relative performance of SB, IB and H?

[Scainer]

I have read a few reviews that show the performance of overclocked Haswell against Ivy Bridge.
These conclude that while heat is an issue and a lower clock is obtained, they are in fact faster.

What I'd love to see is a detailed comparison, showing the relative performance.
e.g.
(I assume these offer safe temps, not maximum clock)

Sandy was generally overclocked to 4.4 - 4.6Ghz (avg. 4.5)
Ivy was generally overclocked to 4.3 - 4.5Ghz (avg. 4.4)
Haswell is generally overclocked to 4.2 - 4.4Ghz (avg. 4.3)


Ivy was about 5% faster than Sandy so relative to Sandy the avg. would be 4400*1.05 = 4620

If Haswell is again 5% faster than Ivy that makes it more than about 7.5% (lets be super conservative) faster than Sandy, so ...
relative to sandy the avg. would be 4300*1.075 = 4622


The questions I'd like to know are......?

In theory a 4.2GHz Haswell offers the performance of a 4.5Ghz Sandy Bridge
Is this true in reality?
What are the comparative thermals like?
Is the heat issue really a problem for performance or is it just a case of my chip has a bigger number than yours, so it must be better?

So if the chips basically offer the same performance, are the other benefits (e.g. power saving) worth the consideration?
e.g. If you leave that computer on 24/7 how much would you save over a year? Would it pay for itself? Would the room be cooler?

[CAT-THE-FIFTH]

Why are you keeping the computer on 24/7 ??

[Scainer]

Why are you keeping the computer on 24/7 ??

Keep the house warm :-P (its a what if)

[CAT-THE-FIFTH]

Keep the house warm :-P (its a what if)

If you want to save power switch your PC off when not in use for extended periods!!

What do you run on your PC anyway??

[Scainer]

To make it easier, ignore the last two lines. Your fixating on an added thought not the content

[CAT-THE-FIFTH]

To make it easier, ignore the last two lines. Your fixating on an added thought not the content

Its not fixating. People worry about a few watts extra power consumption at load,and then keep their PCs on 24/7 consuming more power. Even the load measurements in many reviews are under Prime95,etc which tend not to be as much as situations like gaming.

If you want to save power,your usage habits determine yearly power consumption more in reality.

Moreover,you need to consider every other part of the system too:
1.)Efficiency of motherboard.
2.)Efficiency of PSU under 20% load which is not covered by the 80+ certification scheme. Too many people overrate the PSU they use and never bother to look at the full efficency curve.
3.)Efficiency of screen.
4.)Efficiency of graphics card.

Those three can easily add another 50W to a system or more.

People tend to fixate far too much on pure CPU power consumption than total system component power consumption,and usage patterns.

To answer your original quesion,if you are going for a Core i5 from the beginning,just get the cheapest one. IB does offer you PCI-E 3.0,so probably is the best overall mix.

Haswell,does run hotter than IB,and that has been confirmed by multiple reviews and forums. TS-X might have been a feature that would be useful,but Intel removes that from K series CPUs.

[kalniel]

The questions I'd like to know are......?

In theory a 4.2GHz Haswell offers the performance of a 4.5Ghz Sandy Bridge
Is this true in reality?

The results are real, so what I guess you mean by 'in reality' is something else, like 'are the results noticeable to a user'? To which the answer is the usual 'what are you using the computer for?' For an every day user, doing every day things like web browsing, facebook, the occasional document, spreadsheet and perhaps a bit of photo importing/uploading etc. then no, I doubt they would notice any difference between these three generations in terms of how they performed their every day tasks. However they might notice the difference in terms of the overall package and how long the battery in their laptop etc. would last, or the ability for their laptop to play a particular game.

What are the comparative thermals like?

I think the reviews cover that - it's proportionate to power usage and depends on the tasks, but in general, per task unit the newer generations use less power and thus are cooler.

Is the heat issue really a problem for performance or is it just a case of my chip has a bigger number than yours, so it must be better?

Depends what you mean by performance. Without any clarification I assume you mean the majority of users, in which case, no, heat isn't an issue for performance.

So if the chips basically offer the same performance, are the other benefits (e.g. power saving) worth the consideration?
e.g. If you leave that computer on 24/7 how much would you save over a year? Would it pay for itself? Would the room be cooler?

Again the question is over noticeable differences. Yes the room would be fractionally cooler, but would you notice? No. Would a computer pay for itself over a year in power savings? No.

[Scainer]

Haswell,does run hotter than IB,and that has been confirmed by multiple reviews and forums.

This is kinda my point. Does it run hotter on an even (performance) playing field, or is it just hotter at the equal clock.
e.g.
If Haswell is 5% quicker than IB, it can be clocked lower for the same thing. At this performance point is it actually hotter?



To clarify i am talking about parity. Not who has the biggest. A detailed comparison involving the three architectures.
Not a clock for clock comparison but a relative performance one. I thought I made that obvious in the first post.

[CAT-THE-FIFTH]

This is kinda my point. Does it run hotter on an even (performance) playing field, or is it just hotter at the equal clock.
e.g.
If Haswell is 5% quicker than IB, it can be clocked lower for the same thing. At this performance point is it actually hotter?

I am getting the impression,also after looking on few enthusiast forums,it seems to be generally worst than IB,and SB seemed better than IB. The issue is temperature (not necessarily the heat produced). At least from what I starting to see some reports,is that the current batch of production samples are also worse off in this regards than the ES review chips. The reviewer from Vor tez for example was talking about this on OcUK. Also,considering that in July(IIRC) Intel will fix that USB 3 wake issue with the Haswell chipsets too,I would be inclined to wait until then,and see what the consensus is,after more people have them in their hands,and perhaps more batches of chips are released too.

However,I cannot say about results with delidded IB and Haswell though,and there will be no warranty in this regard.

[Scainer]

I know that generation to generation there should be improvement to the speed, so comparing at a performance parity is wrong, its just that I'm interested to know the answers, in a way to argue against the negativity towards IB and Haswell overclocking.

e.g.
Person A has SB running at 4.5Ghz
Person B has Haswell running at 4.2Ghz

The assumption is that these would be very close from a performance point, but person A is bragging like an A-hole
So what about the other things, such as heat, power usage, etc
I would just like a detailed comparison, for arguments sake, so that person B can shut up person A.


I've also read about retail chips reacting worse to voltage increases, and in a way this is why I started the post.

People are moaning that it only clocks to X.XGhz, and conclude that they are going to buy IB cos its faster.
My argument is that it is not faster it just has a higher clock speed.
Peoples perception is that it needs to have a higher clock to be faster.

[CAT-THE-FIFTH]

I know that generation to generation there should be improvement to the speed, so comparing at a performance parity is wrong, its just that I'm interested to know the answers, in a way to argue against the negativity towards IB and Haswell overclocking.

People on enthuaist forums tend to know this,but the thing is that heat production has never been the issue with SB,IB or Haswell. AMD chips produce more heat for example. However,the temperature is the limiting factor,so in the end people have probably ended up needing to get more expensive coolers to keep the temperature lower,or even go to the extent of delidding their chips,and mounting the cooler on the bare die. SB seems to clock higher on the same cooling on average,so ultimately,the IPC improvements of the later chips are somewhat negated by the lower overclocking headroom on average. You also need to consider how agressive the Turbo is with each generation is too,which is the thing you need to look at,as it affects the actual performance boost from overclocking.

This is why SB is so highly regarded - it was capable of being overclocked a decent amount with even reasonably cheap coolers.

However,Haswell Core i5 CPUs do appear to run at high temperatures in a number of cases, even with good cooling when overclocked,and it tends to throttle in a number of cases under heavy load when overclocked if using stability testing programmes,it appears,so makes that aspect a bit more harder to gauge.

This is why I think its better to wait another month or so,to get more user testing of non-ES samples,to get a better impression on how well it overclocks on average. On top of this you will also get the new chipset stepping too.

If you are running at stock clockspeeds,Haswell is a good overall improvement,but for the overclocker I am not really that sure. Outside of PCI-E 3.0,which again might not be a big requirement in reality for a few years,I still would go for the cheapest setup.

For example Dabs,had the Core i5 2500K for only £130 recently,and I would rather have had one of those and spent the extra £50 to £70 on a more expensive graphics card.

The thing is even the socket 1150 roadmap lists 22NM Haswell as being current for all of next year,and even 14NM Broadwell seems to be more for BGA desktop and mobile systems.

[DanceswithUnix]

It's been a long day already, but by my thinking 5% compounded with 5% is 10.25%, not 7.5%.

4300*1.05*1.05 = 4740

So theory in the original post now holds with observation?

[Willzzz]

When Ivy came out everyone moaned that it wasn't as good as Sandy.
Now everyone thinks Ivy is great and people are moaning about Haswell.

Haswell is a little disappointing, but it's still worth it over older CPUs unless you find the odd bargain.
(open box i5-3550 for £120)

Really the talk about Haswell being a poor overclocker is being blown out of all proportion.
Firstly as you say a 4.2ghz Haswell is actually really a very powerful machine, and secondly people are finding better ways to overclock and are getting some really good results with retail chips, especially with newer BIOS revisions.

Also the new motherboards usually come with better features too.

[Scainer]

5% compounded with 5% is 10.25%, not 7.5%.

Brain is not working, something to do with the sun. I did 5% and then I just added half again for no real reason other than.....
Most quotes said 5% - 15% improvement so I just went with 7.5% to make the examples.
I actually think the 5% was a minimum, but it still illustrates the concept.


To make you happy, here are some updates

Sandy was generally overclocked to 4.4 - 4.6Ghz (avg. 4.5)
Ivy was generally overclocked to 4.3 - 4.5Ghz (avg. 4.4)
Haswell is generally overclocked to 4.2 - 4.4Ghz (avg. 4.3)


Ivy was about 5% faster than Sandy so relative to Sandy the avg. would be 4400*1.05 = 4620

If Haswell is again 5% faster than Ivy that makes it about 10.25% faster than Sandy, so ...
relative to sandy the avg. would be 4300*1.1025 = 4740

Or another way....
SB @ 4.5Ghz = IB @ 4.3Ghz = Haswell @ 4.08Ghz
SB @ 4.8Ghz = IB @ 4.57Ghz = Haswell @ 4.35Ghz
SB @ 5.0Ghz = IB @ 4.76Ghz = Haswell @ 4.53Ghz

[Willzzz]

Indeed, and 5% is a very conservative number.

We tend to see the smaller gains only when the CPU is bottlenecked.