CPU gaming benchmarks: i7-860 to i7-3770k clock-for-clock and OC

[Termie]

Thought some people might be interested in seeing what the potential improvement in gaming is going from an older i7 (the 860) to a new i7 processor (the 3770k). I'm only going to do gaming benchmarks, because that's what I care about and that's all I did on my older i7. I'm doing both clock-for-clock and some moderate OC tests. You can extrapolate from those other levels of OC. The graphics card is an EVGA GTX670 FTW @1215/6500 in all tests.

The settings
(1) i7-860: 159 BLCK with Turbo, translating to 3.5GHz with all cores loaded, or about 4.1GHz in a single thread. HT on.
(2) i7-3770k: either 3.5GHz on all four cores or 4.4GHz on all four cores. Turbo is disabled, HT is on.


Actually, I'll start with a few simple CPU benchmarks I do have:

SuperPi 1M

(1) i7-860 @4.1 on a single core: 10.3s
(2) i7-3770k @3.5: 10.5s
(3) i7-3770k @4.1: 9.0s
(4) i7-3770k @4.4: 8.3s

Clock-for-clock improvement = 14%
OC-for-OC improvement = 24%

IntelBurnTest

This was something I had but had never used as a benchmark, just as a stress test. I'll go ahead and add it for what it's worth.

(1) i7-860@3.5:

I don't have sufficient cooling to run this test on my HTPC. Within seconds it shot to 98C and started throttling. It was running at near 3.5GHz for the few seconds the test ran, however, so this is close enough to compare to the 3770k, and it's consistent with my recollection of prior runs on this CPU when it was in my gaming PC.

(2) i7-3770k@3.5:

(3) i7-3770k@4.4:

Wow, the GFlops results appear to scale entirely linearly with clock speed on the 3770k. Interesting. And I don't know why, but the 3770 is nearly 90% faster clock-for-clock.

 

[Termie]

Now on to a few GPU benchmarks:

3dMark11

(1) i7-860@3.5:

(2) i7-3770k@3.5:

(3) i7-3770k@4.4:

Clock-for-clock improvement in graphics-only score = 2.2%
OC-for-OC improvement in graphics-only score = 2.1%

As you can see, the change in CPU matters very little here, and once you jump to the 3770k, the OC doesn't even register. My guess is that the improvement is almost entirely due to the switch to PCIe 3.0.

Clock-for-clock improvement in CPU physics score = 18.9%
OC-for-OC improvement in CPU physics score = 41.3%

Nearly 20% clock-to-clock improvement is about what I'd expect in theory.


Unigine Heaven 3.0

(1) i7-860@3.5:

(2) i7-3770k@3.5:

(3) i7-3770k@4.4:

I figured this one would be totally GPU-limited. Well, something's giving me a boost of about 4-5%, probably a combination of PCIe 3.0 and the new CPU.


BF3 benchmarks

Swordbreaker Single-Player 60-sec run - All Ultra/No MSAA

Not much of a difference here, particularly considering the inherent margin of error in a real run as opposed to a benchmark. 3% clock-for-clock, 4% with the OC, but most of this is probably the PCIe difference.

Caspian Border Multiplayer 60-sec run - All Ultra/No MSAA

Take these numbers with a grain of salt - it is incredibly hard to get consistent runs out of BF3 multiplayer. In fact, that run on the 860 really isn't representative of what I typically saw, which was more like 67fps. Either way, the minimums are much, much higher, and yes, the average is much higher as well, and if I had to put a percentage on it, I'd say it's somewhere between 10% and 20% clock-for-clock. Suffice it to say, you can see the trend, and it shows that the CPU makes a difference with this GPU. In fact, I'd conclude that the stock 3770k is bottlenecking the 670, whereas at 4.4, the 670 finally becomes the bottleneck. This would tend to support the other threads in this forum showing bottlenecking of 670SLI on the 3770 and 3930 at 4.4. That near-doubling of GPU power simply overwhelms any CPU on the market today.

Borderlands

This test is a bit different due to the in-game framecap, but it's a run I know well, and I was always surprised that it brought my 670 to its knees. Well, it turns out this run was entirely CPU-limited. In fact, the GPU is throttling back to non-boost speeds due to the low GPU load. The average is slightly different, but look at the minimums - they show an 11% improvement clock-for-clock, and a huge 38% improvement OC'd.

 

[Makaveli]

Additional benchmarks to follow...

Sub'd!

And the 20% improvement sounds about right going from lynnfield/Nehalem IPC to Ivy.

If Haswell adds another 10% IPC improvement that will be a no brainer for me and my next upgrade.

 

[Rvenger]

Hey, you switched to an IVB i7!

 

[minitron]

Disable turbo and run the benches with all cores of the 860 at 4.0 Ghz to test IPC (14% according to sPi). Alternatively disable turbo and run the 3770 at 3.5 Ghz.

860 @ 3.5 vs 3770 @ 4.4 sounds reasonable to test 25% overclocks on both systems though its unnecessary if you figure out IPC and do some math.

Gaming benchmarks will be interesting though that 860 could definitely be clocked higher.

 

[Termie]

Disable turbo and run the benches with all cores of the 860 at 4.0 Ghz to test IPC (14% according to sPi). Alternatively disable turbo and run the 3770 at 3.5 Ghz.

860 @ 3.5 vs 3770 @ 4.4 sounds reasonable to test 25% overclocks on both systems though its unnecessary if you figure out IPC and do some math.

Gaming benchmarks will be interesting though that 860 could definitely be clocked higher.

I'll update those SuperPi benchmarks tonight or this weekend, running both at 3.5 for IPC (I've already given you 4.1, however, but Turbo does throw it off a bit). I still have the 860 to do CPU-only benchmarks, but can't run it with the GTX670 anymore.

And yes, both CPUs could be clocked higher in fact, but I'm showing similar overclocks on each (~20% actually, factoring in automatic Turbos at stock: 3.5 vs. 2.93 on the 860, 4.4 vs. 3.6 on the 3770k). I don't like pushing my equipment too hard.

 

[IntelEnthusiast]

I have been saying going from a 1st generation to a 2nd generation Intel® Core™ processor gives around 10% to 15% improvement and then going from a 2nd generation to a 3rd generation Intel Core processor would yeild around 6%; based on your tests that looks to be right on. Thanks for our effort on these tests.

 

[Termie]

Hey, you switched to an IVB i7!

What can I say, I've been following your moves carefully...

I have been saying going from a 1st generation to a 2nd generation Intel® Core™ processor gives around 10% to 15% improvement and then going from a 2nd generation to a 3rd generation Intel Core processor would yeild around 6%; based on your tests that looks to be right on. Thanks for our effort on these tests.

Thanks - happy to help out. Those estimates you provide seem very accurate indeed.

 

[Termie]

Benchmarks added and reorganized.

 

[Lepton87]

The huge difference in IBT is due to AVX. In practice almost no applications use it, so if you want apples to apples comparison run IBT on 3770 without AVX. (without SP1)

 

[Makaveli]

The huge difference in IBT is due to AVX. In practice almost no applications use it, so if you want apples to apples comparison run IBT on 3770 without AVX. (without SP1)

I don't think its just AVX there is a 20% IPC gain from 1st gen i7 to 3rd gen.

But I also agree with that other guy take turbo off and try to test at the same clock speed.

 

[Termie]

The huge difference in IBT is due to AVX. In practice almost no applications use it, so if you want apples to apples comparison run IBT on 3770 without AVX. (without SP1)

Thanks, that makes sense.

I don't think its just AVX there is a 20% IPC gain from 1st gen i7 to 3rd gen.

But I also agree with that other guy take turbo off and try to test at the same clock speed.

Actually, turbo doesn't even come into play in these benchmarks. IBT absolutely locks the CPU at 3.5. Also, I remembered that my 1156 board has a glitch where if you run it at the highest multiplier, turbo is always on, even when it's off in the bios. This prevented me from getting higher overclocks.

 

[Ferzerp]

Your linpack scores are totally too low because you're allowing it to use 8 threads. 4 produces peak performance.

HT is a penalty, not a benefit in linpack, at least on the AVX capable procs. I don't know on the nehalem/westmere.

 

[Termie]

Your linpack scores are totally too low because you're allowing it to use 8 threads. 4 produces peak performance.

HT is a penalty, not a benefit in linpack, at least on the AVX capable procs. I don't know on the nehalem/westmere.

Well, what do you know - you're right! Like I said above, I'd never benched with this program, only stress-tested, but here are some results I came up with while having some undervolting fun:


i7-3770k @ 3.9GHz/1.0v, 8 threads:

i7-3770k @ 3.9GHz/1.0v, 4 threads:

Anandtech was able to get their 3770k to run at 3.9GHz/0.9v, and I've already got it down to 0.95v, but I think this will be my daily setting. Mmmm, I loves me some 1v action!

 

[Termie]

Ok, last gaming benchmarks added, Heaven 3.0.

From the tests I've done, I think it's safe to say that when GPU-limited, games show between 2 and 5% improvement just due to the new platform. When the CPU is a major factor (like BF3 multiplayer), the difference is closer to 10-20% clock-for-clock. And with the good overclocking potential of IVB, there's plenty more potential there. Comparing both my CPUs overclocked by 20%, my IVB system provides around a 25% improvement in gaming (and the GPU becomes the limiting factor).

As I noted in my previous post, I'm able to run 3.9GHz at 1.0v, and that's the happy medium I'm going to go with most of the time, as it will likely eliminate any CPU limitation with a single GPU, while running incredibly cool and efficiently (~60C and 110w in IBT).

 

[Makaveli]

Your linpack scores are totally too low because you're allowing it to use 8 threads. 4 produces peak performance.

HT is a penalty, not a benefit in linpack, at least on the AVX capable procs. I don't know on the nehalem/westmere.

Its the same on nehalem/westmere.

HT on


HT off

 

[Deleted member 4644]

This is a great thread. It makes me feel pretty good that my 860 @ 3.8 is good enough for a while longer. I think the one thing I am missing is 6gbs ports on my mobo.

 

[pantsaregood]

Your SuperPi time and score in LinPack bother me.

My 2500K at 4.4 GHz finishes SuperPi in 8.57 seconds with loads of programs running. Without them, I'm pretty sure it falls in the range of 8 seconds. It also manages 121 GFLOPS in IBT, though that may be a result of AVX.

EDIT: AVX doesn't explain it. Clearly the 3770K is using AVX, but it doesn't make up for the fact that a a 3770K is running at 20 GFLOPS less than a 2500K at the same clock speed. HT disabled may bring the CPU up to ~110 GFLOPs, but that still doesn't explain the loss of power.

Are you sure your CPU is completely stable? IBT is good at spotting "invisible" instability because it gives you some sense of "my PC isn't crashing, but it isn't performing as fast as it should." I can actually drop the voltage on my CPU by about 0.03v and never see a crash or error, but it seems that error checking is to thank for that - it drops to around 110 GFLOPs at that voltage.

 

[Owls]

I have an i7 920 @ 4.11Ghz and I STILL (unfortunately) see no need to upgrade to any of the new CPUs. Intel has either made the old i7s really good or they are purposely holding back the new CPUs.

This is the longest I've ever had a CPU and I bought the i7 920 in Nov 2008 when it was released. Depending on when Haswell gets released I could probably looking at almost 5 years on the same CPU!

 

[Gikaseixas]

Same here. My 860 is "good enough" for all my needs

 

[bunnyfubbles]

I have an i7 920 @ 4.11Ghz and I STILL (unfortunately) see no need to upgrade to any of the new CPUs. Intel has either made the old i7s really good or they are purposely holding back the new CPUs.

This is the longest I've ever had a CPU and I bought the i7 920 in Nov 2008 when it was released. Depending on when Haswell gets released I could probably looking at almost 5 years on the same CPU!

nope, the original i7s really were amazing CPUs, could easily go down as one of the all-time greats

heck, when overclocked they're still effectively just as good if not better than anything AMD has to offer

the flip side to that would be AMD's Cypress GPU that came out this time of year back in 2009. Someone back then could have had a 4+GHz i7 with a pair of 5870 @ 1+GHz and that rig would still be very potent to this day.

 

[beginner99]

nope, the original i7s really were amazing CPUs, could easily go down as one of the all-time greats

heck, when overclocked they're still effectively just as good if not better than anything AMD has to offer

the flip side to that would be AMD's Cypress GPU that came out this time of year back in 2009. Someone back then could have had a 4+GHz i7 with a pair of 5870 @ 1+GHz and that rig would still be very potent to this day.

I have a i7-870 and HD 5850 since over 2.5 years and feel no need to upgrade.
(mainly due to having a 1680x1050 lcd).

 

[jerkan]

test

 

[darckhart]

yup i still rock the 920. was tempted by 2600k, tempted by 3570k, but held out. here's looking at haswell..

 

[RussianSensation]

15% IPC increase from Nehalem/Lynnfield to SB sounds right to me. This has been observed by 2 major websites in a wide variety of tests:

Computerbase: Final Performance Rating
Intel Core i5-760, 4C/4T, 2.80 GHz, 45 nm = 100%
Intel Core i5-2500K, 4C/4T, @ 2.80 GHz, 32 nm = 115%

and

Intel Core i7-870, 4C/8T, @ 2.80 GHz, 45 nm, SMT = 100%
Intel Core i7-2600K, 4C/8T, @ 2.80 GHz, 32 nm, SMT = 116%

The second website that arrived at similar results of Lynnfield vs. SB is iXBT Labs:

Core i5 760 vs. Core i5 2500K at the same clocks: 14%
http://ixbtlabs.com/articles3/cpu/sandybridge-core-vs-lynnfield-p2.html

So we are looking at 14-16% faster IPC from Lynnfield to SB depending on the source, average of 15%.

IVB was only about 3-4% faster over SB:
Intel Core i7-2700K, 4C/4T, 3.50 GHz, 32 nm = 100%
Intel Core i7-3770K, 4C/4T, 3.50 GHz, 22 nm = 103%
http://www.computerbase.de/artikel/prozessoren/2012/test-intel-ivy-bridge/14/

That means a 4.4ghz 3770K should be ~ 4.4 ghz x 1.15 x 1.03 = 5.21 Ghz Core i7 860 (or about 49% faster theoretically).

We are only seeing 41.3% in Physics 3dMark11 score though.

Considering an i7 920 could clock to 3.8-4.2ghz in November 2008 and cost $284, this type of performance progress in the CPU space is nothing short of disappointing given the time frame. That's why I pretty much stopped caring about CPUs and have shifted to GPUs.

That means since 2008, Intel only increased IPC by about 18.5% on average and most of the performance gains instead came from frequency increases via node shrinks. I doubt Haswell will change this and upgrading to it will be more for fun than anything. The main draw will likely be hitting 5.0ghz overclocks due to Intel moving back to fluxless solder, allowing us to take full advantage of the more mature 22nm node. On the positive side this means a gamer can now buy an Intel CPU and overclock it and probably feel safe that in 4-5 years it will still be good enough for games like i7 920 and i7 860 @ 3.8-4.0ghz are with single high-end GPUs.

For anyone who doesn't realize how important it was for AMD to have competitive CPUs, Intel has been shrinking die size and compromising CPU performance by not increase the # of cores at the $300 price level and instead putting in useless GPU performance for enthusiasts:

If someone told me in the summer of 2007 that we'd still be using quad-core $320 CPUs by mid-2013, I wouldn't have believed that person. Unfortunately we are about to have this exact situation play out with Haswell , all because AMD screwed up Phenom and Bulldozer.