Rumor: Intel Ice Lake chip emerges in Sandra with Gen11 graphics

Since it's totally unmoderated, the SiSoft Sandra benchmark results database is kind of a mess to navigate. Even still, TR tipster SH SOTN spotted a result in the database that identifies itself as an "Intel Ice Lake Client Platform." The result is for Sandra's OpenCL-based General-Purpose Processing benchmark, and it identifies the graphics part in our purported processor as "Gen 11 LP" with 48 execution units running at 600 MHz.

With just these scraps of information to go by, there aren't a lot of hard conclusions we can draw, but we can perform a little guesswork. 48 execution units would make it a GT3 part by Gen9 standards, otherwise known as Iris Plus Graphics in the current Intel GPU lexicon. However, the listing does call the integrated graphics processor "Intel UHD Graphics," which might point to a boost in the execution-unit count for Intel's entry-level IGPs. That would be a logical step in light of AMD's Raven Ridge notebook processors and their powerful Vega onboard graphics.

The 600-MHz clock speed is interesting too, because it's 300 MHz faster than the base clock of most existing Intel graphics parts. It's far lower than most of those parts' boost clocks, though. Other results from Intel GPUs on this benchmark list their circa-1 GHz boost clocks. It's possible that this is just early silicon on early drivers with incomplete reporting of performance information.

One other detail of note is that this listing marks the GPU down for having 768 KB of L2 cache—another 50% increase over Gen9. If our assumptions do pan out, it looks like Ice Lake could have a nice little bump in graphics performance compared to Intel's current offerings. We won't know until these 10-nm+ processors begin showing up in the market, and given Intel's still-nascent efforts to bring 10-nm silicon to market, we could be waiting for some time.

Comments closed
    • DoomGuy64
    • 2 years ago

    So now we have 6 core cpus, full avx, and better integrated graphics. Almost as if Intel is making an effort to be competitive, instead of rehashing dated 4 core product segmented chips with new boards for every minor inconsequential update. Of course, the boards thing probably isn’t going to change, but at least we get cpu improvements, all while chuckula vehemently denies AMD has any influence on these competition driven cpu improvements in the comments.

      • chuckula
      • 2 years ago

      Yawn.

      Your pathetic ad-hominem drivel would have been more amusing if you had accused Intel of intentionally gimping its own IGP performance to maintain its monopoly.

      If you’re going to base your entire life on the philosophy of companies you don’t like intentionally gimping their own products, at least try to be consistent.

        • DoomGuy64
        • 2 years ago

        Hi there! It seems I have something pulling on my fishing line…

        Almost as if someone has a search algorithm running 24/7 on TR’s comment section for the sole purpose of anti-AMD shilling, which is either really psychotic, or someone who is being paid to write comments!

    • anotherengineer
    • 2 years ago

    Lakes have 4 feet of ice on them here.

      • Srsly_Bro
      • 2 years ago

      I live a few miles from Lake Michigan and I have no idea if the Great Lakes freeze over in the winter.

    • deruberhanyok
    • 2 years ago

    Would definitely appreciate Intel stepping up their integrated graphics game. If IGPs get to a point where they make 1080p gaming playable, the sting of inflated GPU prices won’t be quite so bad anymore.

      • Ummagumma
      • 2 years ago

      Intel could always respin some form of it’s speculative processing technologies for it’s IGPs….

      And perhaps it might have fewer security issues than those impacting it’s processors??

        • Srsly_Bro
        • 2 years ago

        Of all this time Intel has been messing around and you had the answers. No one at Intel thought of this solution and they are fools. You are so wise…

        Bro..

        Srsly….??

    • Waco
    • 2 years ago

    The biggest shocker here?

    Sandra still exists. HOW?

    • DavidC1
    • 2 years ago

    Interesting thing is the name is Gen 11 LP. The LP designation has been used on parts that go in Apollo Lake and Gemini Lake. It could be that Core M Icelake parts are getting Gen 11 LP. That would explain the clock.

    We can’t tell much from the L2 cache. Gen 9 GT2 has 768KB L3 cache, 512KB being a Data Cache. Sisoftware sometimes reports Gen 9 GT2 as having 512KB L2 cache.

    It means Intel iGPUs have L2 caches as well. I’m not sure whether its reporting the actual L2 cache size or its confusing L3 cache as L2. The latter could be true because 512KB and 768KB has been reported as L2 in Sisoftware database.

      • AnotherReader
      • 2 years ago

      I suspect it is the total L3 cache available as shared memory. For Gen9, [url=https://software.intel.com/sites/default/files/managed/c5/9a/The-Compute-Architecture-of-Intel-Processor-Graphics-Gen9-v1d0.pdf<]according to Intel[/url<], the L3 per slice is 768 KB. Typically, 512 KB is allocated as a data cache for compute applications. The rest is split between system buffers for fixed-function pipelines, and shared local memory. For compute application, that suggests that the typical application has 256 KB of shared memory. Since each slice is usually 24 EUs in Gen 9, 48 EUs correspons to 2 slices and 512 KB of shared memory.

        • DavidC1
        • 2 years ago

        Gen 11 allows for maximum of 8 subslices per slice. Gen 9 only allows 3. Gen 10 in Cannonlake allows maximum of 5 subslices per slice. That’s why the leaks have shown 40EUs for Gen 10.

        That means if EUs/subslice hasn’t changed, we’re looking at potentially a maximum of 64. Though it could be just as well 6 EUs/subslice maximum x 8 subslice for total of 48. So 48-64EUs become the baseline(GT2) in Gen 11, with 1 slice.

          • AnotherReader
          • 2 years ago

          Thanks. Do you have any links handy for Gen 11 documentation?

            • DavidC1
            • 2 years ago

            Sure.

            [url<]https://patchwork.kernel.org/patch/10209709/[/url<]

    • Spunjji
    • 2 years ago

    Could be running wider for power savings and to use up die space needed for pads. That would explain the 600Mhz clock (slightly higher and more stable performance).

      • the
      • 2 years ago

      I’ll second this idea. Having high clocks is difficult to have in a portable design when voltage has to scale upward with it. For power efficiency, this is probably the better trade off.

        • NoOne ButMe
        • 2 years ago

        How much more can they charge OEMs for it, and how much more does it cost Intel?

        I’m expecting the answer to be “OEMs will pay 0 dollars more, with increased costs of OVER 0 dollars”.

        Although as a SKU to appease Apple, given losing them to Ryzen APUs* would probably be really bad.

        Intel has been fine on GPU power for mobile for a while. The performance (or rather drivers?) is the larger issue.

        *i do know AMD no longer calls them APUs.

          • Srsly_Bro
          • 2 years ago

          than?

            • NoOne ButMe
            • 2 years ago

            How much more does it cost them [Intel]?

            Basically, consumers mostly buy on price.
            OEMs, except maybe Apple, who can pass it on, won’t want to pay more.

            But, maybe Intel’s 10nm does somehow actually lower the costs enough they can do it. I still think they would use that cost reduction for high margins, or drop prices to put pressure on AMD…

            If it’s to compete with AMD on iGPUs… all the extra cost it takes should have been put into driver teams.

      • NTMBK
      • 2 years ago

      Or it’s a work around for 10nm heat density problems.

        • Spunjji
        • 2 years ago

        Equally plausible!

    • tipoo
    • 2 years ago

    If someone was too lazy to compare, what sort of uplift over a gen 9 48EU part is that then?

      • chuckula
      • 2 years ago

      From what I’ve seen that’s a low score.
      Like a *really* low score, even for an IGP and especially for one that’s as big as “48EUs” if true.

      So either Ice Lake really does suck that bad or it’s just not optimized to run that test.

        • tipoo
        • 2 years ago

        Are you comparing with non-eDRAM GT3?

      • RAGEPRO
      • 2 years ago

      As far as the benchmark goes? Don’t pay attention to it. On this test Sandra doesn’t differentiate between CPU-only, GPU-only, or CPU+GPU, or CPU+GPUs, or CPUs+GPUs, or whatever. As a result it’s almost impossible to tell what you’re actually comparing against. There are a bunch of machines that list their processor type as Intel HD 4600 (Haswell GT2) and yet score three times higher. Who knows.

        • tipoo
        • 2 years ago

        Ah I see. I was curious since things like the MX150 seem to put the Iris 650 in comparable packages in a very tight squeeze, for not much space or power savings.

    • Neutronbeam
    • 2 years ago

    If they’ve got something that goes to 11 then they’ve finally got my attention.

      • DrCR
      • 2 years ago

      This post now appropriately thumbed up to 11.

        • frogg
        • 2 years ago

        just downgraded Neutronbeam to make you right

    • chuckula
    • 2 years ago

    [quote<]That would be a logical step in light of AMD's Raven Ridge notebook processors and their powerful Vega onboard graphics.[/quote<] Meh. We've been illogical in the face of superior AMD integrated graphics for YEARS! Don't expect us to change now. On a more technical note, I'm curious what that "768KB L2" actually means since it is listed in conjunction with the graphics and not regular CPU cores. GPUs generally don't have huge caches and I'm curious if that number implies that Intel has slapped an L2 cache into the GPU or if it's just an artifact of the CPU cores. Then again a bunch of the information could just be wrong or jumbled too.

      • RAGEPRO
      • 2 years ago

      I spent a lot of time looking into that. In that same spot Sandra reports an L2 size of 256KB of Haswell parts and 512 KB on Coffee Lake parts, but those CPUs also have 256KB per core. … I’m not really sure what that means. I’m kinda betting on “wrong or jumbled” myself.

        • chuckula
        • 2 years ago

        Thanks for the research. We’ll know more when and if Intel ever manages to actually ship these things.

          • willmore
          • 2 years ago

          This says it’s the cache for the IGP:
          [url<]https://wccftech.com/intel-skylake-gen9-graphics-architecture-explained-gt2-24-eus-gt3-48-eus-gt4e-72-eus/[/url<]

Pin It on Pinterest

Share This