From the slides, it looks like this is another X3D CCD + regular CCD? Has AMD's core parking driver improved since launch? If so, how good/bad is it now?
If core parking doesn't work well in a desktop, it's going to be disastrous in a laptop.
Did you invest time into core parking and its related issues, or is this yet another typical smash the internet with the usual bashing because at launch there were some issues and lets keep repeating it...
Why are you so defensive? AMD has always been bad with drivers at launch, but they also tend to improve over time. I am curious if it has matured now, or if it's still crap. If you know the answer, just say it.
It's not a driver issue, it's a Windows scheduler issue. Windows doesn't have a mechanism to determine which programs/games benefit from the additional cache so it ends up migrating threads between arbitrary cores. The solution short term was to just disable the CCD without the cache so that all available cores have the same performance.
Part of the reason that it's hard to get an answer is that your question included invalid assumptions and mis-described the problem, as was pointed out to you.
This is what I think we call 'militant misinterpretation'.
AMD does not write microsofts scheduler. Its not their fault, just like it wasn't intels fault when their E cores and P cores weren't being used properly at launch. That's a tad easier for MS to fix though, sadly with X3d it's not as simlpe as 'fast core/slow core' because that depends on the workload whereas it does not with E/P core.
Will be solved in next year or so. Until then you still got 8 core laptops when you game, which is strong already. (literally the best laptop CPU you could get last year). that cpu will be one of the best for a while, so getting it slightly
it can be easly upgraded, ignoring the workload's for now. just make it so x3d takes care of MAIN focused window, while other cores gets other non-focused processes. it's enough to avoid parking, and that means full power for 70-80% of users. if they give you an option to set that in task manager/something permanent like you could bound to cpu's cores directly, and you have a winner. it does not have to magically know what process is goes where.
And aren't 16 cores on a laptop pretty much useless for 99% of people? Most of these are going to go into gaming laptops where the vast majority of games even today can't use more than 2 or 4 cores anyway.
But it's more due to the other performance limitations that all quad core CPUs have by virtue of being old as shit or spec'd at the very lowest end of the market.
If you take a modern 6 core or greater CPU and disable all but four cores the performance difference isn't very big in most games, I seem to remember.
I think consoles stopping at 8 cores is no accident. Virtually nothing scales to efficiently using 4 cores, not even AAA titles but having 8 available is just perfect to run badly threaded game code while keeping the OS from plaguing performance, and letting 2 of the 8 cores reach high turbo clocks. Beyond 8 it's just a total drag on power, speed and temps.
That's gaming anyway, something X3D targets, so I really want to see an 8 core with 128MB Cache.
Haha! Yeah no, unless there is something very wrong with the CPU or cooling arrangement in question, executing a program will not even damage a chip, let alone destroy it.
For now, by far the best integrated GPU is available in Ryzen 9 7940HS, Ryzen 7 7840HS and Ryzen 7 7840U.
There are many cheap small computers with these AMD CPUs.
This will change only by the end of the year, when Intel is expected to launch Meteor Lake, which might have the best integrated GPU (with 4/3 more execution units than AMD Phoenix and with a little more than 3/4 of the clock frequency of AMD Phoenix), until the launch of AMD Zen 5.
While the M1 GPU has 1024 FP32 ALUs vs. 768 of AMD Phoenix, the M1 GPU has a low clock frequency, IIRC not much above a half of the clock frequency of AMD Phoenix.
At launch, the M1 GPU was indeed better than any other integrated GPU, but meanwhile it has been surpassed.
Moreover, the M1 GPU does not really count, because it is not documented, so its use is limited only to the programs allowed and controlled by Apple.
There are people who have worked a lot to reverse engineer the behavior of the M1 GPU, but I do not consider that such a work is worthwhile when it is much simpler to buy a better supported GPU.
All the benchmarks that I have ever seen published about Apple cores, including here on Anandtech. showed power consumptions above 4 W per core.
All the AMD Zen 4 laptop CPUs and server CPUs have a similar power consumption per core, in the range of 2.5 W to 4 W per core, Only the desktop CPUs, like 7950X or 7900X have a higher power consumption, in the range of 8 W to 10 W per core, due to their high clock frequency.
So I cannot believe that M2 can run Geekbench 6 at 2 W, but 5 W is a plausible value for the single-thread GB 6. 10 W is impossible for the multi-threaded GB 6, unless the M2 is throttled in something like MacBook Air, due to poor cooling. When running at full performance, an M2 consumes well over 20 W.
Where AMD is definitely worse than Apple is in the idle power consumption, especially in the desktop CPUs and in these HX laptop CPUs, which are repackaged desktop CPUs.
The AMD Phoenix 7040 series has acceptable idle power consumption, i.e. a complete computer with those has a wall plug idle power consumption of 4 to 5 W, where the CPU alone has probably much less.
The vast majority of GB run for an M1/M2 is sigificantly below 5w. Most of the run, it's in 1w - 3w. If you don't believe, me, ask anyone with an M1 to run it and show power metrics at the same time. It's easy to verify.
Really wish they just made the laptop 3D v-cache a single CCD 8 core chip but with boost frequencies in line with what they have on the 7945HX3D as if they are targeting the best gaming chip then that would have been the far better option, and cheaper on top of it all, kind of annoying that we're back to Intel and AMD chasing each other with core/thread counts again.
At this point, AMD keeps surprising me, so I will have to guess that they do have a 7745HX3D in the pipeline and haven't announced it yet. Just like how they released 7800X3D well after all the early adopters bought a 7950X3D.
There's 128MB of L3 cache. The 144MB is counting L2 + L3 cache. This is something AMD started doing in their marketing for the last couple years. It's a bit confusing, but it allows them to use an even larger number to advertise total cache.
I think that makes sense. AMD and Intel are using non-inclusive L3 cache right now. So the contents in L2 does not necessarily get duplicated in L3 (like in pre-Ice Lake days), the chip gets all the cache to hold different values
Unlikely. This is a repackaged desktop part, as the article described. The Framework 16 uses true mobile parts, which have completely different (and easier to integrate) constraints. Some things you can expect to be different: mobile chips have lower z-height, more manageable power delivery requirements, better thermal management, and fewer external needs (both traces and additional chips).
You should only expect to see desktop replacement chips in thicker 16 and 17 inch laptops. The Framework 16 is far too slim to support a desktop replacement chip.
I've been coming to this site for many years now and while don't really post or contribute with meaningful comments I've always seen lemurbutton posts as no more than an apple fan making comments to stir the pot. I've never taken him seriously in any way.
Oh let him troll them. If the regulars here haven't figured it out yet, then they deserve what they get. Hang out off to the side and enjoy watching them squabble.
The TDP is troublesome, but of course I would like to see more cache for the CPU land in mobile chips, but the 4 - 15W TDP range that is passively cooled and has high unplugged endurance is more meaningful to me than any DTR chip. So while a few people that are heavily invested in playing games might find these useful, the rest of us are still waiting while capability creeps down the stack to more mundane systems.
I'm not going to say never, but I don't think this will hit mobile chips unless a competitor forces AMD to. AMD has admitted that the stacked cache was a server component that turned out to be good for gaming. The technology is likely to be developed for better server integration for a long time before it gets adapted to fit with mobile packaging.
Reconciling channel data with AMD q2 10K it appears ASUS is AMD gaming division revenue in q2 at $1.581 B. ASUS saves the quarter and well planned by AMD filling a product revenue void. That scores ASUS 2,535.029 7945X3D and an equal number of Rog Ally 7840U at gross $289 each. mb
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
51 Comments
Back to Article
meacupla - Thursday, July 27, 2023 - link
From the slides, it looks like this is another X3D CCD + regular CCD?Has AMD's core parking driver improved since launch? If so, how good/bad is it now?
If core parking doesn't work well in a desktop, it's going to be disastrous in a laptop.
Ryan Smith - Thursday, July 27, 2023 - link
'From the slides, it looks like this is another X3D CCD + regular CCD?"Correct. It's the exact same configuration as a 7950X3D.
duploxxx - Friday, July 28, 2023 - link
Did you invest time into core parking and its related issues, or is this yet another typical smash the internet with the usual bashing because at launch there were some issues and lets keep repeating it...meacupla - Friday, July 28, 2023 - link
Why are you so defensive?AMD has always been bad with drivers at launch, but they also tend to improve over time.
I am curious if it has matured now, or if it's still crap.
If you know the answer, just say it.
hecksagon - Friday, July 28, 2023 - link
It's not a driver issue, it's a Windows scheduler issue. Windows doesn't have a mechanism to determine which programs/games benefit from the additional cache so it ends up migrating threads between arbitrary cores. The solution short term was to just disable the CCD without the cache so that all available cores have the same performance.meacupla - Friday, July 28, 2023 - link
So is the clunky core parking workaround fixed now?Why is it so hard to get a straight answer to a simple question?
bji - Friday, July 28, 2023 - link
Part of the reason that it's hard to get an answer is that your question included invalid assumptions and mis-described the problem, as was pointed out to you.Also, do your own homework.
meacupla - Friday, July 28, 2023 - link
okay, so that's a NO then.Cool product ruined by bad AMD software, again.
Qasar - Friday, July 28, 2023 - link
ahh so something that may not be amd's fault, as it is something windows needs to handle, is STILL amds fault. ok sure.....MrPoletski - Monday, July 31, 2023 - link
This is what I think we call 'militant misinterpretation'.AMD does not write microsofts scheduler. Its not their fault, just like it wasn't intels fault when their E cores and P cores weren't being used properly at launch. That's a tad easier for MS to fix though, sadly with X3d it's not as simlpe as 'fast core/slow core' because that depends on the workload whereas it does not with E/P core.
deil - Wednesday, August 2, 2023 - link
Will be solved in next year or so. Until then you still got 8 core laptops when you game, which is strong already. (literally the best laptop CPU you could get last year).that cpu will be one of the best for a while, so getting it slightly
it can be easly upgraded, ignoring the workload's for now.
just make it so x3d takes care of MAIN focused window, while other cores gets other non-focused processes.
it's enough to avoid parking, and that means full power for 70-80% of users.
if they give you an option to set that in task manager/something permanent like you could bound to cpu's cores directly, and you have a winner.
it does not have to magically know what process is goes where.
Duwelon - Friday, July 28, 2023 - link
And aren't 16 cores on a laptop pretty much useless for 99% of people? Most of these are going to go into gaming laptops where the vast majority of games even today can't use more than 2 or 4 cores anyway.Dante Verizon - Saturday, July 29, 2023 - link
You are kidding? Most current games destroy any quad-core, it's a miserable experience to play with less than 6 cores.garblah - Saturday, July 29, 2023 - link
But it's more due to the other performance limitations that all quad core CPUs have by virtue of being old as shit or spec'd at the very lowest end of the market.If you take a modern 6 core or greater CPU and disable all but four cores the performance difference isn't very big in most games, I seem to remember.
Duwelon - Saturday, July 29, 2023 - link
I think consoles stopping at 8 cores is no accident. Virtually nothing scales to efficiently using 4 cores, not even AAA titles but having 8 available is just perfect to run badly threaded game code while keeping the OS from plaguing performance, and letting 2 of the 8 cores reach high turbo clocks. Beyond 8 it's just a total drag on power, speed and temps.That's gaming anyway, something X3D targets, so I really want to see an 8 core with 128MB Cache.
PeachNCream - Monday, July 31, 2023 - link
"..destroy any quad-core..."Haha! Yeah no, unless there is something very wrong with the CPU or cooling arrangement in question, executing a program will not even damage a chip, let alone destroy it.
tafreire - Friday, July 28, 2023 - link
AMD needs to release a high performance iGPU.nandnandnand - Friday, July 28, 2023 - link
You probably want Strix Halo.meacupla - Friday, July 28, 2023 - link
yeah, this is what I am waiting for.AdrianBc - Friday, July 28, 2023 - link
For now, by far the best integrated GPU is available in Ryzen 9 7940HS, Ryzen 7 7840HS and Ryzen 7 7840U.There are many cheap small computers with these AMD CPUs.
This will change only by the end of the year, when Intel is expected to launch Meteor Lake, which might have the best integrated GPU (with 4/3 more execution units than AMD Phoenix and with a little more than 3/4 of the clock frequency of AMD Phoenix), until the launch of AMD Zen 5.
lemurbutton - Friday, July 28, 2023 - link
The M1 GPU is still more powerful than any AMD/Intel laptop iGPU chips.AdrianBc - Friday, July 28, 2023 - link
This is unlikely.While the M1 GPU has 1024 FP32 ALUs vs. 768 of AMD Phoenix, the M1 GPU has a low clock frequency, IIRC not much above a half of the clock frequency of AMD Phoenix.
At launch, the M1 GPU was indeed better than any other integrated GPU, but meanwhile it has been surpassed.
Moreover, the M1 GPU does not really count, because it is not documented, so its use is limited only to the programs allowed and controlled by Apple.
There are people who have worked a lot to reverse engineer the behavior of the M1 GPU, but I do not consider that such a work is worthwhile when it is much simpler to buy a better supported GPU.
Qasar - Friday, July 28, 2023 - link
lemurbutton.post links to this. or shut up.
Dante Verizon - Saturday, July 29, 2023 - link
Without proper support and a decent OS it's just a paperweight.iphonebestgamephone - Saturday, July 29, 2023 - link
The rog ally igpu is a bit faster than m1 igpu, 5k vs 4.8k 3dmark wildlife extreme.lemurbutton - Friday, July 28, 2023 - link
A CPU that runs at 75w? Probably more because manufactures often boost well beyond the stated TDP.Are you kidding me?
Meanwhile, an M2 will idle at less than 1w and will run Geekbench 6 at around 2-5w, 10w max.
Come on AMD. Do better.
AdrianBc - Friday, July 28, 2023 - link
All the benchmarks that I have ever seen published about Apple cores, including here on Anandtech. showed power consumptions above 4 W per core.All the AMD Zen 4 laptop CPUs and server CPUs have a similar power consumption per core, in the range of 2.5 W to 4 W per core, Only the desktop CPUs, like 7950X or 7900X have a higher power consumption, in the range of 8 W to 10 W per core, due to their high clock frequency.
So I cannot believe that M2 can run Geekbench 6 at 2 W, but 5 W is a plausible value for the single-thread GB 6. 10 W is impossible for the multi-threaded GB 6, unless the M2 is throttled in something like MacBook Air, due to poor cooling. When running at full performance, an M2 consumes well over 20 W.
Where AMD is definitely worse than Apple is in the idle power consumption, especially in the desktop CPUs and in these HX laptop CPUs, which are repackaged desktop CPUs.
The AMD Phoenix 7040 series has acceptable idle power consumption, i.e. a complete computer with those has a wall plug idle power consumption of 4 to 5 W, where the CPU alone has probably much less.
Qasar - Friday, July 28, 2023 - link
adrian. ignore him.. he claims how good apple is, and the m1 m2 are.. but never posts any proof of it.hes just blowing smoke
lemurbutton - Saturday, July 29, 2023 - link
The vast majority of GB run for an M1/M2 is sigificantly below 5w. Most of the run, it's in 1w - 3w. If you don't believe, me, ask anyone with an M1 to run it and show power metrics at the same time. It's easy to verify.iphonebestgamephone - Saturday, July 29, 2023 - link
Link? The a16 goes over 11w in geekbench, though its the peak.Qasar - Saturday, July 29, 2023 - link
lemurbutton...blah blah blah blah. cant post a source cause you have none, and its all in your fantasy world...
IF it was so easy, you would post a link....
erotomania - Sunday, July 30, 2023 - link
My M2 Mac mini can pull up to about 30W from the wall, especially during CPU benchmarks.erinadreno - Friday, July 28, 2023 - link
Why you always seem high? Do you even fact check beforehand?Dante Verizon - Saturday, July 29, 2023 - link
Do you live in a parallel world? Not even smartphone CPUs use that littlelemurbutton - Saturday, July 29, 2023 - link
Except that it does. It's hard to believe but that's why the Air can be fanless and be faster than the fastest AMD laptop chip.iphonebestgamephone - Saturday, July 29, 2023 - link
I thought it was because they let it run hot and throttle when its almost burning.Qasar - Saturday, July 29, 2023 - link
again, post a link. prove it.i also notice, you seem to ONLY make these BS comparison claims to amd, i guess your mighty apple, doesnt compare to intel ?
Shizuma - Friday, July 28, 2023 - link
Really wish they just made the laptop 3D v-cache a single CCD 8 core chip but with boost frequencies in line with what they have on the 7945HX3D as if they are targeting the best gaming chip then that would have been the far better option, and cheaper on top of it all, kind of annoying that we're back to Intel and AMD chasing each other with core/thread counts again.brucethemoose - Friday, July 28, 2023 - link
The 8 core phoenix silicon is fabbed for lower frequency, and it (to my knowledge) does not have the TSVs needed for stacked cache.meacupla - Friday, July 28, 2023 - link
At this point, AMD keeps surprising me, so I will have to guess that they do have a 7745HX3D in the pipeline and haven't announced it yet.Just like how they released 7800X3D well after all the early adopters bought a 7950X3D.
James5mith - Friday, July 28, 2023 - link
The first slide says 144MB of cache, but the entire article and the tables in it show 128MB of cache. Which is it?MrCommunistGen - Friday, July 28, 2023 - link
There's 128MB of L3 cache. The 144MB is counting L2 + L3 cache. This is something AMD started doing in their marketing for the last couple years. It's a bit confusing, but it allows them to use an even larger number to advertise total cache.erinadreno - Friday, July 28, 2023 - link
I think that makes sense. AMD and Intel are using non-inclusive L3 cache right now. So the contents in L2 does not necessarily get duplicated in L3 (like in pre-Ice Lake days), the chip gets all the cache to hold different valuespatel21 - Friday, July 28, 2023 - link
I would love for a chip like this to be available in Framework 16 as an option.lmcd - Friday, July 28, 2023 - link
Unlikely. This is a repackaged desktop part, as the article described. The Framework 16 uses true mobile parts, which have completely different (and easier to integrate) constraints. Some things you can expect to be different: mobile chips have lower z-height, more manageable power delivery requirements, better thermal management, and fewer external needs (both traces and additional chips).You should only expect to see desktop replacement chips in thicker 16 and 17 inch laptops. The Framework 16 is far too slim to support a desktop replacement chip.
SaolDan - Saturday, July 29, 2023 - link
I've been coming to this site for many years now and while don't really post or contribute with meaningful comments I've always seen lemurbutton posts as no more than an apple fan making comments to stir the pot. I've never taken him seriously in any way.PeachNCream - Monday, July 31, 2023 - link
Oh let him troll them. If the regulars here haven't figured it out yet, then they deserve what they get. Hang out off to the side and enjoy watching them squabble.PeachNCream - Monday, July 31, 2023 - link
The TDP is troublesome, but of course I would like to see more cache for the CPU land in mobile chips, but the 4 - 15W TDP range that is passively cooled and has high unplugged endurance is more meaningful to me than any DTR chip. So while a few people that are heavily invested in playing games might find these useful, the rest of us are still waiting while capability creeps down the stack to more mundane systems.lmcd - Saturday, August 12, 2023 - link
I'm not going to say never, but I don't think this will hit mobile chips unless a competitor forces AMD to. AMD has admitted that the stacked cache was a server component that turned out to be good for gaming. The technology is likely to be developed for better server integration for a long time before it gets adapted to fit with mobile packaging.Bruzzone - Wednesday, August 9, 2023 - link
Reconciling channel data with AMD q2 10K it appears ASUS is AMD gaming division revenue in q2 at $1.581 B. ASUS saves the quarter and well planned by AMD filling a product revenue void. That scores ASUS 2,535.029 7945X3D and an equal number of Rog Ally 7840U at gross $289 each. mbsnakeater - Sunday, August 13, 2023 - link
Would be awesome to see 3D V-Cache on Strix Halo class APU