News - Intel Haswell processors to launch on 2nd June

[crossy]

Another new socket ... you kidding me ?
People's motherboards become obsolete so quickly! Am I the only who thinks we are being ripped off ?

Not sure we're necessarily being "ripped off", but being a bit more charitable it does look to me like Intel's kind of "lost the plot" and is flailing around.

AMD FX8350 only £140 in most places. And as a bonus you get a room heater thrown in too. Planning to upgrade my X3 720 to the FX8350 this month.

I take it that the comment means that the FX8350 run's a high TDP then?

Looks like I'm going to be sticking with my old 1090T for a good while longer - especially as jumping platform means having to do a ground up rebuild of Windows (just wish there was an easy way to "encapsulate" apps so I didn't have to reinstall, reconfigure and repatch).

[kompukare]

Another new socket ... you kidding me ?
People's motherboards become obsolete so quickly!
Still remember when I bought P67 last year and it was all the hype!

Socket 775 = P4 / Pentium D / C2D / Quad
Socket 1156 = i3/i5/i7
Socket 1155 = i3/i5/i7
Socket 2011 = i3/i5/i7
And, now Socket G3 = i3/i5/i7

Am I the only who thinks we are being ripped off ?

Yes LGA775 was a record for longevity. And S1156/S1155 (like FM1/FM2) seemed to have been a socket for no good reason. But Haswell's S1150 does have a reason: Haswell CPUs will have the VRMs in the CPU package as Intel is in the slow march to SOA where the mobo will become almost nothing (there are rumours that after Broadwell, Intel might drop PCI-E slots - which frankly seems mad!).

Are you saying that with Ivy Bridge chips FSB overclocking doesn't work or doesn't work well? I know its easier with the K model just to increase the mutlipler but I didn't think increasing the FSB would cause any harm, especially on non K chips?

Yes as watercooled already said with SB and IB Intel basically killed FSB overclocking since the FSB is totally linked to everything. Sort of like how it it used in the early days of overclocking with most chipsets but with the difference that SB/IB CPUs have clock generators onchip where the mobo only gets to supply the FSB and the CPU is now responsible for setting any external clocks like PCI-E/SATA etc. - hence everything quickly becomes unstable 103+ FSB. Overclocking with data corruption is not my idea of fun!

[iranu]

Plus the cost of the motherboard needed for overclocking and the cooler. £160 to £170 is not really a bargain ,when the Core i5 700 series could be had cheaper for around £140 to £150 and also overclocked well too,let alone the Q6600. The Q6600 could be had for as low as £110 delivered in 2008.

My E4500 and Q6600 lasted over 4 years and cost me less than £150 in total. I only moved over to socket 1155 for gaming, after nearly 5 years since the motherboard started to have issues,if not I would still be probably using it now.

Edit!!

Even the Core i5 2500K was closer to £150 in 2011.

Well you've got to buy the board and new ram anyway. Considering inflation and the drop in the pound a 4 core chip that is only £10 more than a Q6600 is good going imho. I always believe in buying quality motherboards anyway because they are likely to last. I'd spend £350 for chip, board and ram on an i5 system anyway so I'm hoping that the price isn't going to be sky high and I can skip 2 generations. I suspect a 4670K will be released at the same price point.

Now if a 3570K was reduced to £120 or less when Haswell debuts then that would be better value, but i5 3570K is only £10 more than a 2500K months after IB came out.

We need more competition from AMD.

[watercooled]

I take it that the comment means that the FX8350 run's a high TDP then?

Not as high as some places would have you believe, and probably a fair bit lower than the Thuban CPUs we both use. It's also important to take the motherboard into account; high-end overclocking boards tend to have relatively inefficient VRMs, and the difference can be easily 20W or so according to what I've seen. Smaller boards with less features also tend to use less power.

Yes LGA775 was a record for longevity. And S1156/S1155 (like FM1/FM2) seemed to have been a socket for no good reason. But Haswell's S1150 does have a reason: Haswell CPUs will have the VRMs in the CPU package as Intel is in the slow march to SOA where the mobo will become almost nothing (there are rumours that after Broadwell, Intel might drop PCI-E slots - which frankly seems mad!).

IIRC, 1156 and 1155 even used the same silicon for the chipsets, just with less fuses blown on the 'newer' ones. I think the same might be true about FM1/FM2 but I seem to remember reading about a genuine reason for the quick jump to FM2, but I could be confusing it with something else, maybe someone else remembers?

As for 1150, there seems to be a few reasons for the change (although they likely could have worked around it very easily if they wanted to). One being this talk of VRMs. Only the PWM controller is likely to end up on the CPU itself, the actual power circuitry i.e. FETs, caps, inductors, will still be around the socket on the motherboard. Without a good explanation, I'm not convinced there's any intrinsic advantage to this, sure some motherboards don't have great circuitry, and better VRM control can improve efficiency and OC headroom, but I can't think of any reason this can't be done by external circuity. However, Intel taking control of more of the board will likely improve their image on some charts i.e. reviewers are less likely to end up with inefficient boards (like a lot seem to have for AM3+), but they would also be able to easily limit power, so overclocking ability.

[CAT-THE-FIFTH]

Well you've got to buy the board and new ram anyway. Considering inflation and the drop in the pound a 4 core chip that is only £10 more than a Q6600 is good going imho. I always believe in buying quality motherboards anyway because they are likely to last. I'd spend £350 for chip, board and ram on an i5 system anyway so I'm hoping that the price isn't going to be sky high and I can skip 2 generations. I suspect a 4670K will be released at the same price point.

Now if a 3570K was reduced to £120 or less when Haswell debuts then that would be better value, but i5 3570K is only £10 more than a 2500K months after IB came out.

We need more competition from AMD.

A Q6600 for £110 to £120 in 2008 would be around £120 to £130 AFAIK in today's money,plus my total E4500 and Q6600 upgrade path cost me around £130 in total in todays money IIRC.

The retail Core i5 2500K was generally around £160 until late 2011 and could be had for £150 quite often. However,the prices are now on average around £175 for the retail K series CPUs and yet none of the other Intel CPUs seem to have show this inflation in price - in fact some are even a tad cheaper than before,than in 2011. Low end locked Core i5 CPUs are now cheaper than the 2011 models and that is without inflation being considered.

Looking at the US,it appears that it is probably a bit price gouging by UK retailers. They are no more a bargain than the socket 1156 Core i5 CPUs,let alone the Q6600. They are good CPUs for the price no doubt.

[Caveman]

Yes as watercooled already said with SB and IB Intel basically killed FSB overclocking since the FSB is totally linked to everything. Sort of like how it it used in the early days of overclocking with most chipsets but with the difference that SB/IB CPUs have clock generators onchip where the mobo only gets to supply the FSB and the CPU is now responsible for setting any external clocks like PCI-E/SATA etc. - hence everything quickly becomes unstable 103+ FSB. Overclocking with data corruption is not my idea of fun!

But also in the early days you could use the Async and Sync options which allowed you to increase the FSB but leave things like the SATA/PCIE at 100Mhz is it not the same for SB/IB? Im just curious, if you can't separate the other buses then meh.

[watercooled]

No, in the past, the clocks were controlled by (often variable) multipliers handled by the Northbridge, now they're all controlled by the CPU, a fixed value, and based on the CPU base clock.

[MercutioUK]

The fact we have a new generation of consoles around the corner is quite likely to give the PC a bit of a shot in the arm. While the "bleeding edge" sector of PC hardware is likely quite small on the whole it's like pretty much every other industry where the top end moves the goal posts a little for everyone/thing else too.
I hate console ports as much as the next guy but it does have to be recognised as one of the drivers of the hardware market. When lazy ports need better hardware to run, the hardware will be made.
The other factor I think in the "expected upturn with Haswell" is the green end of the market and the miniaturisation that comes with it. I think there's a lot of folks who still don't have a PC in the shower/kitchen/bedside table (I'm being at least 50% ironic here) that a smaller, more powerful, less power-hungry chip would help immensely.
Lastly there's the mobile end, given the expected increase in graphical power on the Haswell's there's a decent amount of "average gamer" type laptops that would probably be fairly well represented with a Haswell chip. If a lot of these folks are using older hardware and a dedicated GPU a laptop with similar power might well look quite promising.

As an aside - it also means thunderbolt might get a bit more mainstream which can only be a good thing. High bandwidth, low latency peripherals with a more direct access to the system bus opens a lot of rather interesting possibilities. I'm even betting I see ideas I have for personal setups monetized and offered by the "boutique PC" type places.

[borandi]

is there any reason to buy haswell? it is built on the same tech (22nm) and has either the same TDP or a higher one (due to the IGP).

If you're using a high end Sandy/Ivy bridge processor I don't see any reason to upgrade, it's at most a 17% increase in performance per clock from Ivy bridge. The one thing that has me interested though is the chief designer said it will be "an overclockers dream" in an AMA on Reddit.

Did you pull the number out from thin air or actually have benchmark results?

Regarding higher TDP, it's all about the integrated voltage regulator causing the increase. So if you save 10W of a 77W chip due to the uArch change, add it back due to the iVR. AnandTech covers this in their podcast

[shaithis]

But Haswell's S1150 does have a reason: Haswell CPUs will have the VRMs in the CPU package as Intel is in the slow march to SOA where the mobo will become almost nothing (there are rumours that after Broadwell, Intel might drop PCI-E slots - which frankly seems mad!).

Moving VRMs into the CPU package is probably part of the NTV push intel are on and the need to control the power to the CPU more granularly for it to work.

[watercooled]

They're basically what Intel quoted in terms of CPU performance IIRC.

Again on the subject of the VRM, actual details are scarce, 'the are integrating VRM on the CPU!!' means very little. As I said earlier, integrating the PWM controller is plausible but not necessarily worthwhile, integrating the switching FETs on a logic process would seem infeasible, expensive and pointless, and I doubt Intel would be glueing the storage components like inductors and capacitors to the CPU, so there will still be some motherboard parts to the 'VRM'.

It could be something more to do with power granularity but that would *not* be a replacement for the motherboard DC-DC converter.

I can't seem to find any details about what voltage Haswell takes, if the VRM was fully integrated on the CPU somehow, you'd expect it to directly accept 12v, but I've not seen that mentioned anywhere.

I'll happily read further details to the contrary if anyone has found any, but it seems this VRM acronym is bandied around as an explanation for everything.

The larger TDP is likely more to do with the much larger IGP than anything else, considering they're on the same node as IVB.

[MercutioUK]

One thing that should be nice is thunderbolt is brought into the chipset rather than being an external controller. If we get a decent bandwidth, direct to bus connector as standard you can probably expect quite a few interesting peripheral developments this year.

It's MUCH more interesting in a laptop (external GPU in a docking station type setup) with fairly poor gaming potential built into the chip when the laptops away (or use a decent GPU + a super GPU when docked). Otherwise I'm not so convinced Haswell brings so much more to the table.