The more I read about Haswell's desktop chips, the less impressed I am becoming, all aspects. :\
To summarize:-
- It runs hotter than IB and still uses the same crap-TIM material with the same unwanted 0.2-0.5mm "gap" between core & heat spreader requiring high-risk delidding to reduce temps back to 10-15c lower Sandy / Nehalem performance. Claiming it's due to "cost" is nonsensical given that Sandy & Nehalem Pentium & Celeron chips were soldered without cost issues.
- In theory it's 10% faster, in practise it's as little as 1% in many games (eg, Skyrim is only 1%, WinRAR only 4%)
- It uses more power under load than IB despite the fact the entire "Haswell" concept is about power efficiency. Some benchmarks even say it uses more when idle...
- It's more expensive than Ivy Bridge even factoring in performance.
- i5/i7 "K" owners - It appears to have a 5% (200-300MHz) lower overclocking potential than IB over 4.5GHz instantly nullifying much of the IPC speed gain. +85c at 1.3v @ 4.5GHz with high-end aftermarket coolers? Ouch. Literally. From what I understand, this is due to the now integrated regulator automatically increasing voltage by 0.1v at high loads (whereas before on the motherboard, you had to manually compensate for "droop" which can vary between power supplies). I also understand that whereas before you could tweak the various motherboard settings, now you can only disable the effect by locking the voltage which disables all power-saving technologies, sending Haswell's idle power consumption soaring. So you either put up with higher temperature or sacrifice power savings in idle mode. For a CPU specifically marketed as a "power saver". "Genius". :thumbsdown:
http://www.xbitlabs.com/articles/cpu/display/core-i7-4770k_12.html
- i5/i7 "K" owners - "K" chips don't support TSX-NI, vPro or VT-d extensions for no technical reason whatsoever
- i5/i7 non-"K" owners - It no longer allows +400Mhz "4 bins" overclocking on non-"K" i5 chips. 4570 (3.2GHz / 3.6GHz turbo) is stuck at 3.6GHz vs 3470 (3.2GHz / 3.6GHz turbo + 4-bins = 4.0GHz). Since Haswell hits +10% IPC efficiency only in a very few apps, this means that a 3.6GHz maxed 4570 is now slower on average than its 4.0Ghz maxed 3470 Ivy equivalent...
http://www.hardware.fr/articles/897-6/overclocking-plus-libre-k-plus-strict-par-ailleurs.html
- i3 dual core owners - BCLK overlocking is only allowed on "K" i5/i7 chips (but there are no unlocked i3 "K" chips) meaning 3.3GHz i3 Haswell's stuck at 3.3GHz *still* run no faster than 3 generations old Clarkdale i3-530's which could be BCLK OC'd to +4.2GHz and then some. 0% improvement over 3 generations is a catastrophic failure for dual-cores. Real-world examples:-
i3-530 @ 4.2Ghz = 3,818KB/s WinRAR. (Clarkdale)
i3-2120 @ 3.3GHz = 3,304KB/s WinRAR. (Sandy)
i3-3220 @ 3.3GHz = 3,557KB/s WinRAR. (Ivy)
i3-4220 @ 3.3GHz = 3,702KB/s WinRAR*. (Haswell) - predicted with 4% avg IPC boost vs Ivy.
It's pretty sad if it's going to take a Broadwell i3 just to match a Nehalem due to crippled OCing on dual-cores.
The 4 generation ride was a great seamless stream of improvements (Core 2 Duo -> Nehalem -> Sandy -> Ivy), but this one has somehow managed to screw over almost everyone simultaneously, each for different reasons...
Personally, I'd have been happy with the IPC improvements had say, the chip been soldered and temps were reduced by 10-15c. Or if BCLK OCing had been for all chips, not just limited to "K's". Or if an i3 "K" chip came out for the return of budget overclocking. Or if "K" chips didn't have TSX-NI unnecessarily disabled. So far though, everything seems to be one downgrade / compromise or another that isn't an "upgrade" at all...
Your thoughts?
To summarize:-
- It runs hotter than IB and still uses the same crap-TIM material with the same unwanted 0.2-0.5mm "gap" between core & heat spreader requiring high-risk delidding to reduce temps back to 10-15c lower Sandy / Nehalem performance. Claiming it's due to "cost" is nonsensical given that Sandy & Nehalem Pentium & Celeron chips were soldered without cost issues.
- In theory it's 10% faster, in practise it's as little as 1% in many games (eg, Skyrim is only 1%, WinRAR only 4%)
- It uses more power under load than IB despite the fact the entire "Haswell" concept is about power efficiency. Some benchmarks even say it uses more when idle...
- It's more expensive than Ivy Bridge even factoring in performance.
- i5/i7 "K" owners - It appears to have a 5% (200-300MHz) lower overclocking potential than IB over 4.5GHz instantly nullifying much of the IPC speed gain. +85c at 1.3v @ 4.5GHz with high-end aftermarket coolers? Ouch. Literally. From what I understand, this is due to the now integrated regulator automatically increasing voltage by 0.1v at high loads (whereas before on the motherboard, you had to manually compensate for "droop" which can vary between power supplies). I also understand that whereas before you could tweak the various motherboard settings, now you can only disable the effect by locking the voltage which disables all power-saving technologies, sending Haswell's idle power consumption soaring. So you either put up with higher temperature or sacrifice power savings in idle mode. For a CPU specifically marketed as a "power saver". "Genius". :thumbsdown:
http://www.xbitlabs.com/articles/cpu/display/core-i7-4770k_12.html
- i5/i7 "K" owners - "K" chips don't support TSX-NI, vPro or VT-d extensions for no technical reason whatsoever
- i5/i7 non-"K" owners - It no longer allows +400Mhz "4 bins" overclocking on non-"K" i5 chips. 4570 (3.2GHz / 3.6GHz turbo) is stuck at 3.6GHz vs 3470 (3.2GHz / 3.6GHz turbo + 4-bins = 4.0GHz). Since Haswell hits +10% IPC efficiency only in a very few apps, this means that a 3.6GHz maxed 4570 is now slower on average than its 4.0Ghz maxed 3470 Ivy equivalent...
http://www.hardware.fr/articles/897-6/overclocking-plus-libre-k-plus-strict-par-ailleurs.html
- i3 dual core owners - BCLK overlocking is only allowed on "K" i5/i7 chips (but there are no unlocked i3 "K" chips) meaning 3.3GHz i3 Haswell's stuck at 3.3GHz *still* run no faster than 3 generations old Clarkdale i3-530's which could be BCLK OC'd to +4.2GHz and then some. 0% improvement over 3 generations is a catastrophic failure for dual-cores. Real-world examples:-
i3-530 @ 4.2Ghz = 3,818KB/s WinRAR. (Clarkdale)
i3-2120 @ 3.3GHz = 3,304KB/s WinRAR. (Sandy)
i3-3220 @ 3.3GHz = 3,557KB/s WinRAR. (Ivy)
i3-4220 @ 3.3GHz = 3,702KB/s WinRAR*. (Haswell) - predicted with 4% avg IPC boost vs Ivy.
It's pretty sad if it's going to take a Broadwell i3 just to match a Nehalem due to crippled OCing on dual-cores.
The 4 generation ride was a great seamless stream of improvements (Core 2 Duo -> Nehalem -> Sandy -> Ivy), but this one has somehow managed to screw over almost everyone simultaneously, each for different reasons...
Personally, I'd have been happy with the IPC improvements had say, the chip been soldered and temps were reduced by 10-15c. Or if BCLK OCing had been for all chips, not just limited to "K's". Or if an i3 "K" chip came out for the return of budget overclocking. Or if "K" chips didn't have TSX-NI unnecessarily disabled. So far though, everything seems to be one downgrade / compromise or another that isn't an "upgrade" at all...
Your thoughts?
Facebook
Twitter