Kaby Lake Pentiums gain Hyper-Threading and lose ECC support

Up until now, the biggest difference between Intel's budget Pentium chips and its Core i3 offerings has been the presence of Hyper-Threading. The line between Pentiums and Core i3s is blurring with the release of Kaby Lake Pentium chips, though. Eagle-eyed Hardware.info forum user J_C_W spotted the HT-enabled Pentiums in Intel's ARK and tipped off his forum mates to the change. Here's a list of the SMT-enabled Pentiums arriving with Kaby Lake:

Aside from the clock speed boosts common to many Kaby parts and the addition of Hyper-Threading, there's little difference between these chips and their Skylake counterparts. In its endless march to segment its product lineup, however, Intel has decided to cut out ECC RAM support from these CPUs. That could annoy folks who relied on Pentiums to power home servers or NAS boxes.

The lower-end G4560 and G4560T models come with Intel's HD Graphics 610 IGP, while the more expensive chips come with the HD Graphics 630 processor. The only difference we can spot based on the information in ARK is a 50-MHz clock speed deficit in the HD Graphics 610 when compared to the 630's 1.10 GHz max frequency, though it's possible that there are also differences in shader count in play.

We published a story about rumors of a multiplier-unlocked Core i3 back in November. That story mentioned the possibility that HT-enabled Pentiums and Core i3s with Intel's Turbo Boost could appear. Thus far, no Turbo Boosting Core i3 CPUs have been released, but the unlocked Core i3-7350K is real.  The addition of Hyper-Threading could make the Kaby Lake Pentiums a bit more attractive for usage in an entry-level gaming system, and it's entirely possible they may eventually find their way into some budget builds in our System Guide.

Comments closed
    • Shambles
    • 3 years ago

    Bummer from my point of view. The most useful aspect of the small Intel chips was the ability to use them in a home NAS with ECC support. Here’s to hoping that Ryzen maintains ECC on their CPU lineup.

    • allanmac
    • 3 years ago

    The Core processors get AVX2 while the Pentium & Celeron processors do not.

    • maxxcool
    • 3 years ago

    Dammit … I *JUST* built my media PC to replace my Athlon-640 that ran 7+ years of 1080 goodness. A HT Pentium would have been nice.

    • Chrispy_
    • 3 years ago

    Especially when it came to laptop parts, the only tangible difference between an i3 and an i5 is turbo.

    i3’s are consumer chips, I doubt more than a teeny-tiny percentage of the i3’s target audience even know what ECC support even does.

    • marvelous
    • 3 years ago

    Why would anyone buy an i3 now?

      • flip-mode
      • 3 years ago

      [Deleted] Ah, I misunderstood you. Yes, good question.

      • Krogoth
      • 3 years ago

      More cache, Turbo-clocking and slightly beefer IGP.

        • tipoo
        • 3 years ago

        deleted

      • DPete27
      • 3 years ago

      Probably not for typical consumer use.

    • ronch
    • 3 years ago

    Looks to me Zen is gonna be great.

      • Flying Fox
      • 3 years ago

      Remains to be seen.

    • sophisticles
    • 3 years ago

    Personally I’d rather have a Pentium branded 8 core cpu, they can keep the HT and ECC support, just give an affordable octo core cpu for my VM’s and I’ll be happy.

      • jackbomb
      • 3 years ago

      If they did that, it would probably be based on the slow Atom cores and end up being slower than an i5 even in multithreaded workloads.

    • Eversor
    • 3 years ago

    If you do video encoding or scientific calculations, keep in mind this has no AVX.

    • juampa_valve_rde
    • 3 years ago

    Hooha, intel is trowing a bone to the low end… Mhhh they are worried, ain’t? The ‘rizon is getting darker

      • Magic Hate Ball
      • 3 years ago

      *redder.

      • jessterman21
      • 3 years ago

      *horyzen

    • Tristan
    • 3 years ago

    Meanwhile, AMD may disable HT on low priced RyZens

    [url<]https://twitter.com/CPCHardware/status/818932115270209537?ref_src=twsrc%5Etfw[/url<]

      • Eversor
      • 3 years ago

      The way both AMD and Intel are going with this, we’re not that far from x86 falling into irrelevance.

      Per core, per MHz performance of Apple’s A10 is higher than the i3-7100u on some benchmarks. At 125mm2, it is also not a big chip.

      Qualcomm is faster than Intel’s low power core (wish I had perf per watt data to look at) and is also getting into servers this year. They think they’re still playing a two player game, yet ARM micro architectures keep pushing it more and more.

        • tipoo
        • 3 years ago

        >Per core, per MHz performance of Apple’s A10 is higher than the i3-7100u on some benchmarks. At 125mm2, it is also not a big chip.

        Their chip design does really impress me, but one has to keep in mind, there’s no assurance this efficiency would keep up if trying to scale to a Core I series frequency, no doubt it’s designed around a certain (low) target clock and pipeline depths and stuff may limit its scale-up.

        I would however love to see what a future A10X could do actively cooled…

          • Eversor
          • 3 years ago

          True. It doesn’t matter though, it may not be Apple that grills x86.

          Qualcomm is getting desktop Windows 10 this year (with even proper x86 emulation). Say the software turns out OK and next year they take the upcoming server core and ship it out to desktops, then what? Nvidia already designs high performance ARM SoCs and will keep at it. Oh and they have experience with ARM drivers and will also want to sell you the CPU.

        • odizzido
        • 3 years ago

        My first question would be are those tasks hardware accelerated on the A10 and not on the i3?

          • Eversor
          • 3 years ago

          Afaik, no. You alreaday saw JavaScript benchmarks being very fast since before the A9. We don’t know much about the micro arch and part of the high performance there is due to Safari.

          Where, after the A8, numbers also started to make a show is in stuff like Geekbench v4, which tests a bunch of stuff and (from what I gathered by comparing benchs) reflects in real world code that doesn’t have arch specific optimizations like AVX or NEON. You can also get better multi core scaling in real world vs that but single core seems OK.

          Compound this with the fact that some laptops have soldered single channel DDR, it doesn’t look good if Apple starts making laptops with A10X.

          [url<]https://browser.primatelabs.com/v4/cpu/1566192[/url<] [url<]https://browser.primatelabs.com/v4/cpu/1459730[/url<] Keep in mind the small cores are still quite fast but Apple has no heterogenous scheduler like Qualcomm for the Kryo chips. So this thing could be even more impressive.

          • DPete27
          • 3 years ago

          I’m far from an ARM vs x86 pro, but my surface understanding is that ARM does well primarily because it has instruction sets tailored to it’s architecture and form factor (low power passively cooled). I just don’t see an ARM CPU running Win10 being orders of magnitude more efficient than an x86 CPU which have been continuously optimized for x86 use since the dawn of computing. Intel/AMD have enough money and ARM has been around long enough that if ARM had some magic sauce solution to x86 efficiency, they either would have already exploded into the x86 market, or Intel/AMD would have reverse engineered ARM and implemented those groundbreaking innovations in their own CPUs.

          (keep in mind, I’m not arguing whether or not x86 instruction set is the “best” for modern computing)

          I think of it in terms of Consoles vs Gaming PCs. Consoles do miraculous things with their anemic hardware thanks mostly to their being optimized “close to metal” for gaming first and foremost. They literally have API’s that were custom designed for or in conjunction with their hardware. Gaming PCs have to deal with a much broader usage model, hardware configurations, an pre-existing instruction sets. Some or none of which may be the most efficient choice for their gaming performance.

            • Eversor
            • 3 years ago

            On that subject, the main point is that contemporary x86 decodes into RISC like for execution (I think K7 was the 1st). That is one of the most power hungry parts of the chip. ARM inst are fixed and simpler to decode.

            I wasn’t saying ARM has it there for licensing, but they’re closer now to x86 perf. I was saying that the ISA is and Apple already has an ARMv8 implementation that is world class. Tegra X1 is similar to Apollo Lake but was out in 2015.

        • Anonymous Coward
        • 3 years ago

        [quote<]Per core, per MHz performance of Apple's A10 is higher than the i3-7100u on some benchmarks. At 125mm2, it is also not a big chip.[/quote<] Why don't they license that thing for use anywhere except mobile devices, make some money on the side. Sell it to Nintendo. 🙂 Maybe its too little profit for them to care.

        • Laykun
        • 3 years ago

        Oh man, did you hear? Nvidia’s Tegra 3 outperforms a Core 2 Duo T7200 in some benchmarks, crazy huh?

      • ozzuneoj
      • 3 years ago

      Can anyone translate and explain why these guys would know?

    • EndlessWaves
    • 3 years ago

    Given that the mobile Skylake pentiums had it I’m not sure how much spotting it needed. I certainly checked as soon as I saw that they were available to confirm it and thought no more of it.

    In terms of the graphics, it’s likely it corresponds exactly to the Skylake situation. Namely the HD 510 is Intel’s GT1 graphics which 12 EUs instead of the 24 EUs in GT2 (515, 520, 530), 48 in the GT3 (540, 550) and 72 in the GT4 (580).

    As it’s likely just disabled silicon they may have upped it to 18 as found in the HD 505 but I’d assume 12 until proven otherwise.

    • ozzuneoj
    • 3 years ago

    Even with the things they cut from these, Intel just made the only CPU that 90% of desktop users will need cost $64. Seriously, a 3.5Ghz Kaby Lake is going to have pretty substantial per-core performance, even with less cache. And the extra features and instructions aren’t going to be missed by anyone who would buy one of these without doing their research… no matter what enthusiasts will say.

    I’ve never heard any non-enthusiasts complain about a desktop i3 system being “slow”, even for light gaming, and this is basically an i3. There are still plenty of people out there plugging along with Phenoms, Core 2s, AMD APUs and pre-Sandy i5s without feeling the need to upgrade immediately… this will likely outpace most of them for $64. It’ll also outpace all but the most expensive laptop CPUs… for $64.

    For what its worth, there is a single Pentium G4600 result showing on Passmark right now, and it is a monster for only $75…
    [url<]http://www.cpubenchmark.net/compare.php?cmp%5B%5D=2617&cmp%5B%5D=2921&cmp%5B%5D=2578[/url<] Take this with a grain of salt... but the results show it being close to i3 6100 single threaded performance, beating it in multithreaded performance, beating an i5 6400's single thread performance by about 15%, and only trailing the multithreaded performance by about 16%... for a $75 CPU, that's really impressive. IMO, this is going to be the budget CPU to get, simply because it will be SO cheap and be enough for most tasks for most people. It should easily meet the "real" minimum requirements of most games.

      • Waco
      • 3 years ago

      This. It’s a steal at that cost for what is essentially a high-clocked i3 in all other aspects. The main achilles heel of the Pentium AE was the lack of hyperthreading…

      • Eversor
      • 3 years ago

      Agreed. For light office work machines, my pick is stil the Celeron G1840 for $30. The H110 motherboards also adds about $15 vs the H81 and DDR3 is enough but much cheaper due to less demand. The savings pay for an SSD or a proper PSU that won’t blow up your build.

        • rudimentary_lathe
        • 3 years ago

        The G1840 has no AES support, which would be an issue for an office that encrypts all their drives.

          • stdRaichu
          • 3 years ago

          A quick’n’ditry bench (and ) shows that a single core of my haswell xeon can do 8k blocks of AES-128-CBC at ~800MB/s with AES-NI enabled and ~370MB/s with AES-NI disabled, so I’m not sure if the lack of AES-NI will be a massive bottleneck even for heavy users of encryption.

      • Tom Yum
      • 3 years ago

      Checking Australian pricing, the G4560 is currently available for $79, the nearest i3 (7100) is $169, for which you get an extra 400MHz, or just over 10% extra performance (outside of AVX workloads which are rare for most people). Certainly know which one I’d recommend to my family and friends.

      Really puts the i3 range in a weird spot, AVX and ECC is just not worth paying double for 99% of people.

        • derFunkenstein
        • 3 years ago

        The Core i3 series will still be the best budget option for corporate desktops, because of that AVX advantage. My empoloyer’s policy is that every machine corporate-wide runs with some sort of disk encryption. It may not be a huge deal with spinning disks, but most entry-level corporate boxes from Dell and Lenovo generally have i3 or i5 CPUs anyway.

      • rudimentary_lathe
      • 3 years ago

      Yeah, they certainly seem to be undercutting the i3 with these chips. I’m curious as to why.

        • Tom Yum
        • 3 years ago

        If Cannon Lake had been imminent I would be tempted to think Intel was preparing to drop quad core into the i3 line, quad core with HT into i5 and 6 cores as standard into i7. But Cannon Lake is another 6-9 months away so that doesn’t work.

        Whatever the reason, Intel certainly isn’t the type of company to do this out of the goodness of their hearts. I wonder if its a move to make it difficult for AMD to flush the supply chain of existing AM3/FM2 FX/APU’s, no doubt both AMD and suppliers would like to get rid of them before Ryzen starts shipping on mass.

          • Waco
          • 3 years ago

          I’m hoping it’s because they have some info that Zen doesn’t suck…like at all.

      • mczak
      • 3 years ago

      I think though most people would really want the 75$ chip, not the 64$ one. Let’s face it, the cpu performance of the G4560, G4600 and G6420 is just about identical (that 5% clock difference between slowest and fastest isn’t going to do much…).

      However, there’s something to be said about twice as fast graphics (the HD 610 is GT1, HD 630 is GT2 – twice the number of shader units). Neither one is meant for gaming of course, but this can easily make a difference with non-demanding causal games, or other 3d apps. Except of course if you pair it with a discrete graphics card, then the difference won’t matter.

      I agree though the i3 look terrible. Their only real advantage over these Pentiums is AVX (and AVX 2.0). Chances are if you’re happy with a two core cpu, you’re probably not going to depend on AVX neither… Other than that, they just have a bit more cache and a bit higher clock, which might be good for a ~15% performance increase, but considering they cost twice as much that isn’t enough. If you need more cpu performance than what these Pentiums offer, you’re probably better off with a i5, though of course the price increase is quite a bit bigger again – but at least the cpu performance increase is more substantial (if your workloads are multithreaded).

      • ozzuneoj
      • 3 years ago

      I see that a couple more results for this CPU have been added, so the numbers are changing a bit. Its looking like the G4600 will perform very similarly to the i3-6100. Not bad at all for $75.

      Also, G4560 results are starting to show up as well…

      [url<]http://www.cpubenchmark.net/compare.php?cmp%5B%5D=2925&cmp%5B%5D=2617&cmp%5B%5D=2608[/url<] Not much farther behind the i3-6100 honestly. It should also walk all over an i7-6600u, which makes a $64 Pentium faster than what you'll find in a lot of $1000+ laptops. Desktops FTW? 🙂

    • DPete27
    • 3 years ago

    Oooh Oooh! Jeff! Can we get one of these in the i3-7350K review?? Pretty please!!!

      • EndlessWaves
      • 3 years ago

      This is a far more interesting chip, let’s scrap the 7350K review and just include those results in the G4600 review.

        • DPete27
        • 3 years ago

        While I agree that the 2c/4T Pentiums are going to embarrass the i3 lineup (G4620 is the exact same specs as the i3-6100), I’m equally interested in the i3-7350K and how it will compare to/in the i5 lineup. Granted, overclocking isn’t for everyone, but if you’re buying a 7350K then clearly it’s for you. I don’t see the 7350K being as much of a value disruption as the new Pentiums, but…

      • derFunkenstein
      • 3 years ago

      agreed, it’d be nice to see one of these in there since they’re so close to an i3.

    • Wirko
    • 3 years ago

    The i3 loses ECC support too, save for a couple “E” (“Exotic”) models.

    [url<]https://ark.intel.com/search/advanced?s=t&SocketsSupported=FCLGA1151&MarketSegment=DT&ECCMemory=true[/url<]

    • tipoo
    • 3 years ago

    Seems like an excellent trade-off! Looking at the frame times for a Pentium vs an otherwise equal i3, hyperthreading really made the difference and made i3s far more capable gaming chips. Some games also only boot if they detect 4 threads.

    • NTMBK
    • 3 years ago

    I wonder how these will measure up against an AMD APU?

      • DPete27
      • 3 years ago

      No contest. The G4620 has the same specs as the i3-6100, but suffice to set it equal to the i3-4360 in [url=https://techreport.com/review/26845/amd-a10-7800-processor-reviewed/5<]this TR review[/url<]

        • NTMBK
        • 3 years ago

        Oh no doubt the CPU side will totally thrash the AMD option! Kind of curious how the GPU compares.

          • DPete27
          • 3 years ago

          There are IGP benchmarks on Page 4 of that review. Guaranteed Intel HD630 IGP won’t hold a candle to the $120 A10-7800 IGP, but I’d bet its awful close to the similarly priced A6-7400K.

          • EndlessWaves
          • 3 years ago

          Depending on how much Intel’s drivers have improved it’s either within shouting distance of an A8-7600 or on par with it.

      • Concupiscence
      • 3 years ago

      Unless you’re flogging AES or AVX like mad, the Pentium will smash ’em.

    • colinstu12
    • 3 years ago

    Sounds like a good move to me. SMT will be a welcomed addition to these chips.

    If you wanted ECC on the cheap you’d be better off getting a last-gen used Xeon.

      • rudimentary_lathe
      • 3 years ago

      Any ideas on a cheap mobo to go along with it? Been eyeing a G4400, but the server-grade boards are like 4 times the price.

        • Flying Fox
        • 3 years ago

        Without ECC support on the G4400, what’s the point of still getting a server-grade board?

        Assuming you can still use the C232 chipset (or else you will need to wait for new boards), I can find boards <$200. Not exactly 4 times I suppose. But you have to be aware that “server/workstation grade” is not going to be dirt cheap.

          • rudimentary_lathe
          • 3 years ago

          The G4400 has ECC support.

    • Krogoth
    • 3 years ago

    This move was just to prevent people from making a DIY NAS box on the cheap.

      • MOSFET
      • 3 years ago

      If that’s the case, the move is actually very unimpressive.

      • DragonDaddyBear
      • 3 years ago

      So, you’re not impressed? You must not hold Intel stock 😛

        • MOSFET
        • 3 years ago

        Actually, the everyday performance of the G4620 looks to be right on track with the i3-6100, so it actually is pretty impressive. As far as Intel’s segmentation, ignoring the obvious motivator, greed, it’s just senseless to hack out more ECC from the already thin ECC CPU lineup. Not everyone can or will buy a Xeon, even if it comes strapped to an ITX board, so it’s probably just another opportunity for ARM eventually. (I own several Xeons for home lab use, so I’m not opposed to buying them, used, when necessary.)

      • CScottG
      • 3 years ago

      No. Because of course you can still get the Skylake version and a C232 m-board at a fairly low cost (..and because you really don’t care about a minor performance loss in this use-case).

      -the fakakda unbuffered ECC RAM on the other hand just seems to be escalating in price from about 6 months ago. (..I spent an extra $120 in December getting my 64 gig from what I would have spent purchasing the stuff around July.) Granted though, that I don’t see anyone getting 64 gig for NAS alone (..hell, 8 gig might be overkill).

        • Krogoth
        • 3 years ago

        Skylake-based i3s that have ECC enabled go for doubled the cost of the Pentiums though.

          • CScottG
          • 3 years ago

          ..but you can get Skylake Pentiums that have ECC support as well.

          [url<]http://www.newegg.com/Product/Product.aspx?Item=9SIA1UH3Z96302[/url<] [url<]https://ark.intel.com/products/88179/Intel-Pentium-Processor-G4400-3M-Cache-3_30-GHz[/url<] -and that's more than enough processor for most DIY NAS boxes.

            • Krogoth
            • 3 years ago

            Interesting, I thought that Intel removed ECC support on Skylake Pentiums. It also has VT-D and VT-X support too.

            That chip looks like a cheap solution for a low-end virtual box (driving legacy VMs).

            • CScottG
            • 3 years ago

            Hell, the Celerons even have both ECC and VT-D support..

            I know, odd but true:

            [url<]https://ark.intel.com/products/90741/Intel-Celeron-Processor-G3900-2M-Cache-2_80-GHz[/url<] [url<]http://www.grooves-inc.com/intel-celeron-g3900-prozessor-intel-hardware-electronic-pZZa1-2097676062.html?language=en&currency=USD&utm_source=froogle_us&utm_campaign=froogle_us&gclid=COq486vGudECFQSraQodfEEKNg[/url<]

        • Flying Fox
        • 3 years ago

        [quote<]I don't see anyone getting 64 gig for NAS alone (..hell, 8 gig might be overkill)[/quote<] Not if you are using ZFS with the more resilient redundancy setup, 16gigs seem to be the recommended minimum.

          • CScottG
          • 3 years ago

          Yup, it depends on the setup. Though I still don’t envision 64 gig for a home NAS (that only serves that function). Even 32 gig seems “indulgent”.

      • Flying Fox
      • 3 years ago

      I recently put into home “production” use a Pentium/i3 based File Server. I intentionally picked a workstation board in order to use ECC. No more cheap ECC file server with these CPUs. 🙁

      • derFunkenstein
      • 3 years ago

      yeah, like millions of people are building NAS systems with ECC memory.

        • just brew it!
        • 3 years ago

        They’re probably more concerned with the [i<]NAS vendors[/i<] doing it.

          • derFunkenstein
          • 3 years ago

          Possibly.

        • Krogoth
        • 3 years ago

        NAS hobbyist tend to be hold data integrity in high regard. ECC memory is one of the mainly steps to help ensure this.

          • derFunkenstein
          • 3 years ago

          but are there millions? It’s probably more like what jbi said, it’s a vendor.

      • just brew it!
      • 3 years ago

      If Zen is any good and AM4 continues to support ECC like the AM3+ platform did, there may be a reasonable alternative. Hopefully AMD has their USB3 platform issues sorted, since that’s a “nice to have” on a NAS box.

Pin It on Pinterest

Share This