referring to the quad i5 of the op.

 

[stockwiz]

I wouldn't worry about using a "lot less power" as an excuse to upgrade.. the 5-15 bucks per year in electricity won't be worth it, and if you live in a cold climate you merely turn the extra energy to heat for the home anyway.

Keep it until the next 6 core part comes out.

 

[Yuriman]

Comparatively, SB isn't bad, but when I moved from C2Q to Ivy in my main rig the temperature of my computer room dropped significantly. The air conditioner really couldn't keep up with the few hundred watts of added heat, so it was a matter of personal comfort more than money.

 

[VirtualLarry]

Comparatively, SB isn't bad, but when I moved from C2Q to Ivy in my main rig the temperature of my computer room dropped significantly. The air conditioner really couldn't keep up with the few hundred watts of added heat, so it was a matter of personal comfort more than money.

Same here, moving from a pair of Q9300 @ 3.0Ghz (400FSB) to a pair of G3258 @ 3.6 (36x multi, 100BCLK). I don't think that they are any faster, for DC, in MT, but they do run cooler and consume less power.

 

[BonzaiDuck]

I'm more concerned about taxing the hardware too much than for the power-saving. With EIST and C1E, power-consumption stays low until its needed.

I was looking at my 2700K + 2x GTX 970 system's APC UPS software today. It's showing between 95W and 105W at idle. There's "other stuff" plugged in to the APC, and I can't remember if the BenQ monitor is connected to it, or to another UPS providing power to the other two machines under my desk. Nor do I want to get down on all fours to refresh my memory in the dark, and its late. But for a good Seasonic PSU, I'd expect the actual power-draw of the rig at idle is more like 0.9 x 105 = . . . hmmm . . . ~95W.

But it sleeps for an hour, and then hibernates.

Reminds me. I'd posted a thread a month or so earlier, trolling for thoughts about how C3/C6 affects sleep states. I have now confirmed for my own benefit that enabling these features for high clocks will cause infrequent or occasional idle instability for Sandy Bridge. I'd solved that problem for one machine more than 18 months ago, but I'd turned off several features including the C3/C6 "report" items, so I was never sure with finality that C3/C6 caused the problem. This time around on the second system, it is confirmed.

As to sleep states. I think it only affects hybrid sleep. Let me say that more clearly: I'm absolutely sure of it. Basic sleep and hibernate aren't affected at all by disabling those features. Fur Chrissake! What other "sleep" do you really need besides that? [Yawn. It's getting late . . ]

Moral of the story: Don't enable s*** you don't need. If you want to enable something later -- one thing at a time -- you'll darn well know if it raises a problem, if all was working tip-top before that.

 

[Nhirlathothep]

u use a 670, start changing this thing and then we can talk about cpus




btw: i am with a old sandy-e and seems fine

 

[rumpleforeskin]

Even if OP paid out for new CPU/MOBO and gained 25% cpu performance I doubt he would see any meaningful FPS increase in game. Money would be much better put towards a better GPU in this situation.

 

[alcoholbob]

A 5820K/5930K/5960X will beat a heavily OC'ed 2500K in SLI situations. During regular gaming with a low/mid-end GPU, probably not.

 

[Yuriman]

A 5820K/5930K/5960X will beat a heavily OC'ed 2500K in SLI situations. During regular gaming with a low/mid-end GPU, probably not.

If you drop graphical settings to hit a target framerate, CPU differences will show up regardless of what GPU you have.

I'm more concerned about taxing the hardware too much than for the power-saving. With EIST and C1E, power-consumption stays low until its needed.

I was looking at my 2700K + 2x GTX 970 system's APC UPS software today. It's showing between 95W and 105W at idle. There's "other stuff" plugged in to the APC, and I can't remember if the BenQ monitor is connected to it, or to another UPS providing power to the other two machines under my desk. Nor do I want to get down on all fours to refresh my memory in the dark, and its late. But for a good Seasonic PSU, I'd expect the actual power-draw of the rig at idle is more like 0.9 x 105 = . . . hmmm . . . ~95W.

But it sleeps for an hour, and then hibernates.

Reminds me. I'd posted a thread a month or so earlier, trolling for thoughts about how C3/C6 affects sleep states. I have now confirmed for my own benefit that enabling these features for high clocks will cause infrequent or occasional idle instability for Sandy Bridge. I'd solved that problem for one machine more than 18 months ago, but I'd turned off several features including the C3/C6 "report" items, so I was never sure with finality that C3/C6 caused the problem. This time around on the second system, it is confirmed.

As to sleep states. I think it only affects hybrid sleep. Let me say that more clearly: I'm absolutely sure of it. Basic sleep and hibernate aren't affected at all by disabling those features. Fur Chrissake! What other "sleep" do you really need besides that? [Yawn. It's getting late . . ]

Moral of the story: Don't enable s*** you don't need. If you want to enable something later -- one thing at a time -- you'll darn well know if it raises a problem, if all was working tip-top before that.

There's quite a bit different between our two systems, but I've measured mine idling at ~35w at the wall. Pull the GPU and you can subtract 11w from that. My wife's Haswell i3 is lower still, and Skylake has two generations of power improvements upon that even.

 

[BonzaiDuck]

If you drop graphical settings to hit a target framerate, CPU differences will show up regardless of what GPU you have.





There's quite a bit different between our two systems, but I've measured mine idling at ~35w at the wall. Pull the GPU and you can subtract 11w from that. My wife's Haswell i3 is lower still, and Skylake has two generations of power improvements upon that even.

Now it's a new day, and I'm going to look under the desk to see where I hooked up the BenQ monitor . . . lemme-see . . .

Yeah! Not knowing the BenQ's power specs, I'd say "knock off another 25W." There are a few other devices connected to the APC, but they're minor for wattage.

Of course, as I said, that's when the machine is neither sleeping nor hibernating. And I'm guessing my assessment would be more accurate at 70 to 75W. As you said -- removing the second graphics card brings it to maybe 64W.

This comes a long way from middle of last decade, when my computer of that time was drawing closer to 190W at idle. Of course, if the system sleeps and hibernates on schedule, there's no constant power draw.

In other forum discussions recently, I was asking about the affect of certain BIOS features on sleep states. [Oh. I said that already. It's morning, but I need coffee.] Somebody -- might even have been you -- said something like "I dunno! I never tried it!" [Making the computer sleep and wake up, he meant.]

 

[The Sauce]

Well HardOCP review is out. A whopping 12% improvement per clock compared to Ivybridge (2500/2600k vs. 6600k). And it appears 4.7 is the cap. I'm shocked that they conclude that this is good enough to warrant an upgrade. I don't think so. Including RAM and Mobo I would guess ~$800 for a 12% upgrade?? That money is going into another video upgrade.

 

[StinkyPinky]

I wouldn't bother either with a 2500k clocked that high. 2500/2600k are awesome cpu's, I only replaced mine because my mobo died.

 

[cmdrdredd]

Well HardOCP review is out. A whopping 12% improvement per clock compared to Ivybridge (2500/2600k vs. 6600k). And it appears 4.7 is the cap. I'm shocked that they conclude that this is good enough to warrant an upgrade. I don't think so. Including RAM and Mobo I would guess ~$800 for a 12% upgrade?? That money is going into another video upgrade.

The main benefits are M.2 card support, USB 3.1, and more PCIe lanes to keep everything running at PCIe 3.0 speeds. That's probably why they say it's worth the upgrade. Extra features for future usage. It will depend on whether or not you are going to invest in devices that use those features though.

Me personally I'm eying an upgrade to move away from the mATX board I've been on since 2012. It's a great motherboard but it cramps my devices together and I get some higher than average temperatures from my 2nd 970 because it's starved for air. I can piece this system back together for some other usage at that point.