“What are the advantages of the new 12-core Zen 5/5c CPU in this Asus laptop model over prior models with 8-core Zen 4 CPUs like the 7940HS or 8940HS or 8945HS?”
Answer This Question
1-10 of 10 Answers
Correction - the name of the built-in iGPU graphics engine for the Ryzen AI 9 HX 370 (part of the new Strix Point series of laptop CPUs) is the 890M, not 980M. Per an early performance review the 890M is 60% faster than the graphics iGPU incorporated into the recently announced Qualcom laptop chips, as well as being 40% graphics-faster than AMD’s own last-gen 780M iGPU incorporated into Ryzen 4 8-core APUs such as 7940HS (Phoenix) or 8940HS or 8945HS (both Hawk Point). Also note that AMD still also makes a series of laptop APUs called by the nickname Fire range that typically come with 16 or 12 cores. The new Ryzen 9 HX 370 4+8 = 12 core part seems to be very competitive with the 12-core Fire Range parts, according to very early leaked performance reports on HX 370 engineering samples. But the 3D-Vcache flavor of 16-core Fire Range chips still holds the crown among all laptop chips, at least for now. Hopefully this will help Best Buy shoppers figure out where the HX 370 fits in AMD’s lineup of laptop chips.
Sorry, there was a problem. Please try again later.From Ryzen 9 7940HS to Ryzen 9 8945HS (sometimes shown as 8945H or 8940HS due to AMD renaming it to 8945HS just before release), the 8945HS refresh features an improved memory controller leading to ~10% better performance. It also has a bigger XDNA NPU, going from 10 TOPS to 16 TOPS. From Ryzen 9 8945HS to Ryzen AI 9 370, the 370 has 4x Zen5 + 8x Zen5c cores (vs 8x Zen4 cores), a 16 CU RDNA3.5 iGPU Radeon 890M (vs. 12 CU RDNA3 iGPU Radeon 780M), and a 40 TOPS XDNA2 NPU (vs. 16 TOPS XDNA NPU). The two big advantages of the 370 are expected to be higher CPU performance (~14% single-thread ~45% multi-thread) and much better performance at low power. The Zen5c cores are denser Zen5 cores which run at a moderately reduced frequency and power. While you can still run CoPilot without a 40 TOPS NPU, NPUs are much more power efficient at running AI tasks than CPUs and GPUs are. The Radeon 890M iGPU is expected to perform on par with a RTX 3050 laptop dGPU at max TDP (around a 47.7% improvement over the Radeon 780M at max TDP). Leaked benchmarks show that the 890M at 15W still out-performs the 780M at 54W by ~22.5% (part of that may be due to the presence of Zen5c cores). Note that those are leaked 890M performance for early samples. Production silicon may perform even better. In short, the Ryzen AI 9 370 will give you a much better gaming experience on battery power.
Sorry, there was a problem. Please try again later.Asus please also feel free to comment, but I’ll also weigh in since I’ve fairly thoroughly researched this new APU/CPU announced at June 2024 Comdex and available in July, namely the Ryzen AI 9 HX 370 that has a total of 12 cores. This is a Gen 5 CPU but it has 4 regular Gen 5 cores plus 8 Gen 5c cores and all 12 are double-threaded so it can run 2*12 = 24 threads simultaneously. This is AMD’s first CPU with (sorta) nonuniform cores, but you should not think of them as P and E cores (as in Intel). Rather, think of the 4 Zen 5 cores as being Performance cores that are 15% faster than Gen 4 cores such as the ones in the older HS series chips such as the 7940HS or 8940 HS or 8945HS 8-core CPUs (APUs actually). Think of the 8 Zen 5c cores in this new HX 370 CPU as being new miniaturized modernized 5th generation AMD Performance cores that just run at a 20-25% slower clock rate then the larger new Zen 5 big cores. As such, the new 8 Zen 5c cores give you almost the equivalent of the 8 Zen 4 cores of a 7940HS (for instance) only about 5-10% slower. But on top of that the new Zen 5 big cores essentially give you the equivalent of an extra (1 + 0.15) * 4 Zen 4 cores due to the 15% boost in Instructions per clock of Zen 5 over Zen 4. That’s 4.6 Zen 4 core equivalents. Add 8 times (between 0.9 to 0.95) Zen 4 core equivalents = 7.2 to 7.6 for the new Zen 5c cores. Grand total that’s about 11.8 to 12.2 Zen 4 core equivalents - call it 12 - good enough for napkin math. So on the high end, you might get as much as 50% more Power out of the new HX 370 CPU versus a 7940HS CPU. A more conservative estimate might be a 30% bump. By Power I mean how much computing you can do with all cores busy. On the other hand, the bump in Single Core Speed averages 15% per AMD over a broad range of application types. Depending on app type, Speed bump over 7940HS ranges between +5% to +50%. I read that the 980M iGPU in this new APU is about 40% faster than that of the 7940 iGPU. And the AI engine NPU is anout 50 TOPS compared to 12 TOPS for the 7940HS and 16 TOPS for the 2nd gen 8940 HS and 8945HS.
Sorry, there was a problem. Please try again later.(Zen 5 Laptop and Desktop Performance Update Part 7 (2024 Labor Day) - Max-Core Power for most AMD CPUs - Gen 4/5 Desktop vs Laptop vs Threadripper) Max-Core Power performance data is in for new AMD desktops, and for comparable Gen 4 to Gen 5 models, Max-Core Power is up about +7% to +7.5% Gen 5 over Gen 4 desktop CPUs. This is quite close to my earlier +8% guestimate. The fact that this was nowhere near AMD +16% marketing propaganda caused quite a hatestorm in press and reviewers alike. They didn’t emphasize enough, though, the simple cause of AMD shooting themselves in the foot: simple cherrypicking of test results (which I somehow smelled in advance). Actually, I’m quite pleased with Gen 5 AMD performance - it’s quite near the 8% I hoped for, and I, for one, am buying one. Still TBD is whether I buy a desktop or laptop (Ryzen 9 AI HX 370) variant. Thus, to help me decide, I’ve completely updated my table of AMD-CPU Max-Core Power Performance - listed here (and this time exhaustively covering laptop and desktop and Threadripper CPUs): R5 8640HS = 61530; R5 8500G = 68380; R5 8645HS = 71041; R5 7640HS = 72069; R5 8840HS = 79156; R5 8600G = 80652; R7 8840U = 84453; R5 7645HX = 87033; R7 5700X3D = 90572; R7 8845HS = 92228; R7 7840HS = 93043; R5 7600X = 94609; R5 7600X3D = (TBD September per Moore’s Law is Dead); R9 8945HS = 95227; R7 5800X3D = 97154; R9 7940HS = 97480; R5 9600X = 101831; R7 8700G = 102856; AI 9 365 = 104242; R7 7745HX = 108857; R7 7700 = 115724; R7 7800X3D = 119328; R7 7700X = 120105; AI 9 HX 370 = 121131; R7 9700X = 128552; R9 7845HX = 162792; Threadripper Pro 7945 WX = 177275; R9 7900X3D = 182330; R9 7900X = 185145; Threadripper Pro 5955WX = 185889; R9 7940HX = 194421; R9 9900X = 198903; R9 7945HX = 199251; R9 7945HX3D = 216843; Threadripper Pro 7955WX = 218353; R9 7950X = 229791; R9 7950X3D = 234902; R9 9950X = 246973; Threadripper 3970X = 265660; Threadripper Pro 3975WX = 270714; Threadripper Pro 5965WX = 281075; Threadripper Pro 5975WX = 336064; Threadripper Pro 7965WX = 337754; Threadripper 7960X = 341661; Threadripper 3990X = 410198; Threadripper 7970X = 421584; Threadripper Pro 3995WX = 448607; Threadripper Pro 5995WX = 512978; Threadripper 7980X = 737393; Threadripper Pro 7985WX = 752653; Threadripper Pro 7995WX = 1000000 = basis of comparison = most All-Core Powerful AMD CPU mid-2024 (although some Gen 5 EPYC servers, yet to be deployed, will no doubt be even more powerful).
Sorry, there was a problem. Please try again later.(Performance update part 6 - update for AI 9 365 and 7945HXHX3D) Some minimal amount of data is out such as to make estimates of All-Core Max Power for the little brother CPU. The Ryzen AI 9 365 is at 100444 versus 119620 for the Ryzen AI 9 HX 370. The ratio is 0.83969 or in round numbers you could say that the little brother part has about 84% of the compure power of the AI HX 370. Or equivalently that the 365 that loses two of the 370’s 8 small cores (but keeps its 4 big cores for a total of 10 cores) has about 16% less compute power. However that’s still 6.3% more compute power than the reference Ryzen 9 7940HS (and a similar amount more than the top-end Intel Core Ultra 9 185H. I also had a typo for the Ryzen 9 7945HX3D in the prior update - recall that the 3D-Vcache gives it considerably more effective compute power for things like gaming and creating, and this metric puts an appropriate focus on that. I relist all the top-end Ryzen parts with 8 or more cores Threadripper Normalized All-Core Compute Power: Ryzen 7 8840HS = 86838; Ryzen 7 8845HS = 89605; Ryzen 7 7840HS = 90047; Ryzen 9 8945HS = 92152; Ryzen 9 7940HS = 94464; Ryzen AI 9 365 = 100444; Ryzen 7 7745HX = 105971; Ryzen AI 9 HX 370 = 119620; Ryzen 9 7845HX = 159565; Ryzen 9 7940HX = 188757; Ryzen 9 7945HX = 194856; Ryzen 9 7945HX3D = 212668.
Sorry, there was a problem. Please try again later.(Performance update part 5 - more accurate Normalized Max-Power metrics) in Part 4 was listed Max-Power (that is, max compute power) metrics (normalized to the fastest Threadripper workstation CPU chip that exists) for comparison to the HX 370 used in this ProArt P16 model. Part 5 updates these statistics mostly for accuracy purposes. First of all, we got a couple more data points for the HX 370 and that broader sample results in about a 1% lower score - so might as well use that more conservative score, at least until there are a suitably large enough sample size. Also, I noticed some mistakes mostly in the Intel laptop chips included for comparison (the fault issue was, iPhone calculator sometimes completely ignores you when you click Memory+), so I fixed the Intel comparison numbers and double-checked all stats. Also, this time, for comparison I only included Intel Core Ultra laptops as well as various 8/12/16-core AMD laptops (all desktop chips are gone under the idea that the customer is shopping for a laptop, and all Apple M-Series CPUs are gone under the idea that they are mostly too slow but too expensive if fast enough, and all Qualcomm CPUs have been removed under the idea that they are way too slow compared to the HX 370). So here are the Threadripper-Normalized Max-Power metrics for a bunch of laptop CPU chips (this time I’m only listing max runnable threads and omit core count): Core Ultra 5 125H (18) = 71601, Core Ultra 5 135H (18) = 73291, Core Ultra 7 155H (22) = 82715, Core Ultra 7 165H (22) = 86739, Ryzen 7 8840HS (16) = 86838, Ryzen 7 8845Hs = 89605, Ryzen 7 7840HS (16) = 90047, Ryzen 9 8945HS (16) = 92153, Ryzen 9 7940HS (16) = 94464, Core Ultra 9 185H (22) = 94870, Ryzen 7 7745HX (16) = 105971, Ryzen AI 9 370 (24) = 119620, Ryzen 9 7845HX (24) = 159565, Ryzen 9 7940HX (32) = 188757, Ryzen 9 7945HX (32) = 194856, Ryzen 9 7945HX3D (32) = 194856. Note that the Max-Power score of the original AMD 8-core Gen 4 CPU, namely the Ryzen 9 7940HS (94464), nearly ties the current top-of-the-line Core Ultra 9 185H (94870). But the score of all the low-end 8-core Gen 4 Ryzen HS series exceeds the score of all the low-end Intel Ultra CPUs (that is Meteor Lake generation). But the total Normalized Max-Power of the Gen 5 Ryzen AI 9 370 is about 26.5% higher than the “standard” Gen 4 Ryzen 9 7940HS. [ Recall that in terms of raw Single-Core CPU speed, the HX 370 is about 8% faster. ] If you are perhaps upgrading from a Ryzen 7 8840HS (the slowest 8-core Gen 4 Ryzen) to the HX 370, then you should experience a 37% increase in Max-Core Power. It turns out that my earlier guestimate of a 25-35% bump in power (for the HX 370 was pretty much in the ballpark - but the exact number mostly depends on which (of a plethora now) of Gen 4 8-core Ryzens you are comparing too (note: I noticed that the commonality among the weaker Gen 4 8-core Ryzens is a reduced power profile, so obviously AMD made some variants of the standard-bearer Ryzen 9 7940HS to suit thin-and-light laptops). Note that the only laptop CPU chips that beat the Gen 5 Ryzen AI 9 HX 370 in terms of Normalized Max-Core Power are also AMD CPUs - all from the HX series. [ A note for the confused: after making a big hoopla about its new (at the time) laptop CPU numbering scheme, it turns out that the HX versus HS suffix is the most important for Gen 4, and then AMD went AI-crazy and threw the scheme out the window with Gen 5. ] Finally, here’s an approximate cheat sheet for you with the AI HX 370. Very loosely speaking, in terms of maximal usable compute power, instead of thinking of the HX 370 as 4x Gen 5 (big) cores + 8x Gen 5c (little) cores, you can approximately think of it as 10x Zen 5 (big) cores plus a little bit more. I personally do not like the fact that AMD went to a big.little architecture, but I do like the 26.5-37% bump in maximum available compute power. I guess we gotta accept the annoying with the good.
Sorry, there was a problem. Please try again later.(Performance update part 4 - Threadripper Normalized Max-Core Power) When looking for a fast laptop (as I am, and perhaps you too) the most significant metric is generally Max-Core Power = available compute power with all cores blazing. Often an analogous metric called something like Multithreaded Performance is used for that, but there are too many tests and metrics for that, plus I want a metric that not only measures normal factors of compute power, but that also puts significant weight on factors that make a computer good for both gaming and creating. So I made one, and the first question was, “what units should it be in?” I took a page from middle-age folks who needed units for physical power, and realized that the best choice is relative to a well-known powerful thing - so they defined a horsepower = the power of a horse. Later on, metric came along, so we defined the Watt using horsepower plus a conversion factor. Starting similarly, we define Max-Core Power (2024 version) relative to the all-cores-blazing compute power of the fastest commonly used computer workstation CPU chip, namely the Threadripper 7995WX, which has 96 cores that can run 192 threads (and whose CPU alone will set you back $9999 with a lightly configured workstation probably costing north of $20K). We arbitrarily assign the Max-Core Power of the AMD 7995WX to be one million units. I worked out a method to estimate (well I suppose more like guestimate really) other CPUs’ Max-Core Power relative to that fastest Threadripper that exists in 2024. I’ll list some CPUs below. Unless noted as a desktop in parens, all are laptop CPUs. Also noted are number of cores/threads supported. And if a big.little architecture, big cores are listed first with little cores after a plus sign. Here we go in descending order, skipping Threadrippers since BB doesn’t sell them: Ryzen 9 7945HX3D (16/32) = 230872, Ryzen 9 7950X3D (16/32 desktop) = 230735, Ryzen 9 7950X (16/32 desktop) = 224702, Ryzen 9 7945HX (16/32) = 194856, Ryzen 9 7940HX (16/32) = 188647, Ryzen 9 7900X (12/24 desktop) = 181179, Ryzen 9 7900X3D (12/24 deskrop) = 178409, Apple M2 Ultra (16+8) = 164304, Ryzen 9 7845HX (12/24) = 159660, Apple M3 Max (12+4) = 150755, Apple M1 Ultra (16+4) = 130859, Ryzen AI 9 HX 370 (4/8+8/16) = 121700, Ryzen 7 7700X (8/16) = 116631, Ryzen 7 7800X3D (8/16) = 116486. We’ll pause our list here, just to note that the Ryzen AI 9 HX 370 is actually predicted to be stronger than the two most popular AMD desktop CPU chips including the 7800X3D that is a very strong gaming CPU. Also note that only a combination of the most powerful and most recent Apple CPUs can match or top the Ryzen AI 9 HX 370 CPU (albeit at a much steeper price). We continue our list with more CPUs, all of which are weaker than the AI HX 370 in terms of All-Core Power: Intel Core Ultra 7 155H (6/12+10) = 112867, Apple M3 Mac (10+4) = 110579, Ryzen 7 7745HX (8/16) = 105964, Ryzen 7 5700X3D (8/16 desktop) = 101135, Ryzen 7 8700G (8/16 desktop) = 99818, Ryzen 7 5700X3D (8/16 desktop) = 94674, Intel Core Ultra 7 155H (6+10) = 93538, Ryzen 9 8945HS (8/16) = 92153, Ryzen 5 7600X (6/12 desktop) = 91997, Ryzen 7 7840HS (8/16) = 90039, Ryzen 7 5800X (8/16 desktop) = 89949, Ryzen 7 8845HS (8/16) = 89653, Intel Core Ultra 9 185H (6/12+10) = 88747. I’ll pause here again just to note the inversion in Max-Core Power order among the Core Ultra models listed above. I’ll have to do some work later just to make sure I didn’t make an arithmetic mistake. That would be weird if the top 185H model is less powerful than several lower priced models, but I have seen such inversions before with Intel. Let’s list a few more CPUs even though the rest are far below the Ryzen AI 9 HX 379 in terms of All-Core Power as normalized the top Threadripper workstation: Ryzen 7 5700X (8/16 desktop) = 86823, Ryzen 5 7645HX (6/12) = 84433, Qualcomm Snapdragon X Elite X1E-84-100 (12) = 82823, Apple M4 iPad (4+6) = 81063, Quallcomm Snapdragon X Elite X1E-80-100 (12) = 80135. We’ve gone far enough to show that the top two Snapdragon X Elites have about the same All-Core Power as Apple’s latest iPad, and are about 2/3 the performance of the AMD Ryzen AI 9 370 CPU. The iPad is nice but still an iPad, and the Quallcom X Elites are nice if your primary requirement is to run on battery 20 hours per day. When Apple intriduces the faster/more-powerful M4 versions in computer form, they might compete better with AMD, but not now when Apple is too pricey for some.
Sorry, there was a problem. Please try again later.Performance update part 3 - the big.little elephant in the room) The salient point is that the Ryzen AI 9 HX 370 CPU is AMD’s first foray into using a big.little architecture. For non computer-scientists, that’s the technical term for when a computer CPU maker decides to put (on the same CPU chip) two (or more) different types (in various quantities) of CPU core. It’s called big.little (sometimes stylized as BIG.little) because the little cores often use less die space than the BIG cores, and thus you can fit more of them on a die. In a few cases big.little (also called a hybrid architecture) can be useful, but sometimes they can be a PITA. For instance, since 12th gen Intel famously uses (not always but usually) a big.little architecture where they call their cores Performance (big) and Efficient (little) [and also two more Very Low Power Efficient in Meteor Lake laptops only]. Performance cores are hyperthreaded (each core can keep track of and sort-of run two threads at once) whereas Efficient cores are not. I’ll let you in on a little secret - the main reason Intel went to big,little is that it fell behind AMD in terms of how many (Performant, in Intel terminology) cores it could fit on a CPU chip. This made it look bad next to AMD’s larger number of (Performant, hyperthreaded) cores. So instead Intel went to a small number of big Performant cores plus a bunch of little Efficient cores. That way the total number of Intel cores (and threads that can run at once) kept increasing generation after generation until Intel can currently advertise that it has more cores than AMD. The average computer/laptop consumer knows none of this and simply figures “more is better” and so Intel maintained market share (well it lost a little bit) while AMD continued to lead the technology race. Well AMD got frustrated in losing market share (of less computer savvy users) to Intel via the numbers game of there being more cores if you have both big and little cores, so in essence AMD threw in the towel and finally decided “we can play the big.little game too.” So the Ryzen AI 9 HX 370 is the first AMD (sort of) big.little architecture CPU chip. But what do I mean by “sort of” here? Well first we should say that technically speaking, yes it is indeed a big.little architecture. But it is in some sense different than Intel’s approach, because in AMD’s case the big and little cores are architecturally identical, and the main difference between the two is that the little cores are run at a slightly lower clock rate than the big cores. But in Intel’s case, the Performance cores actually have a dirrerent architectural design than the Efficient cores - for instance Performance cores (currently) are hyperthreaded whereas Efficient cores are monothreaded (thus saving die space). AMD strongly emphasizes that both their new big cores and their new little cores are both hyperthreaded (meaning both big and little cores support persistent context of two threads at once and possibly even share microcode execution of two threads at once (not sure, I’m not an AMD microarchitecture expert)). Actually, AMD marketing so stresses the identical architecture of new big and little cores that it’s almost as if AMD has a case of “big.little embarrassment complex” (a term I Just invented - hopefully the psychologists will add it to their diagnostic manual soon). For instance, AMD calls their new cores Primary and Secondary (as opposed to Performance and Efficient) so as to definitively distinguish themselves from Intel. So actually AMD Primary quite literally means big, and AMD Secondary quite literally means little. One could have alternatively named the AMD big.little cores Fast and Slow - but of course no vendor is going to call some of its cores “Slow.” If forced to use Intel terminology, AMD would probably state that it has big Primary Performant cores plus little Secondary Performant cores. I think the reason for AMD having such a huge case of big.little embarrassment complex stems from AMD having bucked the trend toward big.little for so long. For many years AMD seemed to espouse the tenet that “having just one kind of core is just better.” So now their formal position seems to be “well we still only have just one kind of core, it’s just that some of those cores happen to be bigger than the others.” In fact, AMD’s paranoia seems to be so complete that in their product spec sheet for the Ryzen AI 9 HX 370 it pretty much appears that nowhere thereni do they actually admit that it is a big.little architecture CPU -thus the “(sort of)” in (sort of) big.little. What they do state is that their cores (all of them, lumping big and little cores into one big pile) have a base clock speed of 2.0 GHz and a boost clock speed of 5.1 GHz. Now technically speaking they’re not lying to us, invoking the fact that all these AMD cores have identical architectures. But on the other hand it seems like they’re kinda telling a fib - since we know that they’re using a (sort of) big.little architecture, and given one of their chips plus a scalpel we could prove it. If they were more honest, their spec sheet might say something like Primary Cores: base clock speed of (we refuse to tell you) and boost clock speed of 5.1 GHz; Secondary Cores: base clock speed of 2.0 GHz and boost clock speed of (we refuse to tell you). It appears that AMD’s big.little embarrassment complex is so strong that they appear to prefer to carry on the illusion that they “don’t have a big.little architecture” at the cost of not giving us complete performance specs. I, for one, would like to know how low Primary cores will clock down to under a full all-core load, and how high Secondary cores will clock up to under very light loads. Hopefully once the info embargo is lifted (July 28), some testers will be able to answer these questions by testing the chips with all Secondary cores disabled and then all Primary cores disabled. One good thing is that whereas Linux testers have reported difficulties with Intel’s nonuniform big.little architecture (like in drivers), that should not happen in AMD’s big.little since all cores have the identical core architecture with only a small speed difference. So other than the fact that AMD needs a shrink, what does the Ryzen AI HX 370’s (sort of) big.little architecture mean for prfformance? Probably not much for most users. For the average user running normal applications, my expectation is that the Ryzen AI HX 370 CPU with 4 Primary cores plus 8 Secondary cores should “feel” just like a (hypothetical) last-generation Ryzen laptop CPU that just has 10 Performance cores (co-opting Intel nomenclature) rather than the 8 (Performance) cores that real last-gen Ryzen laptop CPUs actually had. This implies at least a 25% bump in total Multicore Performance (or Max-Core Power as I sometimes like to call it since it represents total available compute power with all cores blazing). It is still unclear whether or not the architechtural improvements will add an additional 8% improvement on top of that (as opposed to the known 8% IPC uplift being prebuilt into the above-estimated 25% bump in Max-Core Power. Doing the math, we can guestimate a 25% to 35% increase in Max-Core Power for the Ryzen AI HX 370 over the baseline (8-core) Ryzen 9 7940HS chip (and its siblings). Of course that doesn’t count the approximate 40% increase in iGPU speed and increase to a 50 TOPS NPU from a 12 TOPS NPU (on the 7940HS) or a 16 TOPS NPU (on its 8000 series siblings). Are there any other performance implications of big.little? Hopefully not (except for people like me who still wish all the cores were the same speed). One thing that I worry about just a little bit, but hopefully will not be a problem, is the possibility that games might stutter when scheduling of their processes gets tossed back and forth between Primary and Secondsry cores. Hopefully there will not be any actual stuttering - but rather at worst just a minor increase in the statistical variance of frame times.
Sorry, there was a problem. Please try again later.(Performance update single threaded part 2 - vs alternate laptop brands) In comparing laptop single-core speed we showed using the Passmark Single-Core stats that the AMD Ryzen AI 9 HX 370 has the fastest single-core speed of all AMD laptop CPUs and even beats the most popular desktop gaming CPU, the Ryzen 9 7800X3D. What about other vendors’ laptop CPUs single-core speeds? Among 4 models of ARM CPUs, the fastest Qualcomm X Elite 84-100 model scores 3966 in Single-Core Passmark versus 4213 for Ryzen AI 9 HX 370. More importantly, the total multithreaded power summed across all cores is 5/8 as large for the top Qualcomm X Elite CPU than for the Ryzen AI 9 HX 370. In spite of both CPUs having 12 cores, the Qualcomm cores are just 37.5% slower on average. The fastest Qualcomm 12-core CPU is also multithreaded-slower than the 8-core Ryzen 9 7940HS (and similar 894?HS) chips that the AI 370 CPU replaces. The only down side of the Ryzen AI 9 HX 370 is: it’s a mouthful to say. Also, most brands are selling you slower Qualcomm CPUs or maybe non-Elite variants, and the prices seem steep for the slow ARM CPU you get (not to mention some of your programs may not work, or may be slowed down by instruction-set emulation). Choose your own adventure but I’m avoiding Qualcomm cuz I want a fast laptop. Apple laptops or iMacs containing M-series CPUs are a feasible alternative, but their main problem is that they are overly expensive. As examples I looked up the Single-Core Passmark score for the beefiest Mac chip of each generation (minus M4 gen which is only on iMac yet and not on any MacBook/iMac), to compare to the 4213 score of the Ryzen AI 9 HX 370: Apple M1 Max 10 Core = 3837, Apple M2 Max 12 Core = 4146, and Apple M3 Max 16 Core = 4780. Only the top-end M3 Max 16 Core beats the 4213 AI 370 Single-Core performance (by 13.5%). It also beats the AI 370 Multicore Passmark performance by 7.7%. All other Apple laptop CPUs of any series (M1/M2/M3) and any other level (Regular or Pro or older Ultra but not Max) is multithreaded-slower as well as singlethreaded-slower than the Ryzen AI 370 CPU. If you seek a fast-CPU laptop, at this point there is only one Apple MacBook I can recommend and it is very pricey - perhaps up to 2x the price of the equivalent Asus ProArt P16 or Asus ROG Zephyrus G16. In fact at this point in time I suggest that fast-laptop buyers select from laptops containing either an AMD Ryzen AI 9 HX 370 (or maybe 365), or an AMD Ryzen 8?4?HX(3D), or an AMD Ryzen 7?4?HS CPU; or else a top-end MacBook if you can afford it; or else if you are an Intel fan then just wait for next-gen Lunar Lake laptops to arrive later this year. Re Intel currently available laptop CPUs I’ve already shown that for single-core speeds the Ryzen AI 370 beats the fastest Intel Ultra 100 series Meteor Lake CPU, and spoiler alert - the AI 370 wins in multithreaded performance too. There’s another issue with Meteor Lake gen laptops that is a bit complex so I’ll skip for now. But what about the fastest 13th and 14th gen Core i9 series laptop CPUs. In principle these all have enough multithreaded power. In terms of single-threaded Passmark score, the following examples are all either a bit lower or a bit higher than the 4213 score of the Ryzen AI 9 HX 370: i7-14700HX = 3857, i9-13900H = 3857, i9-13950HX = 4034, i9-13900HX = 4142, i9-13980HX = 4307, i9-14900HX = 4295. So in principle any of these Intel Laptop CPUs is sufficiently fast in both single-threaded and multithreaded for, say, gaming or creator use. The problem is that they all run hot. Depending on implementation they might consume up to 157 watts (not counting any discrete GPU), since these are adapted desktop chips. And if you haven’t heard, Intell is having a hellish problem with its too-hot desktop CPUs, and it’s as-yet unknown whether the problem will spill over into its desktop-based laptop CPUs. The symptom is that CPUs burn out over time due to too-hot temps, failing at a random time that may be after warranty expiration. The suspicion is that Intel just set the voltage curve too high just to be able to advertise speeds competitive to AMD. Unclear whether problem can be fixed in BIOS, but if it fails it’s a hardware problem needing a new CPU. The hope is that 13th/14th gen laptop CPUs not affected since lower voltage than desktops, but who knows. For my money, I’ll skip 13th/14th gen Intel, and AMD is much better than Meteor Lake Intel on a value per dollar basis, so at least until Intel Lunar Lake laptop arrives, AMD is the most cost-effective option for a fast laptop and also has the fastest options available (especially the Ryzen AI 9 HX 370).
Sorry, there was a problem. Please try again later.(Performance update single threaded) It’s July 24 and 4 days til release of this ProArt P16 and other models sporting the AMD Ryzen AI 9 HX 370 CPU. That will lift the info embargo and we’ll see some print and video reviews. But for now let’s release a bit of single-thread performance data using the single (lol) data point of Passmark performance data (not enough data so consume with grain of salt). We use the Gen 4 Ryzen 9 7940HS 8-core CPU as basis for comparison (which has 949 data points). The Passmark Single-Core performance is about 8% faster on the AI HX 370 CPU due to architectural improvements that increase instructions per clock cycle (IPC). Clock rate changed very little, so Single-Core performance increase should equal IPC Uplift (as it’s called). We need to reconcile that with AMD propaganda from the product announcement that suggested we should expect a 16% IPC Uplift. That 16% was the geometric average of 13 different IPC Uplifts for the following sample programs: FarCry6 +10% Handbrake +11% Kracken +12% WebXPRT4 +12% 3DMarkPhysics +13% Speedometer +15% Adobe Premier Puget Beach +16% Octane +16% Cinebench R23 +17% Geekbench 6 Text +19% League of Legends +21% Blender +23% and Geekbench 5.4 AES XTS +35%. The problem is that AMD cherry-picked their IPC-Uplift examples, and all of them are either games or very specific benchmarks that are not representative of the types of programs the average Joe User typically runs. But all is not lost, and the end result is actually quite good. You can’t fault AMD too much for cherry picking - they were probably following the principle of “what would Intel do?” We should probably put more emphasis on the +8% Single-Thread performance shown by the early results of Passmark testing. I like Passmark because they combine several types of test programs in a way that is fairly representative of typical computer usage. We can expect about an 8% Single-Core speedup in the AI 370 versus the Ryzen 9 7940HS. At the same time the AMD IPC Uplift data is useful. It shows that for certain specialized applications the architectural changes have resulted in huge speedups. For instance the +17% Cinebench R23 results suggest a huge speedup in video applications (so maybe you wanna trade in your aging Macbook for a ProArt P16). And you’re in luck if you play a lot of FarCry6 or Kracken or Octane or League of Legends, since you can expect a Single-Core speedup anywhere from +10% to +21%. And that’s not counting any additional speedup you might get on games implemented via many cores. Just for grins I’ll list the actual raw score Passmark Single-Thread ratings for some CPUs commonly used for gaming: Core Ultra 9 185H = 3712, Ryzen 7 7800X3D = 3749, Ryzen 9 7940HS = 3903, Ryzen 9 7845HX = 3977, Ryzen 9 7945HX = 4049, Ryzen 9 79HX3D = 4130, Ryzen AI 9 HX 370 = 4213. These are all laptop CPUs except the Ryzen 9 7800X3D, which is the most popular desktop gaming CPU. The 7800X3D beats Intel’s fastest laptop CPU - the Core Ultra 9 185H - in Single-Core performance. But all of the AMD laptop CPUs listed above have faster Single-Core performance than the most popular desktop gaming CPU. Note that the new Ryzen AI 370 CPU takes the crown in Single-Core Performance among all available laptop CPUs, even beating the Ryzen 9 7945HX3D, which is nevertheless more powerful overall due to its 16 cores and 3D-Vcache (which it shares with its big-brother 7800X3D desktop part).
Sorry, there was a problem. Please try again later.