We examined Intel’s flagship Lunar Lake Extremely 9 and it is slower than an Extremely 7 in multi-threaded work in the one laptop computer the place it is at the moment on the market [Updated]
[ad_1]
Whenever you purchase by way of hyperlinks on our articles, Future and its syndication companions might earn a fee.
We examined the brand new Asus Zenbook S14 with the Lunar Lake Core Extremely 9 288V processor, and surprisingly, we discovered that the Core Extremely 9 is slower than the Extremely 7 mannequin in threaded workloads, a situation that Intel says Asus will quickly handle with a BIOS repair. Our findings are fascinating as a result of Intel initially shipped the Asus Zenbook S14 to reviewers for its Lunar Lake launch, however the US press acquired the mannequin with the lower-tier Core Extremely 7 258V processor, not the flagship Core Extremely 9 288V processor.
We additionally observed a brand new sort of conduct with the Extremely 200V collection — the Skymont E-cores run at increased frequencies than the Lion Cove P-cores throughout closely threaded work, indicating an distinctive quantity of efficiency and effectivity from the brand new E-cores.
Intel gave us the possibility to check the Core Extremely 9-equipped Zenbook S14, which was initially solely despatched to a couple reviewers in Europe as a result of the design just isn’t at the moment out there for buy within the US. The truth is, in line with our checks, that is the only real Core Extremely 9-powered pocket book out there in the marketplace. Nonetheless, our findings point out that Intel and Asus nonetheless have work to do on the firmware facet to ship the total efficiency of the Core Extremely 9.
The desk above reveals the considerably comparable specs for the 2 processors. With the identical variety of cores and the identical cache capacities, the one actual variations between the 2 chips boil right down to clock speeds and energy limits. The Core Extremely 9 has a 300 MHz increased P-core enhance clock and 1 GHz increased base (P- and E-cores), with an ever-so-slight 100 MHz bump in GPU frequency. Each chips have a peak turbo energy of 37W, however the Extremely 9’s base energy (PL1) weighs in at 30W with a 17W minimal, whereas the Core 7 has a 17W base with an 8W minimal.
The Asus Zenbook S14 could be run at totally different energy ranges, as listed within the desk beneath, by biking by way of totally different fan settings within the MyAsus utility. These settings not solely regulate the fan profile but in addition regulate the configurable TDP. The identical energy settings apply to each the Extremely 9 and Extremely 7, thus eradicating TDP limits as a differentiator between the chips in our testing. Meaning it is right down to the variations in clock speeds to ship tangible benefits for the Extremely 9, and regardless of the comparatively tame deltas, certainly not ought to the Extremely 9 be slower than the Extremely 7.
You’ll be able to see our Zenbook S14 review with the Core 7 258V here, and other than the totally different chips, our Zenbook pattern with the Core Extremely 9 has the identical specs — proper right down to the cooling subsystem. Naturally, that led us to suspect that cooling may have been the wrongdoer to the unexpectedly low efficiency, however as we’ll present shortly, that isn’t the case.
As you may see within the album above, we examined each the Core Extremely 7 and 9 in all 4 modes to measure efficiency variations throughout the total TDP vary. Efficiency is usually inside our expectations for single-threaded work — the Core 9 beats or matches the Core 7 in many of the single-threaded Cinebench and Geekbench 6 checks, however the Core 9 does path barely on the 17W Whisper threshold.
For the Whisper mode, HWiNFO lists the Core Extremely 9’s cTDP setting for PL1 energy at 13W, which is not according to the official specs of 17W. Nonetheless, the Extremely 7 is dialed in on the right 17W PL1. This is able to appear to be a minor repair within the BIOS, however bigger variations reside within the threaded workloads in any respect different energy ranges, and people different energy modes are tuned appropriately for Extremely 9.
Efficiency within the threaded workloads stands out as notably inexplicable. The Extremely 7 took the lead throughout the total vary of Cinebench, Geekbench, and Handbrake benchmarks in any respect energy ranges. These deltas endured regardless of a number of retests below various circumstances, so we contacted Intel, which confirmed our findings.
The Extremely 7 is just ~2 to five.3% sooner in Cinebench than the Extremely 9, 1.4% sooner in Geekbench, and three.5% sooner in HandBrake. These aren’t big variations, however logic dictates it must be slower, not sooner. Moreover, if thermal limitations hamstring the Extremely 9 (i.e., the cooler wasn’t sturdy sufficient), we’d anticipate it to, at a minimal, ship the identical efficiency because the Extremely 7 as a result of they might each encounter the same warmth limitation. Nonetheless, that wasn’t the case.
We charted the common efficient clock speeds, chip temperature, and energy draw throughout two runs of the Cinebench multi-threaded checks. We plot the P-core clock speeds in blue and the E-core clocks in black, with these values listed on the left-hand chart axis. We even have the facility consumption in inexperienced and temperature in crimson, with these values listed on the right-hand chart axis. We recorded these with the processors in ‘Efficiency’ mode.
We’ve got a chart to cowl similar runs for each the Extremely 9 and Extremely 7. We will see that the Extremely 9 peaks at round 80C throughout the multi-threaded checks, so thermals are usually not the limiting issue (the chip has a 100C restrict).
For each chips, the E-cores run at increased clock speeds than the P-cores throughout the Multi-threaded Cinebench checks. This isn’t the case for any of Intel’s earlier hybrid x86 processors with each P- and E-cores for both desktop or cell processors — the P-cores have at all times run at increased clock speeds throughout closely threaded workloads (we’ll display that with Meteor Lake subsequent).
Intel instructed us it is a new, meant conduct to extract the most effective combination of energy effectivity and efficiency in threaded workloads. That’s clearly now supplied by the Skymont E-core structure (affectionately nicknamed ‘Chad’mont by fanatics attributable to its large generational positive aspects) as an alternative of the Lion Cove P-cores — at the very least at these energy ranges. Our laptop computer reviewer, Andrew Freedman, initially observed this new conduct, and I meant to create a voltage/frequency curve to focus on the distinction in energy effectivity. Nonetheless, for the reason that chip isn’t working appropriately, we’ll have to attend for a brand new BIOS.
Given the chips’ specs, we anticipate each the P-cores and E-Cores to run sooner for the Extremely 9 288V, but the Extremely 9’s P-cores run at a mean of three GHz throughout the take a look at, the identical because the Extremely 7’s. Moreover, the Extremely 7’s E-cores run at an efficient 3.5 GHz, 100 MHz sooner than the Extremely 9, but the Extremely 9 pulls extra energy (22W) on common than the Extremely 7 (19W) throughout this portion of the take a look at.
For now, Asus prospects must await a repair. Intel supplied us with the next assertion; “Your outcomes look proper for the present BIOS model. We anticipate Asus to quickly launch a brand new BIOS to handle small efficiency inversions seen in some multi-threaded workloads. You might even see further enhancements in single-threaded workloads.”
Nonetheless, Intel hasn’t given us a timeline for the repair, and Asus has but to answer our queries.
We ran the identical checks on a Meteor Lake Core Extremely 7 155H, additionally utilizing an Asus Zenbook 14-inch chassis. Right here’s an fascinating tidbit that isn’t well-known: Meteor Lake has two sorts of P-cores (extra beneath) and two sorts of E-cores. This implies we had plenty of cores to plot. We plotted the identical collection of checks however put energy and temps on a separate chart to enhance readability.
The massive takeaway right here is that each sorts of P-cores run at increased frequencies than the E-cores throughout the multi-threaded Cinebench take a look at, exhibiting that Lunar Lake does, actually, take a a lot totally different method to delivering power-efficient efficiency in threaded workloads, an method enabled by the energy of the Skymont E-core microarchitecture. Meteor Lake, utilizing the identical ‘Efficiency’ fan profile within the MyAsus app, additionally exhibited a lot increased common energy use in our multi-threaded testing — although it was additionally sooner and has way more whole cores.
A little bit of background about Meteor Lake’s two sorts of P-cores: It isn’t unusual for just a few cores to run sooner on any given chip, however these cores are sometimes sooner just by advantage of the silicon lottery — some cores are naturally able to sooner speeds as a result of variability of the semiconductor fabrication course of, and the sooner cores are recognized within the binning course of after which designated because the ‘favored’ cores.
With Meteor Lake, Intel deliberately designed two cores with sooner, leakier transistors and assigned them 8 Vt factors (threshold voltage factors). The rest of the cores used slower transistors with much less leakage (that means they’re extra energy environment friendly) and solely have 6 Vt factors (the ‘Intel 4‘ course of node enabled the extra two Vt points). The extra two voltage thresholds on the sooner P-cores allow them to hit increased clock charges, thus creating two cores which are particularly designed to be sooner than the opposite cores regardless of using the identical microarchitecture. Intel didn’t publicly announce this design resolution, however the firm confirmed it with me earlier this yr. We have listed the cores as Excessive Efficiency (HP) and Low Efficiency (LP) P-cores within the above chart.
It is well-known that Meteor Lake additionally has two sorts of E-cores — the usual E-cores that reside on the CPU tile and two slower Low Efficiency (LP) E-cores that reside on the SoC tile. Intel intends to make use of the LP E-cores as a lot as attainable to save lots of energy — even to run your complete OS throughout mild workloads like video playback, which permits the chip to close off your complete CPU tile. Nonetheless, the dearth of an L3 cache pressured these cores to incessantly entry principal reminiscence, and lighting up the reminiscence controllers and accessing reminiscence is dear from an influence standpoint, thus offsetting a lot of the facility effectivity gained from utilizing the LP E-cores.
To repair this, Intel added a Reminiscence Facet Cache to Lunar Lake, however that is actually simply analogous to the System Degree Cache (SLC) discovered on Arm chips. In both case, the ‘Reminiscence Facet Cache’ method is rather more efficient.
Conclusion
Intel’s Lunar Lake represents a dramatic step ahead on a number of fronts, particularly in energy effectivity and battery life, however we aren’t in a position to present the deep-dive particulars on lots of the architectural features of the Core Extremely 9 but as a result of we’re testing with a problematic BIOS. We’ll submit extra benchmarks after we get a brand new working BIOS. Naturally, we’ll additionally retest the Core Extremely 7 chips with the brand new BIOS to ensure there aren’t any regressions with these chips.
Neither Intel nor Asus have given us a timeline for the brand new BIOS replace. Given that there’s an obvious lack of Extremely 9 fashions at retail within the US, it’s logical to suppose we’ll see extra Extremely 9 laptops out there in the identical timeframe because the up to date BIOS.
[ad_2]
Source
