Sandy bridge vs haswell, 30% better clock per clock?

[poohbear]

Hey all, getting bored with my 2500k @ 4.5ghz, is haswell 30% better performance wise at the same clock speeds? So 4.5ghz haswell would be 30% faster? Is that a correct assesment? Im asking for gaming @ 2560x1440 resolution. Talking about the 4 core i5 haswells here. Thanks in advance!

 

[sushiwarrior]

Haswell would be perhaps 10-15% better, and it is unlikely you would be able to get to 4.5GHz without de-lidding the processor. More like 4.2-4.3.

I doubt it would be worth it for you at all.

 

[Revolution 11]

30% more IPC? Doubtful except for some special applications like the emulators, Dolphin and PCSX2. These get a 25-30% boost from Sandy to Haswell.

 

[biostud]

For gaming the difference in fps will be less than 5%.

 

[Yuriman]

For gaming the difference in fps will be less than 5%.

This chart has been floating around for a little while now:

Granted, it's a beta. Also granted, not many games are Battlefield 4.

This is a specific usage case where going from a 2500K (3.3ghz) to a 4670K (3.4ghz) nets a 19% increase in minimums and a 21% increase in averages with only a 3% frequency increase. I'd wager it's pretty representative of a "best case" scenario that doesn't take advantage of instruction sets Sandy Bridge doesn't have.

You're as likely as not to lose about 5% clockspeed moving to Haswell so let's call it 15%.

 

[Concillian]

This chart has been floating around for a little while now:

[chart removed]

Granted, it's a beta. Also granted, not many games are Battlefield 4.

This is a specific usage case where going from a 2500K (3.3ghz) to a 4670K (3.4ghz) nets a 19% increase in minimums and a 21% increase in averages with only a 3% frequency increase. I'd wager it's pretty representative of a "best case" scenario that doesn't take advantage of instruction sets Sandy Bridge doesn't have.

You're as likely as not to lose about 5% clockspeed moving to Haswell so let's call it 15%.

2 things:

1) there is already a clock for clock comparison in comparing the OC 2500k and the OC 4670k (both at 4.5Ghz.) That shows a +16% on minimums for Haswell. OC2600k and OC4770k show 14% clock for clock. Actuality is probably 15% since the minimum FPS is a whole number. Averages are +14% for both the 4c/4t and 4c/8t CPUs when comparing Haswell to Sandy. Overall 14-15% is a good estimate.

2) If OCing and cooling is equal and you get an equal draw in the "silicon lottery," then Sandy will have a ~5-10% clock advantage, shrinking the overall Haswell advantage to the 5-10% range.

So Haswell will perform better, but only somewhat. There's also the matter that this is a beta and one of the most CPU intensive games out there. So in gaming cases, other games will likely show even less than a 5-10% advantage for Haswell... or it might show that 5-10%, but FPS is already high enough that the 5-10% don't matter AT ALL.

Point being, yeah, OP may be bored with his Sandy, but it doesn't make financial sense to upgrade to Haswell. The platforms are not really all that different and he will soon be just as bored of Haswell, but now several hundred dollars poorer.

 

[AtenRa]

You will not get more than 5% on average at 1440p. Better get more money and get a 6-core Haswell next year

 

[myocardia]

You will not get more than 5% on average at 1440p. Better get more money and get a 6-core Haswell next year

I couldn't have said it better myself. At that resolution especially, you're wasting your money on a CPU upgrade, unless you go hexacore.

 

[MBrown]

For me my i5 4670k @ 4.3 is .1% slower than my i5 2500k @ 4.5 using the unigine valley benchmark with the extreme preset @ 1080p. This is with my 7950 @ 1150/1425.

 

[schmuckley]

Definitely is NOT the case.
Performance increase is negligible.

 

[BallaTheFeared]

Haswell is roughly 20% faster than SB clock for clock when cpu limited.

If you aren't cpu limited, it is 0% faster.


Also initial batches were pretty garbage, however since then 4.5-4.6 seems much more reasonable with decent cooling, meaning above evo 212, below custom water.

 

[krumme]

Hey all, getting bored with my 2500k @ 4.5ghz, is haswell 30% better performance wise at the same clock speeds? So 4.5ghz haswell would be 30% faster? Is that a correct assesment? Im asking for gaming @ 2560x1440 resolution. Talking about the 4 core i5 haswells here. Thanks in advance!

Well i was bored with my ib and bought hw. Ended up giving hw to one of the kids. If you are bored with sb dont buy hw you will end up beeing depressed. Lol. Trust me you will get next to zerro performance benefit unless eco is important to you. Buy a new gpu it cant be worse

 

[WhoBeDaPlaya]

Nehalem -> Sandy (+10%)
Sandy -> Ivy (+3%)
Ivy -> HW (+10%)

 

[BallaTheFeared]

Nehalem -> Sandy (+10%)
Sandy -> Ivy (+3%)
Ivy -> HW (+10%)

Nehalem -> Sandy (+21% + 600~ MHz)
Sandy -> Ivy (+8% - 100~ MHz)
Ivy -> HW (+10% - 100~ MHz)


Nehalem to Sandy and Sandy to Haswell produces similar IPC gains, the biggest difference is Nehalem avg OC was around 3.6 to 4GHz for most, whereas with Sandy Bridge it jumped to probably 4.4 to 4.6GHz. Haswell didn't get that it actually probably lost some on avg vs Sandy.

Sandy was big because it combined a ~20% IPC increase with a rather decent clock speed gain, whereas Haswell only gets ~20%, a rather large decrease in power at higher clocks, but no additional on average clock speed increase.. It actually lost some I would say, though Haswell chips seem to be getting better in this regard.

 

[Hulk]

I made these calcs shortly after Haswell came out.

http://forums.anandtech.com/showthread.php?t=2324672&highlight=haswell

 

[poohbear]

Ok thanks guys. Ill save my money on cpu & get a new ssd. already running gtx 670s in SLI & very happy on the gpu end.

 

[krumme]

Ok thanks guys. Ill save my money on cpu & get a new ssd. already running gtx 670s in SLI & very happy on the gpu end.

sli 670 is super. And a new monster ssd is always welcome

 

[Brahmzy]

I went from a 4.6 SB to a 4.6 HW (haven't delidded yet) and I do actually feel a slight difference. Things are a bit smoother, a bit snappier. Same memory/timings. Same RAID0 256 840 Pros. Different Z87 board obviously. Gaming feels just a little smoother as well. Most of the benches I've ran show a slight increase to no increase, yet I feel a slight difference.
I was actually expecting to feel no increase from the upgrade, just wanted HW and the power consumption drops across multiple machines...

 

[Makaveli]

I went from a 4.6 SB to a 4.6 HW (haven't delidded yet) and I do actually feel a slight difference. Things are a bit smoother, a bit snappier. Same memory/timings. Same RAID0 256 840 Pros. Different Z87 board obviously. Gaming feels just a little smoother as well. Most of the benches I've ran show a slight increase to no increase, yet I feel a slight difference.
I was actually expecting to feel no increase from the upgrade, just wanted HW and the power consumption drops across multiple machines...

And you sure this isn't just the placebo effect.

i'm going to assume you did a fresh os install also since you changed boards.

 

[Yuriman]

*snip*

1) there is already a clock for clock comparison in comparing the OC 2500k and the OC 4670k (both at 4.5Ghz.) That shows a +16% on minimums for Haswell. OC2600k and OC4770k show 14% clock for clock. Actuality is probably 15% since the minimum FPS is a whole number. Averages are +14% for both the 4c/4t and 4c/8t CPUs when comparing Haswell to Sandy. Overall 14-15% is a good estimate.
*snip*

I chose the lower clocks intentionally to minimize "fast enough" which apparently was starting to happen @4.5 because the lead was closing.

But yeah, overall, probably not worth spending any money on 15-20% unless you have a very specific use case.

 

[Brahmzy]

I've been doing this WAY too long for placebo effects. I know exactly what the "behavior" of each of my boxes feels like at any given time. Always fresh installs / secure erases.

I'll be the first to admit going from a perfectly good OC'd SB to an OC'd HW makes no fiscal sense. That is unless you need a particular feature from Z87 or are looking at power consumption or whatever. And even power consumption would take a while to pay for itself. Most of my hobbies usually don't make fiscal sense, lol.

 

[sniffin]

I feel like smoother and snappier in Windows is probably a placebo if you were already running a modern CPU, an overclocked Sandy no less. 95% of the time I don't notice the difference in Windows between my notebook with a quad core Richland APU and my desktop with a 3770k, both have Samsung 840 Pros. Gaming is another story ofc.

 

[poohbear]

I've been doing this WAY too long for placebo effects. I know exactly what the "behavior" of each of my boxes feels like at any given time. Always fresh installs / secure erases.

I'll be the first to admit going from a perfectly good OC'd SB to an OC'd HW makes no fiscal sense. That is unless you need a particular feature from Z87 or are looking at power consumption or whatever. And even power consumption would take a while to pay for itself. Most of my hobbies usually don't make fiscal sense, lol.

Wouldnt u also attribute that to a fresh OS install? When i do a fresh install its always snappier.

 

[sm625]

Unless you're delidding the haswell and going for a good delidded overclock, I cant see where it makes sense to try upgrading from a 2500K.

 

[Brahmzy]

Maybe, maybe not. Honestly, since W7, I haven't noticed much diff between multiple fresh installs on the same hardware. Once all of the 'crap' has been reinstalled on the fresh install, it feels like a 1yr old install. With the new HW chip and Z87 things actually feel snappier. Yes, we're splitting hairs, but I work on these things all day, everyday and I know how many ms it takes for IE to open the first time on all the boxes and pretty much every other routine task.
HOWEVER, I also went from W8 to W8.1 (Metro BS disabled of course) on these fresh installs. Surely MS didn't make any actual usable performance improvements to the W8 code, or did they? I've heard Server 2012R2 got a bit of work done in that regard. I have been very happy with W8.1's performance - once I get Metro shot down as much as I possibly can. Turning off the active tile crap, running Default Programs to repoint everything back how it ought to be etc. IE11 is another story. Full of freaking bugs.
YMMV

Another thought - I did come from P8P67 (not Z77) boards to Z87s on some of these. Obviously the fast boot is faster, but could the Z87 (2 generations newer) new RST features (accelerated storage) make a difference?