Performance upgrade?

[Corrigan]

I'm considering getting a 2500K and overclocking it to 4.5 GHz. How much of an improvement do you think I'd see in DCS A10-C, based on the system in my sig? Anyone who's made a similar upgrade?

 

Thanks!

[EtherealN]

Hard to say without knowing the exact model of i5 you have currently. There is a LOT of them and their performance can be quite varied.

[Corrigan]

Surely the frequency is enough of a measure? Or are different architectures really that varying in efficiency?

 

Anyway, it's an i5 750 (Lynnfield). Brought up to 3.80 GHz from 2.67 stock.

Edited May 16, 2012 by Corrigan

[metalnwood]

I needed another machine a while back so I got an i5 2500k and clocked it at 4.5. As it was faster I decided to use it for my main machine and put the old one to use in the arcade cabinet that I got the i5 for.

 

My i7 was a 920 clocked at 3.8. I found the i5 @4.5 a good improvement over the i7. I didn't quantify it and remember being happy about doing it.

 

So, if it is an old i5 it could be worth it, if your happy spending the $$

[EtherealN]

Surely the frequency is enough of a measure? Or are different architectures really that varying in efficiency?

 

Indeed they are very varying. :)

It's not really a question of "efficiency", just being different architectures. The clock speed is only a metronome that keeps things happening in synchronization - doesn't say anything really about how much actually happens. Also, different architectures may require a different amount of clock cycles for a given operation - what takes 10 clock cycles to do on one architecture might take a single one on another. (Extreme example - it usually isn't that big of a difference, but you get the point. There are a lot of other things that matter as well, since on-die caches, memory controllers, instruction pipelines etcetera may work in completely different ways between differing architectures.)

 

Anyway, it's an i5 750 (Lynnfield). Brought up to 3.80 GHz from 2.67 stock.

 

You would most likely see a definitive performance increase if you replaced that system with an i5-2500k-based one, but considering that you would also have to replace the motherboard and, potentially, the RAM (the RAM -should- still work, but the sandies are recommended to use 1.5v DIMMs instead of the older standard 1.65v DDR3 DIMMs - when you look at a motherboard you should be able to see it's list of supported RAM modules though and check if the ones you have are on that list) I'm not sure whether I'd recommend you to do that upgrade. Unsure whether it would give you enough to make you feel it was worth the money outlay.

 

If I was you, I would probably hold out until the Haswell chips come around. The Ivy Bridge i5-3570K is another option though, but again you'd require a new motherboard and potentially new RAM. This one runs at 3.4GHz at stock, and while the Ivy doesn't overclock as far as the Sandies do, you should see this one run at 3.8 while playing DCS even without overclocking (thanks to the Turbo feature). You should also be able to get it up a little higher - 4.4 should be achievable I believe.

 

In both cases, do note that these are chips that are overclocked specifically with the multiplier, not the old method of OCing the bCLK as on your current chip. Also note that you need a P or Z series motherboard to be able to overclock, a Z-series is generally recommended.

[Corrigan]

Thanks for an informative and exhaustive reply! I did in fact go ahead and order a new Z68 mobo and the 2500k. I already have 1.5 V DDR3s.

 

Also, thanks metalnwood.

 

I'll report back here when I've built the system.

Edited May 16, 2012 by Corrigan

[galagamo]

I also ordered an I52500k along with an ASRock Z68 hoping to use my current ram with it 1.5v hurray! Just hopping its compatible didn't see it on the QVL but..

I have my current chip OC'd to 4ghz stable Uclk@200 V1.3 on air but it does get a little warm like 70c when stressed.

Hoping for a little more headroom, and a lot more FPS out of this upgrade.

Now I must wait:cry:

[Rhinox]

...I also ordered an I52500k along with an ASRock Z68...

...I have my current chip OC'd to 4ghz stable Uclk@200 V1.3 on air...

...Hoping for a little more headroom, and a lot more FPS out of this upgrade...:cry:

I'd be very surprised if you'd get more than +10% fps with this upgrade. From cpu-architecture point of view, there is not much difference between i7-930 and i5-2500k. I know, it is 45nm vs 32nm, Bloomfield vs SandyBridge, 8MB vs 6MB L3, but "power per MHz" is about the same.

 

With good air cooling you can run i5-2500k @ ~4.5GHz, that is ~10% higher cpu-frequency than your current i7-930 (4GHz). *If* cpu is bottleneck in your system, you get +10% more fps. Not much more...

[galagamo]

I'd be very surprised if you'd get more than +10% fps with this upgrade. From cpu-architecture point of view, there is not much difference between i7-930 and i5-2500k. I know, it is 45nm vs 32nm, Bloomfield vs SandyBridge, 8MB vs 6MB L3, but "power per MHz" is about the same.

 

With good air cooling you can run i5-2500k @ ~4.5GHz, that is ~10% higher cpu-frequency than your current i7-930 (4GHz). *If* cpu is bottleneck in your system, you get +10% more fps. Not much more...

 

+10% would be great! I'm not that optimistic though. I just want to get to 4ghz w/o increasing the Vcore or FSB, and having to worry about the excess heat. I'm mostly interested in the unlocked multiplier.

[EtherealN]

From cpu-architecture point of view, there is not much difference between i7-930 and i5-2500k. I know, it is 45nm vs 32nm, Bloomfield vs SandyBridge, 8MB vs 6MB L3, but "power per MHz" is about the same.

 

Stolen from: http://techreport.com/articles.x/20188/14

 

the i5-2500k turbos to about 12% higher clock than the i7-950 in that test (comparing the single-thread part, obviously, so full apply). If they are "about the same", we'd then expect the 2500K to do roughly 12% better in the singlethread test. Except it does 28% better. This I definitely a significant difference.

 

Add to this how easy it is to overclock it further than the i7-950 (and it's siblings, ofc), and I'd definitely expect more than 10% if overclocking is brought into the picture - even without hardcore overclocking. It gives both better performance per clock and higher clock.

 

In another way:

i7-950 gets 0,348 points per GHz.

i5-2500K gets 0,403 points per GHz.

In other words, we're seeing 16% better performance in that test, per clock.

 

Granted, this is a synthetic test so we cannot draw immediate conclusions about DCS performance from it, but it does give you pretty good indication that the Sandy actually is stronger "per clock", and if we then add a higher easily-achievable clock speed to that equation things get even better for the Sandy.

Edited May 21, 2012 by EtherealN

[Rhinox]

It is not difficult to find another set of benchmarks with quite different results. And you need not to go too far. Look the next page where i7-950 (despite of lower nominal and turbo-frequency) is faster than i5-2500k in MyriMatch and Euler3D. Similar for some game-benchmarks (i.e. source-engine, metro2k33, etc.). Maybe bigger L3-cache (8MB i7-950 vs 6MB i5-2500k) plays role here, or tripple-channel (i7-9xx) versus dual-channel (i5-2xxx) memory controller...

 

I do agree that Sandy is a little stronger per clock, but not much. It is just evolution of previous architecture, where some parts have been even scaled down...

[EtherealN]

Note that Myrimatch and Euler3d are multithreaded. This pollutes the data beyond redemption if we want a per-clock metric that can be used in any way at all to indicate DCS performance.

 

Note for example how the Myrimatch results indicate that the i7-970 is significantly stronger than the i7-2600k, and even more so than the i5-2500k. However, this is because it has both more cores - 50% more cores in fact, and in the comparison with the i5 also offers HT. The data is clearly polluted as far as considering DCS performance.

 

If we compare the i7-950 with the i7-2600k (to remove HT and cache size from the equation) in the Myrimatch data , we get the following:

 

i7-950: 0,28, adjusted for clock frequency and number of cores.

i7-2600k: 0,23, adjusted for clock frequency and number of cores.

 

Where lower is better... ;)

In other words, in this test, the sandies show a 22% performance increase per clock. Non-trivial.

 

Further, in the Metro test, the differences are not statistically significant; something else is limiting it here. And in the source test, it's again a multithreaded test. Thus it shows the benefits of a processor compared to another in heavily multithreaded applications. As you know, this eliminates any deduction towards DCS performance, and pollutes other data. However, in the case of the source engine test, we can compare to the 2600K to again remove HT from the equation:

 

i7-2600K: 52,94

i7-950: 51,31

 

In this case, we do indeed see that the difference is minimal, per-clock. So applications will vary from a small lead for the Sandy, to a huge lead for the sandy, per-clock, PLUS that the sandies will easily overclock to clocks that are higher than the previous i7's will ever achieve without extreme measures.

Edited May 21, 2012 by EtherealN

[Corrigan]

The last point is important in my case. I have an i5 750 at 3.8 GHz, which is more or less as high as I can get it stably.

[EtherealN]

Aye, and considering the GPU you have, it likely is the CPU that is your most likely bottleneck. And with a good mobo, you should expect 4.4GHz to really be the minimum expected overclocking result - indeed, in my case (though that's a 2600K), 4.4 is where the mobos automated overclocker lands me. That's what I get when spending literally about 1 minute of effort to achieve the OC... If I spent more, I could tune it way better and get a much nicer clock for voltage ratio. (But since I usually don't utilize the OC, since it is a potential confounding factor in testing, I haven't bothered.)

 

In my case it's probably the GPU that bottles me at the settings I use (though it's still eminently playable at high settings), so I'm thinking about purchasing a 670.

Edited May 21, 2012 by EtherealN

[galagamo]

My first impressions of the i5 2500k over my previous CPU, as most of the reviews say, its substantially superior in most aspects, and its blatantly obvious to.