Leaked slides expose Haswell’s integrated voltage regulator

Intel is slowly but surely moving more platform components onto the processor. Over the years, the graphics, PCI Express, and memory controllers traditionally found on separate north-bridge chips have migrated onto the CPU die. With Haswell, voltage regulation is also coming onboard. EXPreview has posted leaked slides that detail the voltage regulator inside Intel’s next-generation CPU.

The FIVR, or Fully Integrated Voltage Regulator, promises higher efficiency, finer granularity, and cleaner power delivery than traditional solutions, which rely on the motherboard for voltage regulation. The FIVR apparently consolidates CPU core, graphics, system agent, I/O, and PLL voltage regulation into a single unit. Voltage regulation for the memory is still handled separately by the motherboard, though.

According to the slides, the FIVR is based on a multi-cell architecture with 20 cells per chip. Each cell behaves like a mini regulator and is capable of handling current up to 25A. The switching frequency of the cells is programmable between 30 and 140MHz, the slides say, and there are 16 power phases per cell. You’re looking at 320 phases per chip, which is quite a lot.

The slides credit the large number of power phases with reducing ripple and noise. Voltage ripple is "almost non-existent," with worst-case scenarios yielding no more than +/- 0.002V. The voltage drift due to temperature increases is apparently just 0.001V. Those figures appear to refer to a 90-nm version of the power cell. Haswell’s implementation should be built with more advanced 22-nm transistor tech.

Even when fabbed with 90-nm tech, the integrated regulator is said to be about 50 times smaller than typical motherboard implementations. There is a cost, though. Because the regulator is part of the CPU, it contributes to the thermal footprint of the chip. Perhaps that’s why the TDP of high-end Haswell processors is 84W, up from 77W for Ivy Bridge.

Haswell’s FIVR was surely designed with mobile in mind, and it will no doubt pay dividends in highly integrated systems like tablets and notebooks. Desktop users can probably look forward to greater power efficiency, too. That said, overclockers may hit thermal limits sooner than they did with Ivy-based chips, and motherboard makers will need to find new ways to hype their boards’ power regulation circuitry.

Comments closed
    • Hattig
    • 7 years ago

    Looks like good R&D work from Intel here. Should reduce motherboard price as well, as motherboards slowly move to being mere chip sockets and I/O ports connected together with no complex logic or power regulation of any sort.

      • tipoo
      • 7 years ago

      I guess that would also reduce the impact of CPUs being bolted right onto motherboards like Intel seems to want to move towards.

    • pedro
    • 7 years ago

    The glory days of massive E-ATX boxes sporting ridiculous 2KW PSUs are, it must be said, in the rearview mirror of humanity.

      • tipoo
      • 7 years ago

      Yeah, and it does make technological sense, but I’ll miss those glory days. I wish at least someone still made balls-out performance consumer processors with no thought given to mobile or on chip GPUs. Imagine if IBM was handed an x86 licence and brought Power7 over, lol.

    • DavidC1
    • 7 years ago

    spuppy:

    That’s because the power cell in the article is an experimental chip, not something that’ll be going in Haswell. The project code-name is Ozette, Expreview is just confusing Ozette to be Haswell’s and Tech Report is just regurgitating Expreview’s information.

    Not to say the evolved version won’t be in Haswell, but its not THIS one.

      • spuppy
      • 7 years ago

      Can you locate it somewhere on the package? I can’t locate it on die or off…

    • spuppy
    • 7 years ago

    Looking at leaked Haswell chip pictures, it looks like the desktop version will have 2 of these power cells:

    [url<]http://cdn2.wccftech.com/wp-content/uploads/2013/04/Haswell-Core-i5-CPU-Back.jpg[/url<] You can see the traces for them at the top, above the CPU die. But it looks like there are only 6 phases per cell, rather than the 8 on these slides. Unless I am mistaken, and that's just part of the die. In fact there are leaked delidded pictures out there: [url<]http://www.firegoon.com/u/sikhnj.jpg[/url<] If that's the case, the implementation is nothing like these slides... Instead, the IVRM is on the die itself. If so, you can look for it on leaked die shots: [url<]http://i272.photobucket.com/albums/jj163/idontcare_photo_bucket/Intel%20Core%20i7-2600K/SandyBridgeTempsandDarkSilicon.jpg[/url<]

    • just brew it!
    • 7 years ago

    Some of those slides apparently came from here:

    [url<]http://www.psma.com/sites/default/files/uploads/tech-forums-nanotechnology/resources/400a-fully-integrated-silicon-voltage-regulator.pdf[/url<]

    • Bensam123
    • 7 years ago

    Hmmm I can only imagine the overclocking limitations that could be imposed by this.

    As Geoff mentioned you’re going to have more heat to dissipate, which was a issue with IB while OCing even if you don’t reach the absolute maximum your chip can offer, it starts overheating before that happens and then throttles itself.

    There is nothing stopping Intel from simply disallowing users from over-volting their processors so they can’t OC at all though. For none K series chips, this means they can completely stamp out any round about ways of OCing once and for all. For K series chips this means they can limit how much juice you pump into your processor.

    They could also decidedly start making tiers of OCing processors if they want to, since the binning processes isn’t nearly as fine tuned. They could simply offer K series chips that only over-volt so far and you have to buy the extreme K series in order to over-volt further.

    This is all starting to tie into the comment I made earlier about Intel moving into a microtransaction model by further removing what influence end users have over their chips. It’s all good and fun that having BGA chips will reduce costs to Intel and .002v ripple is nothing to sneeze at, but this is all just foot in the door. Once they have this tech out for a bit, they start regulating it then. Maybe they already have?

      • chuckula
      • 7 years ago

      [quote<]Hmmm I can only imagine the overclocking limitations that could be imposed by this.[/quote<] Indeed, those limits do exist in your imagination. From what we've seen in advance, overclocking Haswell shouldn't bee too much of an issue, and even the non-K series parts now have some interesting OC options. [quote<]There is nothing stopping Intel from simply disallowing users from over-volting their processors so they can't OC at all though.[/quote<] You know Bensam123, have you ever heard of the Strawman argument? There's a new online encyclopedia called "wikipedia" that has a whole article about it! Here's a link: [url<]http://en.wikipedia.org/wiki/Straw_man[/url<] In case the article is too involved, here's a hint: Making "strawman" arguments isn't a good thing.

        • Bensam123
        • 7 years ago

        There being no ‘currently’ imposed limitation with Haswell isn’t the same as Intel having no plans or no ability to implement OCing limitations based on voltage regulation. You’ve never heard of big companies doing something like that though, have you?

        Nothing you said refutes the point that Intel is now fully capable of limiting OCing with on board voltage regulation, all you stated is they haven’t [i<]yet[/i<] and then tried to make some sort of underhanded jab at me as if it related to my original point. More then half your response was an insult (which seems to be back where we started some months ago), especially if you consider the second even more childish post. In order to argue a hypothetical, you have to at least talk on terms of hypotheticals... Although I heard a statistic once that like one in five people are unable to think in terms of hypotheticals, so I suppose it doesn't surprise me that they elude you.

          • chuckula
          • 7 years ago

          [quote<] Nothing you said refutes the point that Intel is now fully capable of limiting OCing with on board voltage regulation, [/quote<] blah blah blah... Nothing you said proves that AMD isn't fully capable of cutting out overclocking of its own chips with zero need to resort to an on-board voltage regulator. (and if you disagree, you obviously think that AMD is full of incompetent morons who couldn't pull that off... you don't think AMD is incompetent do you?) Your problem is that you really really want there to be some big evil Intel conspiracy full of Machiavellian plots that you (and only you!) can see so clearly while everyone else is just too stupid to comprehend your true brilliance. Excuse me while I pause to play my violin for you... OK done. The problem is that if Intel really really wanted to do exactly what you said they are doing (but also not doing), there are about a million easier ways to accomplish the goal than going out and wasting all the development time and engineering work to put the voltage regulator on the chip! You want to stop overclocking? Put in a trivially simple logic block that prevents the chip from running if the detected clock speed exceeds a narrow variance range. DONE! Instead, you think that Intel is full of evil plotters who come up with these grandiose conspiracies to screw you over while wasting billions of dollars in R&D to come to the exact same results that they could achieve far more quickly and cheaply using dozens of other routes. So basically, they are out to steal your money so they can blow the money coming up with ways to steal your money instead of making any profit. Bensam123 logic: YOUR WINNER!

            • Bensam123
            • 7 years ago

            You’re once again trying to turn this into a AMD Vs Intel thing. It’s entirely possible for me to point out a negative point of what Intel is doing that could very likely become a reality without supporting AMD.

            I believe that locking down products and having less options available for consumers is in and of itself BAD. It’s like new cars being made so owners can’t actually work on them and have to take them to the shop instead. If AMD was doing it, then I would also say the same thing. It doesn’t matter whose name you attach to it.

            You’re a smart guy. You at no point said Intel isn’t capable of locking the last bit of OCing down on it’s chips with a on board voltage regulator.

            Once again you spent most of your post trying to insult me instead of actually address the point in a meaningful manner. I’m glad you know some pop-culture psychology terms, go watch more bones.

            I will let you in on my super secret world domination plot behind my posts here at TR though, I actually make posts like these to generate intellectual conversations based off possible hypotheticals. Because anyone can look at AMDs financial earnings and go ‘lol they bad, they going to go bankrupt’. It’s actually fun stretching your brain a bit and not just kicking the guy that’s down cause he’s a easy target so you can get a lot of thumb ups.

            I don’t know dude, maybe there is some sort of PR problem with treating your customers like crap, so it’s not in your best interest to go outright taking things away from them when there is still competition around. Company image and all that jazz. Not everything is so black and white. Two birds, one stone.

            • NeelyCam
            • 7 years ago

            [quote<][b<]very likely[/b<] become a reality[/quote<] Seriously?! If Intel wanted to do that, it would've been done a long time ago already. [quote<]You're a smart guy. You at no point said Intel isn't capable of locking the last bit of OCing down on it's chips with a on board voltage regulator.[/quote<] True, but you missed his main point, which is that it would be much easier for Intel lock down OC by doing something else. Them putting the VRM on-chip just for this would be horribly expensive and stupid, and it's similarly ridiculous for you to suggest that that's what they are doing. In fact, do you remember how Ivy Bridge overclocking was locked down by the very narrow frequency range for bclk? That's all Intel would need to do to lock down OC - disable the multiplier and fix bclk. In contrast, and completely against your conspiracy theory, with Haswell Intel is actually putting effort into making overclocking [b<][i<]easier[/i<][/b<] by enabling various bclk settings. I would even argue that the on-chip VRM makes OC potential better, as the reduced droops don't BSOD your system as easily, allowing lower average voltage to keep the temps down So sorry; I understand your desire to bring up hypotheticals to try to make Intel look evil, but in this case there is plenty of evidence that make your hypotheticals sound just pointless. [quote<]I don't know dude, maybe there is some sort of PR problem with treating your customers like crap, so it's not in your best interest to go outright taking things away from them when there is still competition around.[/quote<] With Haswell, Intel is not "taking things away" regarding overclocking, but instead giving more. So I don't know what you're talking about.. it's almost like you're blindly believing your conspiracy theory, and ignoring evidence to the contrary... If this is how you want to stretch your brain, by all means do it. Try not to sprain anything.

            • Bensam123
            • 7 years ago

            The last paragraph takes into account all the former points and why Intel doesn’t just blatantly do something like lock down frequency.

            You can have more then one reason for doing something Neely. Two birds, one stone. I never said the sole reason for voltage regulation on the chip was JUST to lock down OCing, rather they’re now completely capable of doing it. Like I said, not everything is so black and white.

            Considering how vile Chuckula is, I had expected more out of him as far as intelligence goes, but it seems the limitation of his wits are strawmen and insults. He couldn’t even properly formulate a counterpoint and about the best thing we got out of him was ‘if they were going to do it, they would’ve done it already’, which is a logical fallacy in and of itself. Times, situations, people change… It’s not always the same. Options that were once thought of as a bad idea eventually become good ideas… Escalation is a good example of this.

            I really need to stop responding to people who simply skim posts though, it’s not worth my time.

      • chuckula
      • 7 years ago

      Old McBensam had a troll.
      Ee-[b<][i<]aye[/i<][/b<] Ee-[b<][i<]aye[/i<][/b<] Ee-[b<][i<]aye[/i<][/b<] Trololololo... And in the troll he made up crap about microtransactions Ee-[b<][i<]aye[/i<][/b<] Ee-[b<][i<]aye[/i<][/b<] Ee-[b<][i<]aye[/i<][/b<] Trololololo... With a turboboost charge here And a strawman-post charge there Here a charge, there a charge, everywhere a micro-charge Old McBensam had a troll. Ee-[b<][i<]aye[/i<][/b<] Ee-[b<][i<]aye[/i<][/b<] Ee-[b<][i<]aye[/i<][/b<] TROLLL... Oh wait? AMD is coming out with BGA parts? INTEL STOLE THE IDEA FROM AMD! AMD doesn't make microtransactions, AMD MAKES FREEDOM!!

    • cynan
    • 7 years ago

    There goes Intel slowly but surely chipping away at the margins of motherboard makers. Hopefully, for the consumer, this integration of on-die components isn’t used as an excuse to charge more for their CPUs.

    Perhaps it also stealthily chips away at the value potential of AMD CPUs as respective motherboards will have the added cost of VRMs.

      • curtisb
      • 7 years ago

      How do you figure this chips away at their margins? They’ll charge the same for the motherboards but have to put less components on them…seems like it’ll help their margins to me.

        • jdevers
        • 7 years ago

        It will chip away at the margins because it makes the high margin enthusiast level motherboards harder to justify. Even now, the only real difference between a cheap Z77 and an expensive Z77 motherboard is often just the voltage regulation, color scheme and maybe the number of PCI-E slots. Remove the voltage regulation from consideration and it gets harder and harder to justify a board that costs 2-3x as much with very little in the “extras” department.

          • NeelyCam
          • 7 years ago

          [quote<]It will chip away at the margins because it makes the high margin enthusiast level motherboards harder to justify.[/quote<] Maybe that's a good thing.

          • curtisb
          • 7 years ago

          I don’t see those going anywhere. You’d be surprised what people will pay just to get the custom color scheme. I’m not gonna lie…it was a big part of the decision for my last motherboard purchase (Asus Maximus V Gene). I wasn’t going to sacrifice quality, but I wanted what I wanted.

          Those high-end enthusiast level boards aren’t as big of a part of the market as you might think.

    • willmore
    • 7 years ago

    Does anyone know if the final Haswell implementation will be MCM like the slides show or do they plan on pulling it on-die? Seems a waste of 22nm die area when most of the circuitry won’t scale.

      • chuckula
      • 7 years ago

      MCM. There’s really no crazy advantage to having the VRMs on-die.

        • NeelyCam
        • 7 years ago

        Look at the first slide on the EXPreview link.. it almost looks like the VRM [i<]is[/i<] integrated on-die (instead of MCM), and only the inductors are off-chip (package traces? thin-film deposited somewhere..?). I guess the on-die integration might be the reason for that super-low ripple..

          • willmore
          • 7 years ago

          The control logic and power transistors benefit nothing from being made in 22nm. For that and power reasons, it makes more sense to stick them off chip.

          The super low ripple is due to the massive number of phrases and the high switching frequency.

      • smilingcrow
      • 7 years ago

      Only the Ultrabook chips are MCM I think!

        • willmore
        • 7 years ago

        For the eDRAM, they are going MCM, so it’s not really all that strange to think the whole product line might be in *some* way. I guess we’ll now in a few months.

    • ronch
    • 7 years ago

    I’m waiting for them to also integrate the monitor, keyboard and mouse inside the chip package.

      • DeadOfKnight
      • 7 years ago

      Sound system, too!

      • Wildchild
      • 7 years ago

      I’m waiting for the day I can plug my computer into my brain and just upload information to it.

    • NeelyCam
    • 7 years ago

    The integration of the inductance is particularly interesting

      • just brew it!
      • 7 years ago

      Agreed. This was probably the biggest stumbling block to integrated VRM.

      • sluggo
      • 7 years ago

      The key to the entire scheme. The driver of power switching evolution is higher switching speed, because the faster you switch the smaller the magnetics can be. Switching at 100MHz like this makes the inductors reeeealy small. I doubt that this sort of switching speed is even possible outside of a fully integrated chip solution as the trace and leadframe inductance would be too high.

        • willmore
        • 7 years ago

        It also makes the switching losses a lot higher.

          • sluggo
          • 7 years ago

          Only if you make no changes to the actual switch, which is not the case here.

    • chuckula
    • 7 years ago

    OT but Haswell related, some leaked benchmarks show the HD 4600 vs. the 8670D in the upcoming Richland GPUs (the 8670D is the top-end IGP for Richland): [url<]http://wccftech.com/intel-hd-4600-graphics/[/url<] Spoiler alert: The 8670D in a desktop Richland is about 17.4% faster than the desktop HD-4600 in the 4670K. (8670D: P1508, HD-4600: P1285) BUT: Here's another data point from another article on the same website looking at the HD-4600 in a *mobile* Haswell part (the 47 Watt TDP 4800MQ): [url<]http://wccftech.com/intel-core-i7-4800mq-mobile-haswell-processor-hd-4600-gpu-benchmarked/[/url<] Note that in this benchmark, the slightly higher-clocked HD-4600 GPU gets a score of P1351. Now the delta is down to 11.6% for a mobile part where the IGP is actually important... and this is the supposedly crippled and useless HD-4600 IGP that is going against the top-end AMD IGP.... Neely may not have to cry into his espresso quite as much as he had thought. [P.S. --> Before anyone says that the HD-4600 can only do 3.7 FPS in Unigine Heaven, go look more carefully at the settings that were used for the run including a rather bizzare and non-standard 1920x1067 resolution, 8x anti-aliasing, Ultra quality and extreme Tesselation. With those settings, even my desktop discrete GTX 560 would likely be under 10 FPS, so don't run around saying that Trinity is 10x faster at those settings without proof.]

      • NAG3LT
      • 7 years ago

      But HD 4600 isn’t the top IGP offering in Haswell CPUs. Iris Pro 5200 is Intel’s graphics top dog.

      [url<]https://techreport.com/news/24749/intel-dubs-haswell-igp-iris-promises-2-3x-performance-increase[/url<]

        • chuckula
        • 7 years ago

        Yes, exactly.

          • NeelyCam
          • 7 years ago

          HD4600 might not be HD5X00, but neither is Richland equal to Kaveri.

          In any case, this is cool and all, but what I’m personally more interested in is how the Ultrabook Haswell parts compare to Kabini/Kaveri in a similar power envelope.

            • chuckula
            • 7 years ago

            And neither is Kaveri coming out at the same time as Haswell. I fully expect a “shipping for revenue” announcement in Q4 and actual parts available for sale in Feb/March of next year.

            Broadwell’s GPU won’t be 2x faster than Haswell’s, but it will be better. We are still very much in the dark as to how much faster *mobile* Kaveri’s GPU is going to be compared to what has been available in the past. I’m fully convinced that Kaveri + desktop power envelope will have a very nice IGP, but desktop ain’t where I’m interested in using IGPs.

            • NeelyCam
            • 7 years ago

            My point was that even if Haswell miraculously claims the IGP crown in desktop-class TDP, Kaveri will come and reclaim the win in six months or so (my guess is holiday sales instead of late Q1).

            BTW, do we know for sure if Kaveri is GloFo, or TSMC?

            [quote<]desktop ain't where I'm interested in using IGPs.[/quote<] I'm with you on this one. With a desktop I could always put in a kickass (NVidia) discrete card if I wanted to. Laptop/ultrabook market is where IGPs really matter.

      • ronch
      • 7 years ago

      Yeah, those new Intel IGPs look nice, until you move the mouse and those VLC control buttons come up.

        • chuckula
        • 7 years ago

        YOU PAID INTEL SHILL! HOW DARE YOU SUGGEST THAT INTEL’S SUBSTANDARD GRAPHICS ARE CAPABLE OF RUNNING A VIDEO-PLAYBACK PROGRAM! 😉

      • chuckula
      • 7 years ago

      Yeah just to confirm my earlier suspicion, running with those insane settings in a (more standard) 1920×1080 window with a GTX 560, the most updated Nvidia Linux drivers, and desktop compositing disabled to help performance yields: Max FPS: 10.3, Min FPS: 1.7, Average FPS: 2.9 and a score of 73… all in all the Haswell GPU ain’t that shabby in comparison.

        • NeelyCam
        • 7 years ago

        Yeah but frame times and bad drivers and 23.976. Also something about IQ. And and.. YEAH – AMD just feels faster!!1! And it streams better

      • BestJinjo
      • 7 years ago

      What you posted is worthless data. 3DMark as usual is a complete waste of time when comparing different GPUs. It stopped to matter after 2006.

      [url<]http://www.chinadiy.com.cn/html/24/n-9024-11.html[/url<] GT630 loses to HD4600 in 3DMark but in real world games GT630 absolutely creams the HD4600. Trying to extrapolate real world gaming performance between HD8670D and HD4600 based on 3dMark scores is a waste of time for everyone. People should know better in 2013. Get benchmarks in games where HD4600 is a hair within HD8670D or GT640 and we'll start talking. Intel can throw 3dMark scores all they want for marketing slides. You also lost track that Haswell's GPU competitor is not Richland but Kaveri as Haswell will be around for at least 12 months before Broadwell. Richland is not going to be going against Haswell for 12 months.

        • NeelyCam
        • 7 years ago

        [quote<] You also lost track that Haswell's GPU competitor is not Richland but Kaveri as Haswell will be around for at least 12 months before Broadwell. Richland is not going to be going against Haswell for 12 months.[/quote<] With your logic, Kaveri's GPU competitor is not Haswell but Broadwell. Here's how it goes (assuming AMD doesn't miss their launch dates... which seems to happen all to often): Haswell vs. Richland: 6 months Haswell vs. Kaveri: 6 months Broadwell vs. Kaveri: 6 months Broadwell vs. ?????: 6 months Skylake vs. ?????: ? months BTW, what's after Kaveri? I couldn't find a roadmap.. is AMD killing the whole desktop APU thing after Kaveri?

          • BestJinjo
          • 7 years ago

          It makes no difference anyway as his data is still worthless.

          That is 3dmark scores. We know from recent review of 4770K (retail part) how well it does in 3dmark and how bad it does (vs GT630 that performs “worse” in 3dmark) in actual games, I expect Richland 6700(65W) to just walk away from HD4600 when it comes to actual fps in games .GT630 is 30-130% faster (in actual games,AA or no AA) than HD4600 @ 1350Mhz with DDR3-2600(!). Imagine HD4600 using normal 1600/1866Mhz memory.

          I expect a lot more informed statements from members of a tech site than comparisons of real world gaming performance based on 3dMark scores.

          Regarding Kaveri, it should be coming Q3/Q4 2013 at the latest. Richland is just a stop-gap. Either way, based on real world gaming performance, Trinity is 80-100% faster than HD4000. HD4600 won’t even come close to that and Richland is even faster than Trinity.

    • Waco
    • 7 years ago

    Integrated voltage regulation will make high end boards essentially useless…unless mobo makers try to wring money from people by artificially segmenting their lineups. You want two PCIe slots? Buy the $250 Ultra-Awesome FTW EPIC Level 5 OC Edition SuperClocker board!

      • DeadOfKnight
      • 7 years ago

      I don’t see why this is a bad thing. If it takes something so simple to make mobo VRMs useless then obviously mobo VRMs didn’t add enough value to begin with. I’m all about honesty to the consumer.

      • spuppy
      • 7 years ago

      Motherboards still need to supply this with 2.5v or whatever it will be, so they will still need a good quality VRM from 12v to whatever input these use.

        • Waco
        • 7 years ago

        Yes, but since they don’t have to be variable, the DC-DC circuit will be a lot more simple. All of the control that we’ve become accustomed to on a board level will move almost entirely to the CPU.

          • just brew it!
          • 7 years ago

          Everything’s highly integrated these days though. The programmability of the VRM is already built in to the VRM controller chip, so you’re probably not saving much by eliminating the voltage control.

    • Arclight
    • 7 years ago

    [quote=”Geoff Gasior”<]That said, overclockers may hit thermal limits sooner than they did with Ivy-based chips, and motherboard makers will need to find new ways to hype their boards' power regulation circuitry.[/quote<] Haswell doesn't seem that appealing on the desktop, except maybe for those that want really small mini ITX builds and no dedicated video card. I'm certain that when people will complain about the small performance improvements, Intel will just tell them to buy last gen $1000 CPUs. The only way i'd buy Haswell would be if they SKUs are cheaper compared to IB, which is unlikely.

      • nanoflower
      • 7 years ago

      I don’t know about that. Have you read some of the reports of the overclocking people are getting with Haswell? They are much better than with Ivy Bridge (even with only air cooling) so it seems like integrating voltage regulation hasn’t hurt the overclocking potential of Haswell. At least not in regards to what was possible with Ivy Bridge. It’s certainly possible that Haswell would have done even better if they hadn’t integrated voltage regulation, but then the same thing might be said for them integrating the GPU.

        • Arclight
        • 7 years ago

        I’m not ready to pass judgement yet on the OCing prowess of Haswell. Not until official reviews are out. We have no idea why they haven’t pushed the clocks above 5 GHz or the voltage above 0.9 V.

        My suspicion is that the retail chips won’t do as good as the engineering samples simply because they might have lower [s<]TJ max[/s<] Tcase. Even so, 5 GHz is not that impressive. With SB it was achievable what? 2 years ago? Meanwhile IB could do 4,8 GHz iirc.

          • chuckula
          • 7 years ago

          Lower TJ-Max? NOBODY HAS LOWER PRICES THAN TJ-MAX!

          • NeelyCam
          • 7 years ago

          [quote<]My suspicion is that the retail chips won't do as good as the engineering samples simply because they might have lower TJ max. [/quote<] Always the pessimist... Do you have [i<]any[/i<] reason to believe that Haswell 22nm transistors have a lower TJ-Maxx than Ivy Bridge? I mean, it's the same process..

      • curtisb
      • 7 years ago

      Repost from Friday’s Shortbread:

      [quote<]You're only looking at it from one angle. While there is a slight performance bump, Haswell is all about being more power efficient at the same or better performance levels. While it's true that most of the power efficiency focus has been and will be on battery life, think about enterprise or university/college environments where there are potentially hundreds or thousands of PCs running all day, every day. So at my next batch of PC refreshes you're telling me that at the worst I'll keep the same performance and use less power? Sign me up. There are more ways than just battery life that power efficiencies can be realized. Think about it all the way to how much the AC has to run in the building due to those machines warming it up quicker. Less power usually means less heat. Does this mean I'm going to go out and replace all of my PCs at once with Haswell based PCs? Nope, but the next refresh batch will certainly be Haswell based. And to use even less power they likely won't have the discreet graphics we normally add, and will also likely all have SSDs.[/quote<]

        • Prion
        • 7 years ago

        If you have discreet graphics, do you need to turn on VGA palette snooping to see anything?

    • Mr. Eco
    • 7 years ago

    A long line of PC evolution is about to end in few years, when the whole computer will be a small cube.
    First the I/O cards got integrated, then the video cards and memory boards got created as separate items, much later sound cards were integrated to the mainboard, then memory controller got into the north bridge, then the video cards for the IGP solutions, and now yet another part gets integrated into the CPU.

      • riviera74
      • 7 years ago

      Yes, integration seems bad for us DIY builders. But PCIe still exists for us to get better performance out of daughtercards such as video cards and sound cards.

      Now for notebooks (especially ultrabooks), Haswell is a godsend.

        • spugm1r3
        • 7 years ago

        [quote<]Yes, integration seems bad for us DIY builders.[/quote<] Integration merely means how you DIY will change. Integration tends to lead towards innovation; putting a stereo in a car didn't suddenly end DIY home stereo. As more parts get integrated in the CPU, perhaps innovation will move towards peripherals, which really haven't innovated much in the last thirty or so years. I'm actually looking forward to the point when the ergonomics of the pc become more important than internals.

        • ShadowEyez
        • 7 years ago

        They already are small cubes…

Pin It on Pinterest

Share This