News Intel E-Core-Only N100 Gaming Benchmarks Emerge

[DaveLTX]

It's not 6W in practice. Also, its GPU is weaker than the Steam Deck's, for instance.

Yeah these certainly aren't as power efficient as people think. A quad core Zen 4 tuned for lower power would absolutely smash the N100. And amazingly N100 isn't as cheap as I thought, N305 and the like is even worse!
So it's only good point traditionally namely being cost isn't quite there either. For one of the N100 boxes I could buy a Minisforum 5600H for a bit more
N305? I could buy a whole 5600H for that...

 

[bit_user]

amazingly N100 isn't as cheap as I thought, N305 and the like is even worse!

This shows why. The N-series dies are pretty huge, for what they are. Granted, they incorporate the Southbridge die, but they're still bigger than Gemini Lake in absolute terms.

https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fbucketeer-e05bbc84-baa3-437e-9518-adb32be77984.s3.amazonaws.com%2Fpublic%2Fimages%2Fd6e6e977-3dcc-4a68-8978-e7647a8d82c2_3174x5103.jpeg

Source: https://locuza.substack.com/p/info-snack-alder-lake-m-raptor-lake
​

Sorry, I can't embed the image. You might have to visit the source page, before loading it. It shows a die comparison between all of the Alder Lakes.

The image shows Alder Lake-N die is 60.3% as large as that used for the i9-12900K, which is pretty surprising when you consider the substantial performance gulf between them.

 

[Giroro]

Why not at least put a link in your signature?

The mod here threatened to ban me when I tried linking to youtube in my signature. Mods in a lot of places threaten (and will) ban you just for pointing out you have a youtube channel (unless that channel is really big, because they get special privileges, I guess).

I also tried to promote my twitter in my videos and to post my videos on twitter. In roughly 20 videos of trying twitter, I got up to 2 twitter followers (one is my sister) and 8 Youtube with twitter listed as a source. So I don't try waste time in videos promoting twitter anymore.
I legitimately got more traffic from Parler, which is in itself a giant joke (and my content is in no way political). I get more traffic on my ancient terribly-formatted blogger and myspace than I do on twitter.

This stuff is monopolized and just completely impenetrable to an outsider, especially people who have a job. It's built where only popular content with an existing audience is allowed to be seen by new people. I've even bought youtube ad space. I feel like I would be far more likely to have my content seen if I just started to blind mail DVDs to people, which at this point would have been far cheaper.
It's all one big skinner box treadmill to trick people into engaging with the platform, which I unfortunately got suckered into. Turns out people who falsely think they can make content and grow spend a lot more time on the platform and watch way more ads, which is all they want.

 

[JamesJones44]

I'd like to see the N100 tested against AMD's dual-core (e.g. Athlon Gold 7220U) and quad-core (e.g. Ryzen 3 7320C) Mendocino chips to see which one is the best for a "portable console".

Add ARM variants to those tests. I would love to see the performance per watt battles at the low end to give some perspective of where we are with the two architectures.

My guess is these are pulling more than 6 watts of power (as we all know TDP != power draw), but without a proper set of tests it's all just speculation, these really could just be pulling 6 watts.

 

[DaveLTX]

Add ARM variants to those tests. I would love to see the performance per watt battles at the low end to give some perspective of where we are with the two architectures.

My guess is these are pulling more than 6 watts of power (as we all know TDP != power draw), but without a proper set of tests it's all just speculation, these really could just be pulling 6 watts.

Alder lake N will need a lot more to boost and will need 25W at least for PL2 state
Even tablet and high end phone processors have a TDP of 10W or more, usually undisclosed but they can hit 10W when given enough cooling and for phones, only momentarily before the saturation temperature is hit and backs off

 

[LuxZg]

I'd love to get a more general benchmark. Gaming obviously isnt the goal!

For example, those could work great as home server/router or NAS.

Indeed , as someone said this is roughly Skylake i5 6500 performance, and that's exactly what I have in local Nextcloud box. But even if N100 peaks at 20-30W it's still improvement over Skylake. And performance is way better than RPi . So for a NAS for family or small office, I think these would be great.

 

[cyrusfox]

Yeah these certainly aren't as power efficient as people think. A quad core Zen 4 tuned for lower power would absolutely smash the N100. And amazingly N100 isn't as cheap as I thought, N305 and the like is even worse!
So it's only good point traditionally namely being cost isn't quite there either. For one of the N100 boxes I could buy a Minisforum 5600H for a bit more
N305? I could buy a whole 5600H for that...

I have come across some barebone boxes for $120-140 price mark(8GB vs 16GB DDR5), I thought it was pretty compelling. The Minisforum I see above $300 but perhaps I am looking at the wrong site.

Seems like a steal to replace any ancient skylake generation quad core wit hmore than enough performance for aux PC work.

 

[bit_user]

Add ARM variants to those tests. I would love to see the performance per watt battles at the low end to give some perspective of where we are with the two architectures.

Yeah, but the amount of cross-platform benchmarks is really small. I guess there's pretty much just GeekBench (which has extremely limited GPU testing) and 3DMark/Android. Actually, it would be interesting to see if 3DMark has a native ARM/Windows port.

BTW, there's apparently a 3DMark/Linux port, but it's only available to select users due to the lack of a flashy GUI and maybe other known issues.

Taking a step back and thinking about it, the only ARM SoC I think would stand a decent chance against it is the Snapdragon 8cx Gen 3, like what's in the Lenovo Thinkpad X13S or this Windows/ARM dev kit:

Windows Dev Kit 2023

Powered by Arm64 and running Windows 11, this desktop device enables you to develop Windows apps for Arm, on Arm. Find device specifications, set up instructions, Arm-native developer tools, Support, and FAQs.

learn.microsoft.com

Or maybe the 7cx equivalent. But I don't expect the RK3588 is a decent match for it, except maybe on all-thread workloads (thanks to the A55 cores). A76 just isn't quite a Skylake equivalent, though.

these really could just be pulling 6 watts.

Please. Let's be realistic. These are Alder Lake chips, after all.

 

[bit_user]

I have come across some barebone boxes for $120-140 price mark(8GB vs 16GB DDR5), I thought it was pretty compelling.

Not bad, but I'd worry either the fan is annoying or that it's undercooled and cooking in there.

Seems like a steal to replace any ancient skylake generation quad core wit hmore than enough performance for aux PC work.

I wouldn't trust a cheap no-name bargain brand, for a machine I actually depended on.

 

[cyrusfox]

Not bad, but I'd worry either the fan is annoying or that it's undercooled and cooking in there.

Holds up per the reviews on cooling/noise (Assuming they are reliable). interesting to see only x2 lanes for the nvme drive

I wouldn't trust a cheap no-name bargain brand, for a machine I actually depended on.

What do you mean you have never heard of the WooYi Brand, as common as HP or Dell 😀
Check out the RGB on it, can't spell quality without RGB

 

[RedBear87]

I'm sure if paired with a reasonably decent GPU, it would've performed better.

The problem, as bit_user has already pointed out, is that these things come with very limited PCIe lanes, indeed as he said they come with just nine PCIe 3rd Gen lanes; even ASrock's N100M, which comes with what might look like a regular x16 PCIe Gen 3 slot, supports only x2 connections through that slot. At that speed any GPU would be severely bandwidth limited, operating at around 1/8 of its regular speed. Mind you, it would be interesting to see some GTX 1650, RX 6400 and ARC 380 benchmarked using that motherboard, it's kind of weird that no one bothered doing it because the motherboards appear to be available right now, but I suppose that it would be just a curiosity for most people.

 

[thestryker]

My point is that it's almost certainly using more than 6 W. We just don't know much.

This claims the default PL2 of the N200 (another PL1=6W SoC) is 25W:

​

Intel introduces Alder Lake N-series chips for laptops priced below $400 - Liliputing

Intel introduces Alder Lake N-series chips for laptops priced below $400

liliputing.com

​

Of course, these can be customized by the OEM, so whoever is doing the benchmarks should run a utility that can check. Or, better yet, just collect data on the package power during their tests.

STH doesn't do a lot of benchmarking, but they have a broad range of generations and products on compare and have power ranges: https://www.servethehome.com/fanless-intel-n100-firewall-and-virtualization-appliance-review/

I wouldn't trust a cheap no-name bargain brand, for a machine I actually depended on.

I'm not sure who makes Zotac's, but basically every single other minipc are made in effectively the same place. Most of these types of things there simply won't be an equivalent to in NA/EU period, or it'll be a company charging more money for the exact same thing.

 

[bit_user]

... even ASrock's N100M, ...

That foolish board uses DDR4. Otherwise, I might've already bought one.

 

[bit_user]

STH doesn't do a lot of benchmarking, but they have a broad range of generations and products on compare and have power ranges: https://www.servethehome.com/fanless-intel-n100-firewall-and-virtualization-appliance-review/

Yes, but...

"In terms of power consumption, both the N100 and N200 consumed around 10.5-12W at idle. That feels high, especially since the 35W TDP-fanned 1L PCs like the HP Elite Mini 600 G9 is under 8W. On the maximum power consumption side, these only raised to 22-23W for the N100."
​

So, I'm going to put that high idle power mostly down to the Ethernet? I guess the M.2 SSD could be contributing, if something weird is going on with it. It's certainly not contributing more than a couple W, max.

The HP machine they reference uses a SK Hynix PCIe 3.0 drive. If it's anything like their P31 Gold, it should be very efficient. Both machines use DDR5, but the HP is using an i7-12700T, which is socketed!

Anyway, I have to agree that the idle power is a lot higher than I'd like to see. 8 W would be okay. 22-23 W is in the range of what I'd expect for peak active power. The main question is whether the manufacturer customized the PL2 threshold. Given that it's fanless, I'd have to guess they did.

BTW, I do like what they did with the chassis! It looks like a pain to dust, but otherwise pretty neat.

 

[thestryker]

Yes, but...

​

"In terms of power consumption, both the N100 and N200 consumed around 10.5-12W at idle. That feels high, especially since the 35W TDP-fanned 1L PCs like the HP Elite Mini 600 G9 is under 8W. On the maximum power consumption side, these only raised to 22-23W for the N100."​

​

So, I'm going to put that high idle power mostly down to the Ethernet? I guess the M.2 SSD could be contributing, if something weird is going on with it. It's certainly not contributing more than a couple W, max.

The HP machine they reference uses a SK Hynix PCIe 3.0 drive. If it's anything like their P31 Gold, it should be very efficient. Both machines use DDR5, but the HP is using an i7-12700T, which is socketed!

Anyway, I have to agree that the idle power is a lot higher than I'd like to see. 8 W would be okay. 22-23 W is in the range of what I'd expect for peak active power. The main question is whether the manufacturer customized the PL2 threshold. Given that it's fanless, I'd have to guess they did.

BTW, I do like what they did with the chassis! It looks like a pain to dust, but otherwise pretty neat.

If it's anything like my box the default tuning in the BIOS is atrocious. I haven't felt like hooking mine back up to a display to further tune it, but it has no qualms about hitting a 20W PL2. For the old 5105/6005 boxes there were settings tweaks to lower idle so it might just be an optimization thing. That is why I'm waiting a bit longer before upgrading mine though I'd like to see some of the base operation stuff cleaned up. Below is the one I've got, and I bought the cheapest at the time as I didn't know how much I'd like it and should have spent the extra $40-50 or so the better case would have cost (I don't use the fan anymore, but before I installed a 40mm inside I did).

 

[DaveLTX]

Yeah, but the amount of cross-platform benchmarks is really small. I guess there's pretty much just GeekBench (which has extremely limited GPU testing) and 3DMark/Android. Actually, it would be interesting to see if 3DMark has a native ARM/Windows port.

BTW, there's apparently a 3DMark/Linux port, but it's only available to select users due to the lack of a flashy GUI and maybe other known issues.

Taking a step back and thinking about it, the only ARM SoC I think would stand a decent chance against it is the Snapdragon 8cx Gen 3, like what's in the Lenovo Thinkpad X13S or this Windows/ARM dev kit:

​

Windows Dev Kit 2023

Powered by Arm64 and running Windows 11, this desktop device enables you to develop Windows apps for Arm, on Arm. Find device specifications, set up instructions, Arm-native developer tools, Support, and FAQs.

learn.microsoft.com

​

Or maybe the 7cx equivalent. But I don't expect the RK3588 is a decent match for it, except maybe on all-thread workloads (thanks to the A55 cores). A76 just isn't quite a Skylake equivalent, though.


Please. Let's be realistic. These are Alder Lake chips, after all.

Indeed, 8cx Gen 3 sounds like it could wipe the floor with N100. Definitely still absolutely no match for a low power optimized Zen 4 APU, at 15W the Z1E/7840U is able to outmuscle a 10600K very easily and not far off a 11800H
The issue with gracemont is that it is Skylake IPC in only a very narrow set of workloads. Introduce what we usually consider as "multitasking" and they are no longer skylake IPC.

The problem, as bit_user has already pointed out, is that these things come with very limited PCIe lanes, indeed as he said they come with just nine PCIe 3rd Gen lanes; even ASrock's N100M, which comes with what might look like a regular x16 PCIe Gen 3 slot, supports only x2 connections through that slot. At that speed any GPU would be severely bandwidth limited, operating at around 1/8 of its regular speed. Mind you, it would be interesting to see some GTX 1650, RX 6400 and ARC 380 benchmarked using that motherboard, it's kind of weird that no one bothered doing it because the motherboards appear to be available right now, but I suppose that it would be just a curiosity for most people.

And also, the nine PCIe Gen 3 lanes can't be tied together more than a x4 or even x2. Not sure about specifics but look around and the asrock N100 motherboard has a x2 lane for the x16 physical slot

STH doesn't do a lot of benchmarking, but they have a broad range of generations and products on compare and have power ranges: https://www.servethehome.com/fanless-intel-n100-firewall-and-virtualization-appliance-review/

I'm not sure who makes Zotac's, but basically every single other minipc are made in effectively the same place. Most of these types of things there simply won't be an equivalent to in NA/EU period, or it'll be a company charging more money for the exact same thing.

PCPartner is the parent company of Zotac. They are a big company and they make their own mini PCs!

 

[thestryker]

PCPartner is the parent company of Zotac. They are a big company and they make their own mini PCs!

Yeah I couldn't remember who owned them, but I thought they did all their own stuff still.

And also, the nine PCIe Gen 3 lanes can't be tied together more than a x4 or even x2. Not sure about specifics but look around and the asrock N100 motherboard has a x2 lane for the x16 physical slot

Yeah the N series are only meant for laptops/NUCs these are drop in replacements for Tremont.

Indeed, 8cx Gen 3 sounds like it could wipe the floor with N100. Definitely still absolutely no match for a low power optimized Zen 4 APU, at 15W the Z1E/7840U is able to outmuscle a 10600K very easily and not far off a 11800H

While academically these comparisons might be interesting the price categories are polar opposites. N100 boxes start around $120-150 and the cheapest Zen 4 I've seen starts around $520 and the Qualcomm chips can only be found in Microsoft products (I'm hoping the Nuvia designed chips will change this). There aren't many N305 boxes yet, but the router type ones cost just under $100 more than the N100 versions so one of these would still be under half the cost. There's a ton of applications where the extra power simply isn't worth the extra cost.

The issue with gracemont is that it is Skylake IPC in only a very narrow set of workloads. Introduce what we usually consider as "multitasking" and they are no longer skylake IPC.

Yeah that's not actually an accurate statement just check the STH piece I linked the 6500T is 4c/4t Skylake and the chips perform within margin of error of each other. The biggest limitation on Gracemont is that they'll never have hyperthreading and the limited instruction set.

 

[bit_user]

Indeed, 8cx Gen 3 sounds like it could wipe the floor with N100. Definitely still absolutely no match for a low power optimized Zen 4 APU, at 15W the Z1E/7840U is able to outmuscle a 10600K very easily and not far off a 11800H

Yeah, it's hard to get good benchmarks on it, because it's mainly just supported under Windows and I think a lot of the stuff people are running is x86 programs under emulation. Even so, there's no way X1 cores are going to beat anything newer than Skylake.

If you're curious, the best review you'll probably find is:

​

Lenovo ThinkPad X13s G1 Laptop review: Introducing the Qualcomm Snapdragon 8cx Gen 3

Notebookcheck tests the Lenovo ThinkPad X13s, an efficient, light, and silent subnotebook for business customers.

www.notebookcheck.net

​

The issue with gracemont is that it is Skylake IPC in only a very narrow set of workloads. Introduce what we usually consider as "multitasking" and they are no longer skylake IPC.

Tell me more. Is that because of the cache situation, or what? Do you have any sources with more info?

And also, the nine PCIe Gen 3 lanes can't be tied together more than a x4 or even x2.

Yes, and for practical reasons you wouldn't want to, any how.

Not sure about specifics but look around and the asrock N100 motherboard has a x2 lane for the x16 physical slot

And that's actually a big step up from their previous gen, which gave us only a PCIe 2.0 x1 slot!! That wasn't even quite enough to run a SATA 3 controller at full-rate.

In contrast, PCIe 3.0 x2 is enough to run a 10 Gigabit Ethernet card. That's the key thing, for some.

 

[bit_user]

Yeah the N series are only meant for laptops/NUCs these are drop in replacements for Tremont.

Tremont is the core. Jasper Lake and Elkhart Lake are the two popular SoC lines, with the latter being more commercial/industrial-focused.

We were promised in-band ECC support, on Elkhart Lake, but it's been virtually nonexistent. It's exciting, because it means you can get ECC functionality with non-ECC memory. The tradeoff is a little capacity and performance. Anandtech benchmarked it here:

ASRock Industrial NUCS BOX-1360P/D4 Review: Raptor Lake-P Impresses, plus Surprise ECC

www.anandtech.com

the Qualcomm chips can only be found in Microsoft products

There are Laptops from multiple vendors with the Snapdragon 8cx Gen 3. The most high-profile seems to be Lenovo's X13S, which I linked above. It's intended to be a laptop chip, after all. Mini-PCs are too far "down market" for it, and don't really play to its strengths.

check the STH piece I linked the 6500T is 4c/4t Skylake and the chips perform within margin of error of each other. The biggest limitation on Gracemont is that they'll never have hyperthreading and the limited instruction set.

If you're comparing with a Skylake i5, that doesn't have HT either. As for instruction set, Gracemont has a superset of what the Skylake client CPUs have.

 

[thestryker]

If you're comparing with a Skylake i5, that doesn't have HT either.

Yeah that was the point: the IPC is the same so what the poster was claiming isn't accurate.

As for instruction set, Gracemont has a superset of what the Skylake client CPUs have.

Yeah I meant in comparison to modern AMD/Intel CPUs.

There are Laptops from multiple vendors with the Snapdragon 8cx Gen 3. The most high-profile seems to be Lenovo's X13S, which I linked above. It's intended to be a laptop chip, after all. Mini-PCs are too far "down market" for it, and don't really play to its strengths.

Totally didn't realize the gen 3 hit other manufactures since the first two didn't really even though they were available. Still same difference though it's pointless to point at something that will never be in the same field and say it's better.

Tremont is the core. Jasper Lake and Elkhart Lake are the two popular SoC lines, with the latter being more commercial/industrial-focused.

We were promised in-band ECC support, on Elkhart Lake, but it's been virtually nonexistent. It's exciting, because it means you can get ECC functionality with non-ECC memory. The tradeoff is a little capacity and performance. Anandtech benchmarked it here:

​

ASRock Industrial NUCS BOX-1360P/D4 Review: Raptor Lake-P Impresses, plus Surprise ECC

www.anandtech.com

No Tremont cores had over 8 lanes of PCIe as far as I'm aware, but the Gracemont C1100 series does which is why I phrased it the way I did.

 

[bit_user]

Totally didn't realize the gen 3 hit other manufactures since the first two didn't really even though they were available. Still same difference though it's pointless to point at something that will never be in the same field and say it's better.

If you go back to where it first got raised, I mentioned it in response to a question about how this would compare to various ARM SoCs. I took the question to be in the spirit of technical inquisitiveness, not one of commercial competitiveness or product positioning. That's what Interests me, anyhow.

I'm interested in how all sorts of different cores and CPUs compare, on all sorts of tasks. Commercial considerations come second, but the technical analysis can often shed insights into the commercial side.

No Tremont cores had over 8 lanes of PCIe as far as I'm aware, but the Gracemont C1100 series does which is why I phrased it the way I did.

PCIe is a function of the SoC, which is independent of the core microarchitecture. We can look at Snow Ridge, for instance, which had up to 24 Tremont cores and 32 PCIe 3.0 lanes:

​

Intel Atom® Processor P5362 (27M Cache, 2.2GHz) - Product Specifications | Intel

Intel Atom® Processor P5362 (27M Cache, 2.2GHz) quick reference with specifications, features, and technologies.

ark.intel.com

​

...but it's somewhat academic, because those are sold into a narrow vertical (5G basestations, I think) and I doubt you can buy them for use in other applications. At least, not without dealing somewhat directly with Intel.

 

[DaveLTX]

Yeah, it's hard to get good benchmarks on it, because it's mainly just supported under Windows and I think a lot of the stuff people are running is x86 programs under emulation. Even so, there's no way X1 cores are going to beat anything newer than Skylake.

If you're curious, the best review you'll probably find is:

​

Lenovo ThinkPad X13s G1 Laptop review: Introducing the Qualcomm Snapdragon 8cx Gen 3

Notebookcheck tests the Lenovo ThinkPad X13s, an efficient, light, and silent subnotebook for business customers.

www.notebookcheck.net

​

Tell me more. Is that because of the cache situation, or what? Do you have any sources with more info?


Yes, and for practical reasons you wouldn't want to, any how.


And that's actually a big step up from their previous gen, which gave us only a PCIe 2.0 x1 slot!! That wasn't even quite enough to run a SATA 3 controller at full-rate.

In contrast, PCIe 3.0 x2 is enough to run a 10 Gigabit Ethernet card. That's the key thing, for some.

It's most likely due to the cache but also the "low power ambitions" nature of it much like ARM's low power cores.
Unfortunately its not low power enough

Yes, the last generation had 8 lanes from the integrated PCH

Yeah I couldn't remember who owned them, but I thought they did all their own stuff still.


Yeah that's not actually an accurate statement just check the STH piece I linked the 6500T is 4c/4t Skylake and the chips perform within margin of error of each other. The biggest limitation on Gracemont is that they'll never have hyperthreading and the limited instruction set.

They definitely do their own stuff, PCPartner has always owned Zotac.

I am aware that at peak performance it runs like a skylake clock for clock. But do a wider gamut of benchmark and Gracemont starts to show that its only skylake at peak performance.
For instance, half width AVX. AVX is there as a "ISA compatible" situation but AVX on gracemont falls behind very quickly
It is perfectly matching with integer workloads but how often are integer only workloads around? Loads of vector workloads (practically anything using the FPU) exist and gracemont's efficiency becomes really especially poor

 

[bit_user]

I am aware that at peak performance it runs like a skylake clock for clock. But do a wider gamut of benchmark and Gracemont starts to show that its only skylake at peak performance.
For instance, half width AVX. AVX is there as a "ISA compatible" situation but AVX on gracemont falls behind very quickly
It is perfectly matching with integer workloads but how often are integer only workloads around? Loads of vector workloads (practically anything using the FPU) exist and gracemont's efficiency becomes really especially poor

Do you have any benchmarks to share?

From what I've seen, its floating-point performance is still fairly capable.

Source: https://www.anandtech.com/show/1704...hybrid-performance-brings-hybrid-complexity/7
​

This shows its SPEC2017int score is 64.5% as fast as a single-threaded Golden Cove (DDR5). Meanwhile, its SPEC2017fp score is 54.1% as fast as a single-threaded Golden Cove.

If we compare it with Coffee Lake-R, which is essentially built from Skylake cores, it manages 88.2% of the SPEC2017int and 79.5% of the SPEC2017fp. If you compensate for the clockspeed difference (3.9 vs 5.0 GHz), you actually get int and fp performance 113.1% and 101.9% as great, respectively.

So, even of float-point and vector workloads, it's definitely defensible to say that Gracemont has Skylake-level IPC.

 

[DaveLTX]

Do you have any benchmarks to share?

From what I've seen, its floating-point performance is still fairly capable.

​

​

Source: https://www.anandtech.com/show/1704...hybrid-performance-brings-hybrid-complexity/7​

​

This shows its SPEC2017int score is 64.5% as fast as a single-threaded Golden Cove (DDR5). Meanwhile, its SPEC2017fp score is 54.1% as fast as a single-threaded Golden Cove.

If we compare it with Coffee Lake-R, which is essentially built from Skylake cores, it manages 88.2% of the SPEC2017int and 79.5% of the SPEC2017fp. If you compensate for the clockspeed difference (3.9 vs 5.0 GHz), you actually get int and fp performance 113.1% and 101.9% as great, respectively.

So, even of float-point and vector workloads, it's definitely defensible to say that Gracemont has Skylake-level IPC.

It's not quite so simple. And in any case even Zen 2 cores have higher efficiency than gracemont cores. Which is unfortunate.

 

[bit_user]

It's not quite so simple.

Yes, but we have to do the best we can with the data available. If you have better data, please share.

And in any case even Zen 2 cores have higher efficiency than gracemont cores. Which is unfortunate.

This would be a big revelation, if you could substantiate it. For instance, if true, then Rome (Zen 2-based EPYC) should've left no market opportunity for Sierra Forest (Gracemont-based Xeon). All AMD would've had to do is port Zen 2 to TSMC N5 or N4 and call it a day.

I've seen only one detailed performance comparison involving Zen 2 and Gracemont. However, there's a massive disclaimer right at the start of it:

"Please take power measurements in this section with a grain of salt. Research suggests that AMD doesn’t have power measurement hardware, which limits the accuracy of their RAPL counters"
​

You can go on to read the article, but that basically makes their Zen 2 comparisons irrelevant, other than perhaps the shape of the curves.

Alder Lake’s Power Efficiency – A Complicated Picture

Reviews across the internet show Alder Lake getting very competitive performance with very high power consumption.

chipsandcheese.com

The best we could do is compare package power of a Zen 2 APU against a N300, in order for it to be a fair test. If you're aware of such a comparison, please share.