How to achieve your goal: Start with a dumpster fire like Kaveri as the baseline. Relative improvements are easy when you can set the bar beneath ground level.
Indeed, and it also means that it can't be expected that they will improve by another factor of 25 in the next five years. That and the fact that it's getting more challenging to get process improvements past 7nm.
Compared to Tiger Lake that is allegedly on an "inferior" process I'm not particularly impressed by anything that AMD has done in mobile at all. That includes blatantly superior GPU performance and CPU parity even in most multi-threaded workloads. On top of that I don't see USB 4.0 anywhere in AMD's mobile stack (or anywhere else).
When you consider that 90% of the improvements being attributed to AMD are really due to simply not using GloFo as a foundry, it really comes into focus that Moore's law isn't really that dead and AMD gets to ride on the coattails of other companies.
Tiger Lake doesn't exist yet. Neither does USB 4.0.
And AMD created an entirely new design and applied it to everything from laptops to high-end servers while Intel was putting out its fourth respin of Skylake.
CajunArson already thoroughly tested a Tiger Lake platform and was positively flabbergasted by the performance, especially when combined with the USB4 devices they had around. They also have those rare GloFo manufactured Ryzens that performed so poorly that the only conclusion is that TSMC's manufacturing makes up 104% of the reasons for Ryzen 4000 series performance. They're basically just K6-2s with slightly smaller wires and whatnot inside. The Intel offering is also cheaper I'm sure.
In the meantime in the real world all we have to judge Tiger Lake is Intel's own marketing. While this may very well turn out to be accurate at this time it's coming from the same company that was demoing seemingly incredible frequencies to gullible journalists not seeing through the smoke, mirrors, and 1HP chillers just a short while back. Also it's comparing a CPU that is "coming soon" to one that's been on the market for months now. Why not compare it to Cézanne (Zen 3 + RDNA2) if we're comparing future products?
Through the magic of the internet and anonymity we get to hear some dope talk with such certainty about some stuff they don't know or understand but heard a rumor.
Are you telling me a future product may be faster than a something currently available? Shock, isn’t that which way tech is supposed to go? Also, you’re a bit clueless if you ask me, first of all, Intel 10nm is equivalent to TSMC 7nm, secondly, they don’t have to rely on multi core count at all, they are in fact currently ahead in IPC with a comparable CPU clocked at the same speed, ironically, it’s actually Intels 14nm fab that’s allowed Intel to be competitive, allowing very high clock speeds.
So let’s look at the here and now, the 4900hs CPU beats every mobile CPU Intel has, at 35w, only when you push the Intel chips to 90w do they start to become competitive, this is a fact, as per every review site. The 4000u series, they’ve managed to place double the amount of cores at the same TDP as Intel, even their 10nm parts and nearly double the performance.
Yet you’re down playing their achievements and banging on about some unreleased Intel part beating their current crop of CPUs. You sound like a very ignorant Intel fan.
What a funny post. AMD have already demonstrated significant improvements in die size and efficiency *on the same manufacturing process*, and your "analysis" completely ignores AMD's measurable progress with idle power efficiency and frequency ramping, not to mention casually bypassing IPC gains which have the square root of fuck-all to do with process.
Pointing to an unreleased product proves nothing besides your desire to only see your favoured team as the "winner".
"... it really comes into focus that Moore's law isn't really that dead and AMD gets to ride on the coattails of other companies." Well, one of the main reasons TSMC is so dominant is ultimately because Intel they were unwilling to go for the lower margin stuff. A lot of this all goes back before the phone explosion, when Intel decided that low margin stuff was beneath them. Then by the time they realized that Atom had to offer more than an excuse for Intel Manufacturing to have something to do with old fabs, it was mostly too late. Eventually, they did realize something hence a few years ago they started dumping $billions into dumping (well the called it "contra-revenue" to hide from any anti-competitive charges) Atom on the market but too late. In the meantime, all the cumulative revenue of all those lower margin vendors meant that eventually - and largely predictably – TSMC were able to beat Intel Manufacturing. Another reason this is significant is back in the Athlon64 days, Intel were able to keep the P4 relevant by having a better process. Now they are playing catch-up in manufacturing and while Zen isn’t as radically ahead as A64 was, it is a very solid foundation and even though masks are running to $100 millions now, the chiplet design enables a smaller company to compete very well.
It does just show that the ultimate metric is price.. that AMD were able to carry on even when shovelling out underpar like Kaveri/carrizo simply becoz the price is right. They're a bit 2faced regarding ARM, at one point it was the future. It still seems to me that ARM will overtake all, except the very HEDT.. especially now that APPLE has signalled the switch. Kudos to AMD tho they needed the power efficiency most of all
Ignore Kaveri. The subsequent year to year improvements thereafter are not as terrible as you make it out to be. Do all processors have ONLY the same time duration parameters defined in E? Also, eq E is not weighted as shown. How did you calculate the idle efficiency?
The baseline was from a target with a new ceo, lisa was elected ceo in 2014, she was probably making bold claims of what to expect as a ceo usually do. Kaveri just happened to be the product at that time.
Let's flip your assessment: Producing an architecture that's this efficient, performant and competitive - from what was effectively a standing start - is one hell of an achievement.
Getting from "bargain-basement craptop chip" to "used in Microsoft Surface devices" within 5 years is fine progress.
Isn‘t that what Intel usually does in their marketing materials, i.e. show that a current laptop with faster memory (not theirs), a faster SSD and a better GPU (nVidia, so again not theirs), implying it‘s due to their CPU which hardly saw a performance increase ?
Anyhow, AMD, a small (compared to their competition) company with limited resources and close to bankruptcy set a goal *for the future* in 2014 and they exceeded that goal, so cudos to them.
Second paragraph from the end, I think you meant to say they're *not* going to be resting on their laurels anytime soon. Thanks for the article and the update, these have been improving years for computing to be sure.
"Going forward, AMD plan to continue to be lead partners on upcoming process node technologies." Missing "s": "Going forward, AMD plans to continue to be lead partners on upcoming process node technologies."
Actually, if when you say "AMD," you are speaking of a number of people inside AMD, then "AMD plan" is correct, as in, "The people inside AMD plan...", etc.
This illustrates why it's doubtful Intel will be catching AMD anytime soon, imo--while Intel was milking its x86 monopoly, AMD was working the chart Ian created for this article. And so it goes. AMD will be a moving target for Intel for the foreseeable future.
To expand on that: it's generally an Americanism to think of a company as if they were an individual. You'll notice all British publications will use the plural forms.
It's not about companies being individuals, it's companies being referred to as a singular composite entity. For example we also say "New York plans to introduce a new subway system". Do you in England say "London plan to increase policy presence" or "London plans to increase police presence"? London is a collective of people just like AMD is a collective of people. It makes sense to use the same verb form when referring to such collective entities. Which is why we say "AMD plans ..."
@ian is CPU performance measured by the INT or FLOAT unit? I ask because Zen2 did a lot of work on the FPU, but no mention was made on the efforts to improve the INT unit.
And thanks for the article, Ian! I've been wondering about this!
" What's wrong with using plain old 8.76?" Nothing. What's wrong with "8760/1000"? It makes it clear where the value comes from (8760 hours in a year and 1000 W = 1 kW and the result being in kWh). Pretty normal way to phrase general engineering calculations. Very intuitive even for laymen, I would assume.
Maybe not skip the actual words of the article? :D It is being calculated by using a real test platform, so we get actual power usage here, not TDP definitions.
Intel's Lakefield and Alder Lake approach with big/small cores (copying ARM's big.LITTLE/DynamIQ) looks like it could work really well. AMD should do the same with a Jaguar-like small core. For desktop, they could put 32 of them in a separate chiplet or something like that.
Yes it is promising but it’s typical Intel in the sense it’s really stingy: only 1 big core and 4 little cores. At the very least, they should have included 2 big cores. No wonder Apple is ceasing their relationship with them. Rumour has it their first Mac ARM chip will be 8 big cores and 4 little cores.
It's stingy, but it can be improved later on and it might perform better than expected. It also has DRAM stacked on the CPU, which could help it punch above its weight. Benchmarks and reviews next week?
The Atom cores are starting to get pretty powerful anyway. The upcoming Gracemont cores should perform like Ivy Bridge, except a lot smaller.
The 1:4 ratio seems good too, with the 4 small cores taking up the die space of about 1 large core. I hope we see 2+8, 4+16, and 8+32.
They'll perform like Ivy, but the clocks are going to be low, so they'll still be significantly slower than the majority of Ivy Bridge systems that ever shipped. I have a couple of devices with similarly-clocked dual core Ivy chips on them and they're not really suitable for daily usage anymore - so a lot of it's going to be down to how parallel the user's tasks are, and how well that one Sunny Cove core is being used to improve UI snappiness.
It depends what your priorities are. This initial Lakefield is both a technology demonstrator and a product for low power/ultraportable devices. Even Apple doesn't ship its current SoC with more than 2-3 big cores. But, I agree that an actual head-to-head comparison would be interesting (despite all the caveats).
Did AMD provide the settings those individual CPUs were running at? Several are running significantly below their real-world power levels, so may be noticeably undervolted compared to the chips you encounter in actual devices.
I must have missed it, where did you see "several [CPUs] running significantly below their real-world power levels"? I didn't see any absolute power usage numbers for the CPUs.
From TFA: "Note that some of these platforms are not running at their standard TDP designations."
e.g. the 4800H from it's default of 45W was run at 35W, the 2800H from it's default of 45W was run at 35W, the FX-9830P from it's default of 45W was run at 35W. The 3750H, 2700U, FX-8800P, and FX-7600P were run at stock.
AMD definitely improved their CPU performance in leaps and bounds since Kaveri (I always mislabeled it as "Kadaveri", because it was dead to me). From my POV, the one big fly in the Renoir ointment is AMD's decision to not make Renoir APUs available to those of us who like to build our own - it'll be OEM only, at least for the time being. That's a real bummer, as I had my eyes on one for an HTPC project. Now it's either wait and see or go Intel; neither one is a good alternative.
I've been waiting on one for my own HTPC (case, HSF and PSU exist already - looking to upgrade mobo, CPU and RAM). I'll probably just bite the bullet and go for one of the older APUs instead, see how far I can tweak it!
I read your first paragraph and I couldn't understand what the metric was, are you saying 25X increase in perf/Watt?
I had to scroll down a ton to see some table with some weird formula, maybe some equation with capacitance? Maybe say in your first paragraph you're using some hybrid CPU/GPU formula or something because its not obvious what you're talking about.
I do know in general AMD CPUs are dirty cheap right now and super fast, but its not 25X cheaper or 25X faster than the 2014 CPU....maybe perf/watt?
This compute metric seems random by spliting it 50/50 between CPU/GPU. Just throwing numbers if A improves CPU/GPU by 10/1000x but B improves CPU/GPU by 100x each, I would argue B is superior (still majority of workloads are CPU dependent), but metric will highlight A as a superior result. Am I missing anything?
"Am I missing anything?" Yes, the part where what you said matters in any way shape or form to their original goal? The article clearly states that AMD thinks (thought) heterogenous system architecture was going to be a big thing. So a 50/50 split made sense then. If all you care about is CPU increases, then by all means dismiss that side of the equation. With 25x20, AMD didn't set out to show the world a distinct increase in perf./W according to industry accepted metrics that take into account each user activity spread over a whole year and divide that neatly into GPU and CPU chunks and measure and test accordingly. They set out to improve the efficiency of their APUs for 'gaming and compute' in a broad, easily understood manner. Don't be sad that not everything is catered specifically to you.
I'd be safe, just expect 15% performance and assume that. 15% performance gen over gen versus sandy->kaby is a nice gain so it's not very bad. Maybe games are up to 25% ? like farcry and starcraft 2. conservative expectations are safest.
Facinating how C and E are almost the same within the same process node (counting 14 and 12 nm as the same). Except the starting point, Kaveri, that has an E that are way behind the rest of the 28 nm products. Why is that?
I believe I found the source. Charrizo was AMDs first APU to support Heterogeneous System Architecture (HSA), the dynamic shifting the power budget between CPU and GPU. I.E. Using 30W CPU power in Cinebench and 30 W GPU power in 3DMark, yet still have a roof of 35W total in both cases. Kaveri did not have that feature, meaning it would score far worse on power efficiency (E).Carrizo are also their fist APU to use Adaptive voltage and frequency scaling.
I don't like Intel but they're going to be serious now that Apple is a competitor. It is going to be interesting since 2015. Of course, AMD will not be far behind.
The comments regarding ARM seem incredibly naïve. It's only the powerful force of inertia keeping x86 in the game right now. It was clear that's changing, well before Fujitsu and Apple became the ARM standard-bearers this week.
Intel and AMD are both in trouble. ARM's new X program is the foundation of taking over mainstream desktop CPU's. Yes, ARM won't have performance monsters like the i7/9 and Ryzen x8/9xx cores, but those are 10% of all CPUs being sold. 90% are i3/i5 and lower equivalent. Not to mention, that laptops/ultrabooks want battery life and connectivity above all and next-gen Qualcomm products will demolish AMD and Intel in that space with their beefed up 875 variant for PCs
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
82 Comments
Back to Article
CajunArson - Thursday, June 25, 2020 - link
How to achieve your goal: Start with a dumpster fire like Kaveri as the baseline. Relative improvements are easy when you can set the bar beneath ground level.mjz_5 - Thursday, June 25, 2020 - link
thats the CPU they had at the timequadibloc - Wednesday, July 15, 2020 - link
Indeed, and it also means that it can't be expected that they will improve by another factor of 25 in the next five years. That and the fact that it's getting more challenging to get process improvements past 7nm.Sahrin - Thursday, June 25, 2020 - link
It's not just "25x faster" - it's power efficiency as well.Compared with June 2014 competition, Kaveri socred 235 on r15nt, and Haswell-U scored 291.
AMD took that CPU from 2014, and soundly beats Comet Lake with its successor - in *both* power *and* performance.
Alistair - Thursday, June 25, 2020 - link
exactlyCajunArson - Thursday, June 25, 2020 - link
Compared to Tiger Lake that is allegedly on an "inferior" process I'm not particularly impressed by anything that AMD has done in mobile at all. That includes blatantly superior GPU performance and CPU parity even in most multi-threaded workloads. On top of that I don't see USB 4.0 anywhere in AMD's mobile stack (or anywhere else).When you consider that 90% of the improvements being attributed to AMD are really due to simply not using GloFo as a foundry, it really comes into focus that Moore's law isn't really that dead and AMD gets to ride on the coattails of other companies.
PixyMisa - Thursday, June 25, 2020 - link
Tiger Lake doesn't exist yet. Neither does USB 4.0.And AMD created an entirely new design and applied it to everything from laptops to high-end servers while Intel was putting out its fourth respin of Skylake.
close - Monday, June 29, 2020 - link
CajunArson already thoroughly tested a Tiger Lake platform and was positively flabbergasted by the performance, especially when combined with the USB4 devices they had around. They also have those rare GloFo manufactured Ryzens that performed so poorly that the only conclusion is that TSMC's manufacturing makes up 104% of the reasons for Ryzen 4000 series performance. They're basically just K6-2s with slightly smaller wires and whatnot inside. The Intel offering is also cheaper I'm sure.In the meantime in the real world all we have to judge Tiger Lake is Intel's own marketing. While this may very well turn out to be accurate at this time it's coming from the same company that was demoing seemingly incredible frequencies to gullible journalists not seeing through the smoke, mirrors, and 1HP chillers just a short while back. Also it's comparing a CPU that is "coming soon" to one that's been on the market for months now. Why not compare it to Cézanne (Zen 3 + RDNA2) if we're comparing future products?
Through the magic of the internet and anonymity we get to hear some dope talk with such certainty about some stuff they don't know or understand but heard a rumor.
GigaFlopped - Friday, June 26, 2020 - link
Are you telling me a future product may be faster than a something currently available? Shock, isn’t that which way tech is supposed to go? Also, you’re a bit clueless if you ask me, first of all, Intel 10nm is equivalent to TSMC 7nm, secondly, they don’t have to rely on multi core count at all, they are in fact currently ahead in IPC with a comparable CPU clocked at the same speed, ironically, it’s actually Intels 14nm fab that’s allowed Intel to be competitive, allowing very high clock speeds.So let’s look at the here and now, the 4900hs CPU beats every mobile CPU Intel has, at 35w, only when you push the Intel chips to 90w do they start to become competitive, this is a fact, as per every review site. The 4000u series, they’ve managed to place double the amount of cores at the same TDP as Intel, even their 10nm parts and nearly double the performance.
Yet you’re down playing their achievements and banging on about some unreleased Intel part beating their current crop of CPUs. You sound like a very ignorant Intel fan.
brantron - Friday, June 26, 2020 - link
Uh...AMD does use Global Foundry. This is one of the most confused comments I think I've ever seen lol.csutcliff - Friday, June 26, 2020 - link
only for the i/o die in non-mobile chips, zen2 cores/apu & Navi/Vega7nm are fabbed at TSMCGigaplex - Saturday, June 27, 2020 - link
They only use Global Foundries for the IO die. The CPU is TSMC.Spunjji - Friday, June 26, 2020 - link
What a funny post. AMD have already demonstrated significant improvements in die size and efficiency *on the same manufacturing process*, and your "analysis" completely ignores AMD's measurable progress with idle power efficiency and frequency ramping, not to mention casually bypassing IPC gains which have the square root of fuck-all to do with process.Pointing to an unreleased product proves nothing besides your desire to only see your favoured team as the "winner".
KompuKare - Friday, June 26, 2020 - link
"... it really comes into focus that Moore's law isn't really that dead and AMD gets to ride on the coattails of other companies."Well, one of the main reasons TSMC is so dominant is ultimately because Intel they were unwilling to go for the lower margin stuff.
A lot of this all goes back before the phone explosion, when Intel decided that low margin stuff was beneath them. Then by the time they realized that Atom had to offer more than an excuse for Intel Manufacturing to have something to do with old fabs, it was mostly too late.
Eventually, they did realize something hence a few years ago they started dumping $billions into dumping (well the called it "contra-revenue" to hide from any anti-competitive charges) Atom on the market but too late.
In the meantime, all the cumulative revenue of all those lower margin vendors meant that eventually - and largely predictably – TSMC were able to beat Intel Manufacturing.
Another reason this is significant is back in the Athlon64 days, Intel were able to keep the P4 relevant by having a better process. Now they are playing catch-up in manufacturing and while Zen isn’t as radically ahead as A64 was, it is a very solid foundation and even though masks are running to $100 millions now, the chiplet design enables a smaller company to compete very well.
MASSAMKULABOX - Saturday, June 27, 2020 - link
It does just show that the ultimate metric is price.. that AMD were able to carry on even when shovelling out underpar like Kaveri/carrizo simply becoz the price is right. They're a bit 2faced regarding ARM, at one point it was the future. It still seems to me that ARM will overtake all, except the very HEDT.. especially now that APPLE has signalled the switch. Kudos to AMD tho they needed the power efficiency most of allneblogai - Thursday, June 25, 2020 - link
Should they have chosen some other mainstream laptop chip of theirs? Which one?logainofhades - Thursday, June 25, 2020 - link
So they should start their goal, with a CPU that wasn't even out at the time then?psychobriggsy - Thursday, June 25, 2020 - link
That's what they had, and certainly the 25x20 was recognition that it wasn't good enough and they needed this.I guess they could have started with an even slower 15W Kabini of course. I wonder how that compares on this metric to a 15W 4800U.
YB1064 - Thursday, June 25, 2020 - link
Ignore Kaveri. The subsequent year to year improvements thereafter are not as terrible as you make it out to be. Do all processors have ONLY the same time duration parameters defined in E? Also, eq E is not weighted as shown. How did you calculate the idle efficiency?mode_13h - Thursday, June 25, 2020 - link
It's not as if they intentionally made Kaveri bad, just so they would have a low baseline to improve from.oleyska - Monday, June 29, 2020 - link
The baseline was from a target with a new ceo, lisa was elected ceo in 2014, she was probably making bold claims of what to expect as a ceo usually do.Kaveri just happened to be the product at that time.
Cellar Door - Thursday, June 25, 2020 - link
Just like every other self proclaimed expert on the internet - I'm sure you could build a better one in your shed.Meteor2 - Saturday, June 27, 2020 - link
self-proclaimedSpunjji - Friday, June 26, 2020 - link
Let's flip your assessment: Producing an architecture that's this efficient, performant and competitive - from what was effectively a standing start - is one hell of an achievement.Getting from "bargain-basement craptop chip" to "used in Microsoft Surface devices" within 5 years is fine progress.
sseemaku - Thursday, June 25, 2020 - link
The invention of this marketing style should be credited to apple. Compare the new soc with 5 generations old one and make the graph look exponential!R0H1T - Thursday, June 25, 2020 - link
Yeah except AMD "presented" this goal back in 2014 (2015?) & set out to achieve it! The findings are from AT, AMD isn't trumpeting their own horn 🙄Irata - Thursday, June 25, 2020 - link
Isn‘t that what Intel usually does in their marketing materials, i.e. show that a current laptop with faster memory (not theirs), a faster SSD and a better GPU (nVidia, so again not theirs), implying it‘s due to their CPU which hardly saw a performance increase ?Anyhow, AMD, a small (compared to their competition) company with limited resources and close to bankruptcy set a goal *for the future* in 2014 and they exceeded that goal, so cudos to them.
It‘s not something they came up with now.
Spunjji - Friday, June 26, 2020 - link
I had forgotten about Intel's "lookit are aewsoem gameng performances* (*with an Nvidia GPU)" slides. Mendacity pays! :Dksec - Thursday, June 25, 2020 - link
You should try look further back then presentation in the past 5 years son.MrSpadge - Thursday, June 25, 2020 - link
Those 3x increases every major iteration (every ~2.5 years) is by definition an exponential.Kaveri to Carrizo was 3.5x,
Bristol to Raven was 2.2x,
then Picasso to Renoir was 2.92x
PixyMisa - Thursday, June 25, 2020 - link
The graph *is* exponential.Spunjji - Friday, June 26, 2020 - link
That's still impressive. I don't know why you think progress like that is inevitable - AMD spent 7 years from 2007 to 2014 proving that it isn't.Moizy - Thursday, June 25, 2020 - link
Second paragraph from the end, I think you meant to say they're *not* going to be resting on their laurels anytime soon. Thanks for the article and the update, these have been improving years for computing to be sure.Moizy - Thursday, June 25, 2020 - link
*important* - dang autocorrectipkh - Thursday, June 25, 2020 - link
Minor typo in next to last paragraph. Is AMD sitting on their laurels or not?I was told that AMD is going to sit on its laurels any time soon, regardless of where it sits in the competitive landscape.
1_rick - Thursday, June 25, 2020 - link
While we're at it, the phrase is usually "rest on [one's] laurels" or "sit back and rest on" them.YB1064 - Thursday, June 25, 2020 - link
Typo: "I was told that AMD is going to sit on its laurels any time soon, regardless of where it sits in the competitive landscape."Should read: "is NOT going to..."
yogeshm.007 - Thursday, June 25, 2020 - link
"I was told that AMD is going to sit on its laurels any time soon, regardless of where it sits in the competitive landscape."I'm really sad to hear that.
RinzImpulse - Thursday, June 25, 2020 - link
R7-2800Humm... did I miss something?
Slash3 - Thursday, June 25, 2020 - link
That's the older model.https://www.amd.com/en/products/apu/amd-ryzen-7-28...
neogodless - Thursday, June 25, 2020 - link
I believe Ryzen 4800H is a 45W TDP, although the 4800HS and 4900HS have very similar performance and a 35W TDP.neogodless - Thursday, June 25, 2020 - link
Oops I see the note that it's running at 35W for the tests.ballsystemlord - Thursday, June 25, 2020 - link
Grammar error:"Going forward, AMD plan to continue to be lead partners on upcoming process node technologies."
Missing "s":
"Going forward, AMD plans to continue to be lead partners on upcoming process node technologies."
WaltC - Thursday, June 25, 2020 - link
Actually, if when you say "AMD," you are speaking of a number of people inside AMD, then "AMD plan" is correct, as in, "The people inside AMD plan...", etc.This illustrates why it's doubtful Intel will be catching AMD anytime soon, imo--while Intel was milking its x86 monopoly, AMD was working the chart Ian created for this article. And so it goes. AMD will be a moving target for Intel for the foreseeable future.
mkozakewich - Thursday, June 25, 2020 - link
To expand on that: it's generally an Americanism to think of a company as if they were an individual. You'll notice all British publications will use the plural forms.bji - Thursday, June 25, 2020 - link
It's not about companies being individuals, it's companies being referred to as a singular composite entity. For example we also say "New York plans to introduce a new subway system". Do you in England say "London plan to increase policy presence" or "London plans to increase police presence"? London is a collective of people just like AMD is a collective of people. It makes sense to use the same verb form when referring to such collective entities. Which is why we say "AMD plans ..."ballsystemlord - Thursday, June 25, 2020 - link
@ian is CPU performance measured by the INT or FLOAT unit? I ask because Zen2 did a lot of work on the FPU, but no mention was made on the efforts to improve the INT unit.And thanks for the article, Ian! I've been wondering about this!
mode_13h - Thursday, June 25, 2020 - link
> - CPU Performance from Cinebench R15 nT ScoreThat's going to be mostly FPU performance.
PixyMisa - Thursday, June 25, 2020 - link
Hence the huge leap from the 3750H to 4800H. Twice as many cores and twice the FP width.brucethemoose - Thursday, June 25, 2020 - link
"Such a goal would consider a number of performance aspects, perhaps relating to AI acceleration"Yes please!
But I hope they aren't talking about low-precision zen instructions, as they've been awfully quiet when it comes to RDNA and AI.
Tomatotech - Thursday, June 25, 2020 - link
That equation for E .... why put 8760/1000? What's wrong with using plain old 8.76?Death666Angel - Friday, June 26, 2020 - link
" What's wrong with using plain old 8.76?" Nothing. What's wrong with "8760/1000"? It makes it clear where the value comes from (8760 hours in a year and 1000 W = 1 kW and the result being in kWh). Pretty normal way to phrase general engineering calculations. Very intuitive even for laymen, I would assume.Oxford Guy - Thursday, June 25, 2020 - link
Perhaps this is a stupid question but — does this rely on TDP? I skipped over all the math bits.After all, we all know how fudgy TDP can be.
Death666Angel - Friday, June 26, 2020 - link
Maybe not skip the actual words of the article? :D It is being calculated by using a real test platform, so we get actual power usage here, not TDP definitions.Oxford Guy - Saturday, June 27, 2020 - link
Maybe not use charts that show only the claimed TDP? :Dnandnandnand - Thursday, June 25, 2020 - link
Intel's Lakefield and Alder Lake approach with big/small cores (copying ARM's big.LITTLE/DynamIQ) looks like it could work really well. AMD should do the same with a Jaguar-like small core. For desktop, they could put 32 of them in a separate chiplet or something like that.brucethemoose - Thursday, June 25, 2020 - link
As a plus, the big.LITTLE style scheduling work in Windows/Linux would be pushed along by Intel anyway.nandnandnand - Thursday, June 25, 2020 - link
Lakefield can be the beta test. Apparently it should be available starting July 3 in the Samsung Galaxy Book S.https://www.notebookcheck.net/Samsung-Galaxy-Book-...
andykins - Thursday, June 25, 2020 - link
Yes it is promising but it’s typical Intel in the sense it’s really stingy: only 1 big core and 4 little cores. At the very least, they should have included 2 big cores. No wonder Apple is ceasing their relationship with them. Rumour has it their first Mac ARM chip will be 8 big cores and 4 little cores.nandnandnand - Thursday, June 25, 2020 - link
It's stingy, but it can be improved later on and it might perform better than expected. It also has DRAM stacked on the CPU, which could help it punch above its weight. Benchmarks and reviews next week?The Atom cores are starting to get pretty powerful anyway. The upcoming Gracemont cores should perform like Ivy Bridge, except a lot smaller.
The 1:4 ratio seems good too, with the 4 small cores taking up the die space of about 1 large core. I hope we see 2+8, 4+16, and 8+32.
Spunjji - Friday, June 26, 2020 - link
They'll perform like Ivy, but the clocks are going to be low, so they'll still be significantly slower than the majority of Ivy Bridge systems that ever shipped. I have a couple of devices with similarly-clocked dual core Ivy chips on them and they're not really suitable for daily usage anymore - so a lot of it's going to be down to how parallel the user's tasks are, and how well that one Sunny Cove core is being used to improve UI snappiness.eastcoast_pete - Thursday, June 25, 2020 - link
It depends what your priorities are. This initial Lakefield is both a technology demonstrator and a product for low power/ultraportable devices. Even Apple doesn't ship its current SoC with more than 2-3 big cores. But, I agree that an actual head-to-head comparison would be interesting (despite all the caveats).Meteor2 - Saturday, June 27, 2020 - link
A12X and Z has four big cores.edzieba - Thursday, June 25, 2020 - link
Did AMD provide the settings those individual CPUs were running at? Several are running significantly below their real-world power levels, so may be noticeably undervolted compared to the chips you encounter in actual devices.Death666Angel - Friday, June 26, 2020 - link
I must have missed it, where did you see "several [CPUs] running significantly below their real-world power levels"? I didn't see any absolute power usage numbers for the CPUs.edzieba - Friday, June 26, 2020 - link
From TFA: "Note that some of these platforms are not running at their standard TDP designations."e.g. the 4800H from it's default of 45W was run at 35W, the 2800H from it's default of 45W was run at 35W, the FX-9830P from it's default of 45W was run at 35W. The 3750H, 2700U, FX-8800P, and FX-7600P were run at stock.
eastcoast_pete - Thursday, June 25, 2020 - link
AMD definitely improved their CPU performance in leaps and bounds since Kaveri (I always mislabeled it as "Kadaveri", because it was dead to me).From my POV, the one big fly in the Renoir ointment is AMD's decision to not make Renoir APUs available to those of us who like to build our own - it'll be OEM only, at least for the time being. That's a real bummer, as I had my eyes on one for an HTPC project. Now it's either wait and see or go Intel; neither one is a good alternative.
Spunjji - Friday, June 26, 2020 - link
I've been waiting on one for my own HTPC (case, HSF and PSU exist already - looking to upgrade mobo, CPU and RAM). I'll probably just bite the bullet and go for one of the older APUs instead, see how far I can tweak it!webdoctors - Thursday, June 25, 2020 - link
I read your first paragraph and I couldn't understand what the metric was, are you saying 25X increase in perf/Watt?I had to scroll down a ton to see some table with some weird formula, maybe some equation with capacitance? Maybe say in your first paragraph you're using some hybrid CPU/GPU formula or something because its not obvious what you're talking about.
I do know in general AMD CPUs are dirty cheap right now and super fast, but its not 25X cheaper or 25X faster than the 2014 CPU....maybe perf/watt?
Death666Angel - Friday, June 26, 2020 - link
First paragraph, 2nd sentence: "25x improvement in ‘Performance Efficiency'". I'm no native English speaker, but that seems pretty clear to me.saj4u - Friday, June 26, 2020 - link
This compute metric seems random by spliting it 50/50 between CPU/GPU. Just throwing numbers if A improves CPU/GPU by 10/1000x but B improves CPU/GPU by 100x each, I would argue B is superior (still majority of workloads are CPU dependent), but metric will highlight A as a superior result. Am I missing anything?Death666Angel - Friday, June 26, 2020 - link
"Am I missing anything?" Yes, the part where what you said matters in any way shape or form to their original goal? The article clearly states that AMD thinks (thought) heterogenous system architecture was going to be a big thing. So a 50/50 split made sense then. If all you care about is CPU increases, then by all means dismiss that side of the equation. With 25x20, AMD didn't set out to show the world a distinct increase in perf./W according to industry accepted metrics that take into account each user activity spread over a whole year and divide that neatly into GPU and CPU chunks and measure and test accordingly. They set out to improve the efficiency of their APUs for 'gaming and compute' in a broad, easily understood manner. Don't be sad that not everything is catered specifically to you.deil - Friday, June 26, 2020 - link
I still wait for ZEN3 as it might be lit. IF rumors are true, and they really got another ~15-20% IPC, those chips will be incredible.oleyska - Monday, June 29, 2020 - link
I'd be safe, just expect 15% performance and assume that.15% performance gen over gen versus sandy->kaby is a nice gain so it's not very bad.
Maybe games are up to 25% ? like farcry and starcraft 2.
conservative expectations are safest.
Simen1-Norge - Friday, June 26, 2020 - link
Facinating how C and E are almost the same within the same process node (counting 14 and 12 nm as the same). Except the starting point, Kaveri, that has an E that are way behind the rest of the 28 nm products. Why is that?Simen1-Norge - Friday, June 26, 2020 - link
I believe I found the source. Charrizo was AMDs first APU to support Heterogeneous System Architecture (HSA), the dynamic shifting the power budget between CPU and GPU. I.E. Using 30W CPU power in Cinebench and 30 W GPU power in 3DMark, yet still have a roof of 35W total in both cases. Kaveri did not have that feature, meaning it would score far worse on power efficiency (E).Carrizo are also their fist APU to use Adaptive voltage and frequency scaling.zodiacfml - Friday, June 26, 2020 - link
I don't like Intel but they're going to be serious now that Apple is a competitor. It is going to be interesting since 2015. Of course, AMD will not be far behind.wow&wow - Friday, June 26, 2020 - link
Does it simply mean that AMD's mobile processor chips were 25 times worse than Intel's in energy efficiency 6 years ago but are now competitive?Starting from 7nm Zen 2, is AMD now finally competitive?
MASSAMKULABOX - Saturday, June 27, 2020 - link
Imagine where CPU's would be if it wasnt a cosy duopoly , say ten or twenty co's were competing?Oxford Guy - Saturday, June 27, 2020 - link
Too many competitors would drive cost up and reduce innovation due to the effect of volume on price.Too few competitors, though, is also a big problem.
Meteor2 - Saturday, June 27, 2020 - link
The comments regarding ARM seem incredibly naïve. It's only the powerful force of inertia keeping x86 in the game right now. It was clear that's changing, well before Fujitsu and Apple became the ARM standard-bearers this week.tkSteveFOX - Sunday, July 19, 2020 - link
Intel and AMD are both in trouble.ARM's new X program is the foundation of taking over mainstream desktop CPU's. Yes, ARM won't have performance monsters like the i7/9 and Ryzen x8/9xx cores, but those are 10% of all CPUs being sold. 90% are i3/i5 and lower equivalent. Not to mention, that laptops/ultrabooks want battery life and connectivity above all and next-gen Qualcomm products will demolish AMD and Intel in that space with their beefed up 875 variant for PCs