Intel "Haswell" Quad-Core CPU Benchmarked, Compared Clock-for-Clock with "Ivy Bridge"

[btarunr]

Russian tech publication OCLab.ru, which claims access to Intel's next-generation Core "Haswell" processor engineering-sample (and an LGA1150 8-series motherboard!), wasted no time in running a quick clock-for-clock performance comparison with the current Core "Ivy Bridge" processor. In its comparison, it set both chips to run at a fixed 2.80 GHz clock speed (by disabling Turbo Boost, C1E, and EIST), indicating that the ES OCLab is in possession of doesn't go beyond that frequency.

The two chips were put through SuperPi 1M, PiFast, and wPrime 32M. The Core "Haswell" chip is only marginally faster than Ivy Bridge, in fact slower in one test. In its next battery of tests, the reviewer stepped up iterations (load), putting the chips through single-threaded SuperPi 32M, and multi-threaded wPrime 1024M. While wPrime performance is nearly identical between the two chips, Haswell crunched SuperPi 32M about 3 percent quicker than Ivy Bridge. It's still to early to take a call on CPU performance percentage difference between the two architectures. Intel's Core "Haswell" processors launch in the first week of June.



View at TechPowerUp Main Site

 

[btarunr]

Many Thanks to NHKS for the tip.

 

[syeef]

Funny.

 

[dj-electric]

I don't believe in clock to clock benchmarks and find them pointless.
1. A CPU can be as fast as another C2C but the other comes naturally clocked higher
2. A CPU can be overclocked much further than another
So, of course a bugatti veiron at 60km/h would be as fast as fiat uno at 60km/h

For those who want examples, benchmark an i7 920 against 3770K C2C and see what im talking about.

 

[The Von Matrices]

I think we all expected Haswell to focus on graphics so a lack on improvement in x86 performance is not a surprise.

For those who want examples, benchmark an i7 920 against 3770K C2C and see what im talking about.

Does a 3770K really overclock much better than an i7-920? I agree that the 3770K uses a lot less power and generates a lot less heat, but the maximum 24/7 clocks (without extreme cooling) for both are still in the low ~4GHz range. The architectural differences account for a lot more of the performance difference between those chips than do the maximum clocks.

 

[...PACMAN...]

I don't believe in clock to clock benchmarks and find them pointless.
1. A CPU can be as fast as another C2C but the other comes naturally clocked higher
2. A CPU can be overclocked much further than another
So, of course a bugatti veiron at 60km/h would be as fast as fiat uno at 60km/h

For those who want examples, benchmark an i7 920 against 3770K C2C and see what im talking about.

You're wrong, clock to clock is the best way to benchmark and show differences in the architecture of two different chips at the same speed. It's all about IPC.

I understand where you are coming from with regards overclocking headroom but that normally forms part of a FULL review also.

Needless to say, my FX 4100@3.6 is miles behind a 2600K@3.6

 

[RejZoR]

I guess i'll be keeping my trusty Core i7 920 for another year or two. The chip was so good it's probably the longest owned single CPU in my systems ever. It's so long i don't even remember what year i bought it, which is unusual...

 

[dj-electric]

You're wrong, clock to clock is the best way to benchmark and show differences in the architecture of two different chips at the same speed. It's all about IPC.

And what if a certain cpu is as fast as the other C2C but at stock is clocked much much higher thus being faster? than, what's the point?

 

[repman244]

And what if a certain cpu is as fast as the other C2C but at stock is clocked much much higher thus being faster? than, what's the point?

The point is to see if the architecture had progress on the IPC field. Intel isn't changing it's design drastically (like going for a high clock low IPC), they have a certain pattern in the last few years.

A good example was when BD launched and was compared to the older gen per clock and you could see that the IPC decreased.

I agree that at the end it doesn't matter much since you can "hide" that with clock speed, but for architecture comparison it's the only way.

 

[dj-electric]

The point is to see if the architecture had progress on the IPC field. Intel isn't changing it's design drastically (like going for a high clock low IPC), they have a certain pattern in the last few years.

Progress could also be the ability to work at a higher frequency. That's all i'm saying.

 

[...PACMAN...]

And what if a certain cpu is as fast as the other C2C but at stock is clocked much much higher thus being faster? than, what's the point?

That's a different point entirely. Obviously if they are the same speed as each other at the same clocks, then the one clocked higher is faster. However, this is when other elements of the benchmark come into effect, i.e. does it clock higher? Has there been any power revisions or die shrink?

In the case of various phenom II revisions they could all pretty much be clocked to the same area of 3.8/4Ghz but effectively(as they were the same architecture) gave the same performance at the same clocks.

C2C is best used when it's between two different architecures.

 

[repman244]

Progress could also be the ability to work at a higher frequency. That's all i'm saying.

Indeed, but it's still interesting to see what happened to the IPC.

 

[DaJMasta]

I think it all comes down to the fact that they're testing engineering samples. Generally, these chips (especially early revisions) can't clock up to where the final product will be and come with dramatically reduced clocks as a result. Clock vs. clock comparisons are valid in comparing different architectures (even if they're very much the same, as we see here), but in this case they may be all we'll see until we hear word of launch pricing and clock speeds.

There's still plenty of room for haswell to be an impressive option compared to ivy bridge - clock speed, power consumption, graphics performance, etc - but we at least now know that in some kinds of tasks, performance is basically identical. I think their architecture improvements will make some benchmarks show a much larger difference, but apparently calculating pi or prime numbers hasn't gotten a performance boost since ivy bridge.

 

[LAN_deRf_HA]

So who get's to do the preview this time? Anandtech got it the first time. Tom's next. I hope I it goes back to Anandtech.

 

[Mathragh]

When it comes to the superpi benchmarks: this can be simply the result of intel moving away from optimising the ancient x87 instructionset for something a bit more modern.

I'm not sure however what is up with the wprime results, as that program should be able to use newer instruction sets? According to http://www.realworldtech.com/haswell-cpu/, Haswells IPC should be substantially higher than ivy bridge. I guess we can only wait and see, but I dont believe these results are 100% accurate.

 

[Pehla]

i belive intel say'd they are working on power consumption on haswel cpu and igpu.. they aim at notebooks with this one..i may be wrong but this comparsion just confirm my toughts..

 

[RejZoR]

Generally, if per clock performance is high, you also get very high performance with higher clocks. Unless CPU cannot be clocked high fopr some weird reason. But pretty much all go to 4GHz these days...

 

[SonDa5]

Cool to see some leaked benchmarks. I think they are legit.

All of these benchmarks are greatly effected by memory bandwidth as well. I am looking forward to seeing how well Haswell Integrated Memory Controller has improved for over clocking memory. These scores with the CPU at same speeds and max memory overclock on each cpu will show just how much stronger Haswell is over all.

 

[Prima.Vera]

How high in GHz those guys can go??

 

[NC37]

Hey uhh Intel...yeah, AMD called, they want their mediocre speed bump back.

 

[Nordic]

This is just one side of the story, practically just a teaser. Now we just need a full cough tpu cough review

 

[Dent1]

Needless to say, my FX 4100@3.6 is miles behind a 2600K@3.6

Are we supposed to be surpised that your cheap low end FX 4100 quadcore is mile behind an expensive high end quadcore 8 threaded 2600K.

 

[Rowsol]

Hey uhh Intel...yeah, AMD called, they want their mediocre speed bump back.

 

[qubit]

So there's almost no performance boost with a "brand new" architecture?

Nice to see AMD providing stiff competition to Intel.

We'll only gain if it clocks higher and has a proper soldered heatspreader and that remains to be seen.

 

[Prima.Vera]

So there's almost no performance boost with a "brand new" architecture?

Nice to see AMD providing stiff competition to Intel.

We'll only gain if it clocks higher and has a proper soldered heatspreader and that remains to be seen.

Maybe I am wrong, but probably this next gen will be some minor tweaking over prev gen, an increase in transistor count and maybe higher frequencies. This is how Intel plans for idiots to change their mobos into new one. Dark deal made with the mobo manufacturers.