Haswell alleged benches

Status
Not open for further replies.
Isn't this sort of what people were expecting? Haswell was targeting power use most aggressively I thought. Besides if it overclocks really well who cares.
 
^ Yeah. Intel doesn't really need to push as hard anymore. Wouldn't be surprised if the IHS has the same 'problem' as ivy did. It should overclock better hopefully at least at a lower voltage would be alright too.
 
Haswell's claims to fame are supposed to be power usage and graphics, right? So that's good for mobile, even if it's less exciting for those desktop users who use discrete graphics instead of on-CPU graphics.
 
I guess. But I think it should be a wee bit faster. Just a hair. But we'll see.
 
I wouldn't listen to Francois Piednoel, as he's known to toss around quite a bit of crap.

Nevertheless, those benchmark numbers do look fake. Haswell has a 10-15% bump in IPC over Ivy. Though the clock speeds haven't budged, the performance should still be an okay step up.
 
blastingcap said:
Haswell's claims to fame are supposed to be power usage and graphics, right? So that's good for mobile, even if it's less exciting for those desktop users who use discrete graphics instead of on-CPU graphics.
Click to expand...



True, and honestly that was kind of Intel's strategy from Sandy Bridge through Broadwell from what I've been hearing these past few years. Intel's main focus is power consumption and the mobile segment and probably wont refocus that until Skylake or later IMHO. Sandy Bridge was kind of a freak of nature in it being a very solid leap ahead of Nehalem. Ivy Bridge wasn't supposed to be anything major and not a lot of people had high hopes for it. I honestly blame the Tech media. For some strange reason, months prior to its release they were pumping it up like crazy. Then what went from just a die-shrink with minor tweaks turned out to be an expected solid 10% bump clock for clock to Sandy. That didn't happen to say the very least and it struggled between 3-7% on the CPU side.

Because of that it basically got crapped on despite the lower TDP and higher IGPU because of the IHS overclocking heat temp fiasco and not really seeing any performance leap. The Intel team did everything right, but the journalists had their head in the clouds, perhaps reading too many "rumors".

Haswell has been pumped up for so damn long (since Sandy) and after Ivy I don't think Intel wants to miss out on selling this generation to the masses outclassing all other previous models. Whether that remains to be seen is uncertain. But I can tell you that an early engineer sample doesn't really tell you much. I expect CPU's will be a snoozefest for a long time to come thanks to the mobile market, AMD's lag, and already being excessively powerful for modern machines. Not to mention any new CPU extensions will probably get thrown to the wayside for many years until developers start actually taking advantage of them.
 
InternationalHat said:
Isn't this sort of what people were expecting? Haswell was targeting power use most aggressively I thought. Besides if it overclocks really well who cares.
Click to expand...

Agreed. I never expected a huge performance increase, but then again CPU performance isn't a huge deal to me. I keep my CPU/Mobo combos for 3-4 years and just upgrade my graphics card every 1-2.

I am expecting a huge dip in power consumption though, and have been waiting on Haswell to get a new laptop.
 
It seems, by biggest hardware upgrades this or next year will be PS4/X720 :). No point of getting rid of my 2500K
 
pelo said:
Haswell has a 10-15% bump in IPC over Ivy. Though the clock speeds haven't budged, the performance should still be an okay step up.
Click to expand...

They said the same thing about Ivy Bridge. It ended up being more like a 5-9% bump.
 
DeathFromBelow said:
They said the same thing about Ivy Bridge. It ended up being more like a 5-9% bump.
Click to expand...

The Ivy figures were regarding performance bump, not IPC bump. With the small jump in clock speed and the single-digit IPC increase, Ivy ended up being ~10-15% faster than Sandy. Haswell isn't getting a clock speed bump but rather the increase in performance is coming strictly from IPC gains.
 
pelo said:
The Ivy figures were regarding performance bump, not IPC bump. With the small jump in clock speed and the single-digit IPC increase, Ivy ended up being ~10-15% faster than Sandy. Haswell isn't getting a clock speed bump but rather the increase in performance is coming strictly from IPC gains.
Click to expand...


It's too early to tell that Haswell will only be IPC improvements leading to its figures, much less if they figures hold true. I have yet to see a review where Ivy held a solid 10% in more than a couple tests w/o OC and when comparing SB OC to IVB OC I'd take Sandy.

It wasn't that it was bad, it was just misleading in its claim to be that big of an improvement over SB. The wikipedia article quotes 5-15% improvement clock for clock compared to SB and that just wasn't true. That's a pretty big gap, but for the same price you weren't really losing much.
 
Liger88 said:
It's too early to tell that Haswell will only be IPC improvements leading to its figures, much less if they figures hold true. I have yet to see a review where Ivy held a solid 10% in more than a couple tests w/o OC and when comparing SB OC to IVB OC I'd take Sandy.

It wasn't that it was bad, it was just misleading in its claim to be that big of an improvement over SB. The wikipedia article quotes 5-15% improvement clock for clock compared to SB and that just wasn't true. That's a pretty big gap, but for the same price you weren't really losing much.
Click to expand...

I saw similar in many reviews especially on [H] that the speed improvements between SB and IVB were nowhere near the supposed 10% to 15% increase. In only very, very few certain applications did IVB show a big boost over SB, but it wasn't 15%. And, on websites like [H], at the same clock speeds, SLI and Crossfire cards of 2 or more cards showed better improvements and benefits with IVB over SB.

It's those same reviews that made my friend decide to wait until Haswell is released before he upgrades his i7 920 system. His Core i7 920 served him so well and plays all of his games without issue. He didn't see a cost benefit to go to Sandy Bridge when the performance and gaming experience would still be the same regardless. That's especially since he'll need a new board to go with it. To him, Haswell has to be SIGNIFICANTLY faster than his Core i7 920, Sandy Bridge and Ivy Bridge for him to upgrade. It has to show significant performance improvements in the games he currently plays-- games from two years ago and to be released within the next two years-- for him to be willing to spend the extra money on a new system build. We're not talking about 5 extra frames per second here, we're talking about 10 to 15 FPS or higher at 1080p resolution with a single high-end card according to him.

For example, if an Nvidia GTX 680/Sandy Bridge is 65 FPS with all settings on very high in one game, and a GTX 680/Ivy Bridge is 70 FPS with the same settings, he wants a GTX 680/Haswell combo hit 80 to 85 FPS.

Then again, my friend probably has too high of expectations. :p
 
Well, IPC will vary depending on the workload.

1335129654uPyIt9VZzQ_4_3.png


1335129654uPyIt9VZzQ_4_2.png


Xbitlabs is down, but iirc they did a comprehensive IPC test between the two. The IPC varied by nothing to ~16%. The average was in the mid single digits, and with the 100mhz bump it's roughly ~10%+/-. That's roughly the same range you get from Zambezi > Vishera (perhaps this one is slightly higher?), so it isn't an intel-only thing. We're just not going to make big jumps in CPU performance year-to-year anymore. Steamroller could be the exception here, but only because AMD royally screwed up with the Bulldozer uArch and has a lot of room for improvement. Intel can't just keep widening the cores and expect not to pay for that in die space and power consumption. Enthusiasts would love for them to do that, but enthusiasts only make up a tiny percentage of their market.

Here are the Vr-zone benchmarks:

sandyivy48-1.jpg.jpeg


sandyivy48-2.jpg.jpeg


sandyivy48-3.jpg.jpeg


Clearly there's a pretty big variance and the gains weigh heavily on the workload. To make a good judgment on IPC you'd have to compare the processors under a variety of workloads and then take the average - or even a weighted average; for example, for some folks the AVX performance in X application is far more important than gaming or what have you. So when we say "IPC" it's a very very broad term that mean a whole variety of things, but it's only with the graphs like the one above with which you can make an assertion regarding the average IPC.
 
Last edited:
Status
Not open for further replies.
Back
Top