Just found it, 122mm^2. So a bit bigger, but not a whole lot bigger given it packs much more graphics power, 50% more cache and what else. Maybe they've improved 10nm density or size again?
Ok, those are some interesting number for sure. So, seems like Tiger Lake is close to Renoir in die size. In that equality, GPU performance will be better on TGL, given Renoir is estimated as 28% better than ICL and TGL is said to be 50% better than ICL. On the cores side of things, ICL is ~10% better in IPC compared to Zen 2. TGL is said to add another 10% on that. So that is 20% IPC advantage over Zen 2. In single core, Zen 2 will get a beating for sure, if thermals are better on TGL In MC, I there is no chance that 20% IPC advantage will make up for the 4 cores less, I truly hope they plan a 6 core version of this or even an 8 core.
I have seen a Tiger Lake Xe 2x performance claim vs Ice Lake gen11, even though there are only 50% more execution units. Perhaps Intel can just clock the 10nm+ higher for the extra Xe performance.
It's hilarious you can't do simple maths. Both ICL and Zen2 performances are well benchmarked without a shred of doubt, the only unknown is TGL's IPC and it's an optimization, and by definition it simply won't have a huge IPC lift. Intel has never delivered more than 10% IPC uplift one year after a major new architecture, nobody can. 10% is a pretty good guess. But alas that's more than what your brain could handle.
For zen2 vs icl if your talking about the 4800h realease it's not just an ipc comparison since amd hits 4.2 ghz while icl hits 3.9 ghz so at the same frequency icl has has a 10% ipc advantage
The hilarity of this comment of yours is when its convenient.. intel is good to have high freq, lower ipc (9900k) vs in this case, the opposite is also good for intel? XD
The successor to Renoir will be releasing next year. If Tiger Lake comes out in a year, then its competition may be the next 7nm+ AMD APU. (Don't know the name).
Great article - but why jump to the assumption that it's Golden Cove? Surely Willow Cove seems more likely, as the next step on the microarchitecture roadmap.
Ian, right before CES Intel was expected to reveal a next-gen cooling solution for its Project Athena ultra-light laptops during the show. I haven't heard anything about it afterwards. Did Intel mention anything about it or was it all just a rumor?
I wonder if this Horseshoe bend graphite panel description is related to the new cooling solution...
"When Horseshoe Bend is placed in full-width landscape mode, the cover has a kickstand to hold the screen up. And inside the cover, graphite panels conduct heat away from the processors, letting the laptop run at higher clock speed when the kickstand is out."
Four cores though...If AMD really did substantially close the battery life gap, it's going to get hard to argue with 8 cores in the same 15-28W packages.
Looks like AMD downsized their GPU performance in favour of cpu aggregate raw power. Mainly they gave up on their APU concept, pretty funny to see Intel faster in GPU department. Anyway pretty easy to assume another one or two dies from Intel with 32 EUs and two times the cpus with Tunderbolt 4 away. Intel always had at least three different dies for different needs.
Actual Tiger Lake looks done to satisfy Apple (great GPU and Tunderbolt 4)
Personally I'm good with that tradeoff, if I needed GPU I'd get a dedicated GPU laptop, where my work is 90% CPU hitting. The IGP just needs to be "competent", which both of them are at this point.
Not entirely sure AMD has given up on the concept. Leveraging a shared, common memory address space makes sense given the target market. The removes some CPU-GPU communication bottlenecks.
The big bottleneck is memory bandwidth and I would have thought we'd be seeing some high end mobile part with an HBM stack to resolve that by now. Just going with a shared 16 GB of HBM in the package for mobile would not only solve the bandwidth issue for mobile but also save on board space. Costs are still seemingly too high for this and HBM is a bit of a power hog for mobile vs. LPDDR4.
AMD did some arch changes to VEGA for Renoir. each CU has 60% more performance than last iteration, so Renoir has the equivilant to 18 Picasso VEGA cores
Their GPU *performance* has gone up, not down. The number of CUs went down because they don't need as many of the newer CU designs to reach the same point of being memory bandwidth limited.
Moar Cores! is like the Moar Megapixels! of yesteryear. Additional cores are a benefit to highly threaded tasks, which also tend to be large batch tasks. There, you end up thermally limited in very short order in a laptop, so the theoretical advantage may not actually appear in real world testing.
It takes more power to scale frequency than core counts, so even in a thermally limited environment, all things the same more cores of equal quality will outperform less in aggregate performance.
Your assumption is correct only if the task you are talking about is capable of using more than 8 threads. General usage of thin and light notebooks is not content creation or very multi threaded so tigerlake will do just fine. It won't win benchmarks but as a sense of speed it will be better than Zen 2 because of much higher IPC
What if having 8 high performance cores expands the use of thin and light devices...It does for me, my work is on a biggish Thinkpad P1 but Renoir brings much smaller laptops right up against its performance.
You go on the idea that people use a laptop the way they use a tablet or phone, only doing one thing at a time. Remember in 2017 when Intel was still pushing the 4 core/8 thread 7700k vs. the AMD 6 and 8 core chips? Two years later, anything less than six cores is seen as entry level garbage on the desktop.
For laptops, AMD moves to 6 and 8 core chips, and you are saying that more than four cores isn't important, except for in select situations. The "real world testing" that many Intel lovers keep doing is all about how quickly you can run a single application without anything else running at the same time.
Highly threaded tasks *or* multiple single-threaded tasks. That has almost nothing to do with thermal limitations, though - you're oversimplifying. If it were just more threads = more heat then Intel would be running their 4 TGL cores within a 9W thermal envelope, instead of struggling to get 4 cores to run within the same TDP that AMD will be running 8. :)
I can tell you that the laptops in which these CPUs will end up won't be used for content creation, just in very rare cases. So what it does matter for the target users is single core performance which tigerlake will have. Actually according to preliminary figures it will be some 20% better at same frequency compared to zen 2 and given turbo speeds on zen stop at 4.2ghz we can safely assume tgl will exceed that. From leaks it is said to be 4.3ghz. so all in all this will translate in a much better experience for a general user, that is faster loading web pages, faster light gaming, faster short bursty tasks. I don't really see a problem yet with having only 4 cores in 15w
Intel seems to disagree as they have a 6c/12t u series 10th gen which performs 30 percent better than last years 4 core in multithread but just as good in single core
Please, in future just post "no matter what AMD do Intel will be better" once and leave it at that, instead of cluttering up the thread with this ill-informed dreck. Your 20% number is nonsense and we still have no idea what the clock speeds of TGL will be.
96 EU should easily beat AMD low power APU even without even counting the architecture that Xe brings. Also, Tiger lake is rumored to use LPDDR5 memory
I feel like Tiger lake will beat AMD new APU in GPU and per core performance. While AMD APU is going to win in multi-core performance (intel IPC advantage and higher clocks will not be enough to offset the 8 core advantage that AMD have)
AMD new APU focus more CPU performance than GPU. THey are using 512 vega cores which is less than today APU, but clock speed will be higher. It will be slightly faster than current APU in GPU performance.
Honestly i can't see Renoir H beating actual Coffee Lake cpu line in commanding a discrete GPU. Without a steady turbo clock speed you can't do much. So basically AMD is in a segment of the market with not so high GPU needs and some multithread desires without much performance in single thread. I don't know if they will penetrate much in Intel mobile dominium. Pretty strange choice.
just lol so AMD has an 8 core 7nm chip with a newer GPU coming in the form of a new Navi implementation lets see how they stack up, in the meantime we can buy AMDs current product that beats intels current product, when exactly will this be on the shelf?
like I always say competitive products benefit us all however predicting a product thats probably at least 6-7 months as a minimum off release is the best and therefore the world is blue ....well thats just daft
Given thermals on Zen 2 h series will probably be better than coffee lake h, I expect Zen 2 to close the frequency gap and maybe even beat coffee lake h products in gaming given they can sustain turbo for longer
Yep. Gpu in tgl will be faster. Cpu in single core will be some 20% better which is huge given that even in the multi core era, speed of usage is still dictated by IPC. So I am sure tgl will do just fine if efficiency is better or the same as ice lake.
"Easily beat" might be taking it a bit far, given their TDP issues in Ice Lake. Increasing the width of their GPU will improve performance if they can hold clocks steady, but we don't yet know that. New architecture also requires new drivers, and the last lot still don't perform as well in games as in synthetic benchmarks.
LPDDR5 = rumoured by whom? Last I knew, even standard DDR5 isn't due to arrive in servers until Q4 2020 at the earliest.
I'd be shocked if that was the case. Intel's very rarely (if ever) used the same die for U and H series parts, and they've had more cores on H-series than U-series since Skylake.
Isn't Intel's 64 EU iGPU already bandwidth constricted in most games? So, having 96 EUs isn't going to make it any faster. Meanwhile AMD reduces the number for CU, but increases the clock frequency giving a smaller die for the same pref.
Increasing the GPU clock speed you eat the TDP slowing down the cpu clock speed. This is not Gen 11, it is a brand new solution named Xe. You can't do comparisons without knowing the amount of cache inside the GPU to mitigate the limits of bandwidth. Apple to Orange.
you also cant do comparions on unreleased products with NO performance numbers.. fact vs fiction.. come on gondalf.. your intel fanboyism is really showing the last few days..
"Now we know already from Intel’s disclosures that an Xe graphics unit is different to a Gen graphics unit, with an Xe unit capable of doing SIMT work (working on data on its own) individually or SIMD work (wider vector units) collectively by switching modes through software."
Ian, I'm not sure if they'll do this for the client. They gave up that approach to reduce EU size in the Gen 9 generation.
In 10nm, 2 cores is about 25mm. so 50mm for the cores. 50mm for graphics in ICL, and about 25mm for Thunderbolt etc. In 10nm+, 2 cores is about 25mm.so 50mm for the cores. 75mm for graphics. and about 25mm for Thunderbolt etc. In 14nm, 4 cores + 32 EU's was also 150mm. So about 50mm cores, 75mm EU's. 25mm other. ICL cores have 80 million more transistors per core than SKYL, so 320m more per 50mm area. So an increase of 6.4m transistors per mm about.
I dont think they hit their 2.8x density target of original 10nm
14nm was 220m transistors per core. New core is 300m. 440m per 50mm = 17.6m per mm. 10nm increases it by 6.4m per mm. So 24m per mm. 25% increase in density.
Okay, so I'm going to trust your numbers for a second. At your numbers, that means the transistor density per mm2 on the CPU is 38.70m transistors. Versus 17.6mm. That's 2.19x better. Not 2.8x which was their original 10nm claim.
Imagine the efficiency gains at 5nm in two years. AMD may be able to power a 16-core desktop Ryzen at 25W in a laptop with incredible performance per watt.
Is there die imagery of 14nm chips where ti is possible to identify the thunderbolt segment? It might give some indication as to transistor density changes...
This is an interesting approach from Intel - they are releasing their newest generation chips directly into the mobile platform first - and they have been doing that for some time... admittedly because they seem to be hitting power limits preventing them from making it a full desktop product... but they are really well suited to the mobility solution they are being sold for. Typically AMD is releasing product to the mobile segment that is a half step or full generation behind their desktop solutions - with their latest products still using the Vega arch.
These will go up against AMDs next generation of APU in 2021 - which will be based on RDNA 2 GPU cores and Ryzen 3 CPU Cores on a monolithic 7nm+ die out of TSMC. It will be the first time we will be able to directly compare Ryzen at 7nm+ with Intel on 10nm+. Both companies are touting large IPC gains... which based on the core size of these makes a lot of sense... I wonder if they might be thinking about going for more than 3 threads per core?
Never thought AnandTech would be glorious again as he left big shoes to fill, but I’m glad to say reviews have been great and your CES contributions are excellent. Thank You
Tiger Lake is supposed to have an 8-core 16-thread H-series CPU as well for end-2020, so I presume that these kind of differences (4-core vs 8-core vs xx-core) can be die specific and we see here the 4-core die.
IF this is 2022, when are we going to see 7nm using EUV. I thought that 7nm plans had not been impacted by the 10nm debacle and would follow on quickly from 10nm.
Why don't they call the chip "Ice Lake" or "Tiger Lake" instead of "Core i10 11980 Xe-what-the-freak-ever" and add some nice shots of frozen lakes, and large Cats walking by said frozen lakes on the cover of the boxes...?
Some wasted marketing opportunity here. Take the Epson route, with beautiful landscapes on their cartridge covers.
How are you associating a process version with a core number? I don't understand why you say Tiger Lake = 4 cores when Intel always releases chips at different size and price levels for each "optimization." This is especially the case for more important and widely marketed process versions, which you have suggested Tiger Lake will be. How can you conclude that they will not make dies of other sizes and configurations just based on the 4+IGPU one they decided to show the public?
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
107 Comments
Back to Article
yeeeeman - Monday, January 13, 2020 - link
Very nice article Ian. How does the die size compare to Ice Lake?yeeeeman - Monday, January 13, 2020 - link
Just found it, 122mm^2. So a bit bigger, but not a whole lot bigger given it packs much more graphics power, 50% more cache and what else. Maybe they've improved 10nm density or size again?Ian Cutress - Monday, January 13, 2020 - link
I actually just did a die size table for an article going up tomorrow. I'll copy it in here.nandnandnand - Monday, January 13, 2020 - link
How did the taste compare, and when will you reach Singularity?yeeeeman - Monday, January 13, 2020 - link
Ok, those are some interesting number for sure. So, seems like Tiger Lake is close to Renoir in die size. In that equality, GPU performance will be better on TGL, given Renoir is estimated as 28% better than ICL and TGL is said to be 50% better than ICL. On the cores side of things, ICL is ~10% better in IPC compared to Zen 2. TGL is said to add another 10% on that. So that is 20% IPC advantage over Zen 2. In single core, Zen 2 will get a beating for sure, if thermals are better on TGL In MC, I there is no chance that 20% IPC advantage will make up for the 4 cores less, I truly hope they plan a 6 core version of this or even an 8 core.zodiacsoulmate - Monday, January 13, 2020 - link
Remember there is a huge space for thunderbolt 4 which is not in RenoirJayNor - Monday, January 13, 2020 - link
I have seen a Tiger Lake Xe 2x performance claim vs Ice Lake gen11, even though there are only 50% more execution units. Perhaps Intel can just clock the 10nm+ higher for the extra Xe performance.yeeeeman - Monday, January 13, 2020 - link
Or the perf per Eu îs increased since there are now 8 threads instead of 7 executed by one EU.jOHEI - Monday, January 13, 2020 - link
Someone should teach u middle school math dude. thats not how % work.Also Zen2 = ICL in cinebench for example in single core
TGL will beat Zen2 in Single core marginally while having half the cores. how is that a beating???
plus let's not forget 7nm power consumption is looking than 10nm, so let's wait to see 10nm+...
eek2121 - Monday, January 13, 2020 - link
It's hilarious to me that people are speculating on the performance of unreleased products.Korguz - Monday, January 13, 2020 - link
tell that to gondalf....levizx - Tuesday, January 14, 2020 - link
It's hilarious you can't do simple maths. Both ICL and Zen2 performances are well benchmarked without a shred of doubt, the only unknown is TGL's IPC and it's an optimization, and by definition it simply won't have a huge IPC lift. Intel has never delivered more than 10% IPC uplift one year after a major new architecture, nobody can. 10% is a pretty good guess. But alas that's more than what your brain could handle.eek2121 - Tuesday, January 14, 2020 - link
The Zen 2 cores on the low power parts have much less cache...Daironhorse - Tuesday, January 14, 2020 - link
For zen2 vs icl if your talking about the 4800h realease it's not just an ipc comparison since amd hits 4.2 ghz while icl hits 3.9 ghz so at the same frequency icl has has a 10% ipc advantagetamalero - Tuesday, January 14, 2020 - link
The hilarity of this comment of yours is when its convenient.. intel is good to have high freq, lower ipc (9900k) vs in this case, the opposite is also good for intel? XDKorguz - Tuesday, January 14, 2020 - link
yep.... when amd does it.. its wrong.. when intel does it... its ok...Spunjji - Tuesday, January 14, 2020 - link
This only makes sense if you've not actually read any of the IPC comparisons that exist and just assume whatever outcome benefits Intel.outsideloop - Monday, January 13, 2020 - link
The successor to Renoir will be releasing next year. If Tiger Lake comes out in a year, then its competition may be the next 7nm+ AMD APU. (Don't know the name).Rudde - Tuesday, January 14, 2020 - link
"Intel has confirmed that Tiger Lake will be shipping this year."Korguz - Tuesday, January 14, 2020 - link
untill people can buy it.. its vapor ware....Spunjji - Tuesday, January 14, 2020 - link
I'll eat my hat if TGL significantly (as in, more than 10%) outperforms Renoir in actual games.I'm also not sure on ICL being better than Zen 2 by 10% - indications so far look much close to a draw on that front.
Regardless of the specifics, though, they ought to be more evenly matched than Intel's desktop offerings.
QChronoD - Monday, January 13, 2020 - link
... but what did it taste like?Gondalf - Monday, January 13, 2020 - link
Ummm the cpu blocks are huge... really large, I am surprised. Intel is really serious about increase the per core performance.Ian Cutress - Monday, January 13, 2020 - link
Also AVX-512Kevin G - Monday, January 13, 2020 - link
What was the comparison with Cannon Lake since that also had AVX-512? Key difference would be VNNI for Tiger Lake.SarahKerrigan - Monday, January 13, 2020 - link
Great article - but why jump to the assumption that it's Golden Cove? Surely Willow Cove seems more likely, as the next step on the microarchitecture roadmap.Ian Cutress - Monday, January 13, 2020 - link
That's my mistake. I assumed I correctly remembered what order they went in. I've updated itscorpio187 - Monday, January 13, 2020 - link
Ian, right before CES Intel was expected to reveal a next-gen cooling solution for its Project Athena ultra-light laptops during the show. I haven't heard anything about it afterwards. Did Intel mention anything about it or was it all just a rumor?Ian Cutress - Monday, January 13, 2020 - link
I didn't hear anything.JayNor - Monday, January 13, 2020 - link
I wonder if this Horseshoe bend graphite panel description is related to the new cooling solution..."When Horseshoe Bend is placed in full-width landscape mode, the cover has a kickstand to hold the screen up. And inside the cover, graphite panels conduct heat away from the processors, letting the laptop run at higher clock speed when the kickstand is out."
https://www.cnet.com/news/intel-horseshoe-bend-fol...
Seraphimcaduto - Monday, January 13, 2020 - link
Who ever said scientists couldn’t run? That made my morning!Adonisds - Monday, January 13, 2020 - link
Tiger Lake is 10nm+, Ice Lake is 10nm. Hmm... I know what Cannon Lake was. A mistake!yeeeeman - Monday, January 13, 2020 - link
It was a mass produced fab process hence it oficially never existed.yeeeeman - Monday, January 13, 2020 - link
It wasn'tFreeb!rd - Monday, January 13, 2020 - link
You misspelled mass... should be mess.... ; )Spunjji - Tuesday, January 14, 2020 - link
They failed to mass produce it, so it wasn't mass produced... round of applause for Captain Logic here.tipoo - Monday, January 13, 2020 - link
Four cores though...If AMD really did substantially close the battery life gap, it's going to get hard to argue with 8 cores in the same 15-28W packages.Gondalf - Monday, January 13, 2020 - link
Looks like AMD downsized their GPU performance in favour of cpu aggregate raw power.Mainly they gave up on their APU concept, pretty funny to see Intel faster in GPU department.
Anyway pretty easy to assume another one or two dies from Intel with 32 EUs and two times the cpus with Tunderbolt 4 away. Intel always had at least three different dies for different needs.
Actual Tiger Lake looks done to satisfy Apple (great GPU and Tunderbolt 4)
tipoo - Monday, January 13, 2020 - link
Personally I'm good with that tradeoff, if I needed GPU I'd get a dedicated GPU laptop, where my work is 90% CPU hitting. The IGP just needs to be "competent", which both of them are at this point.Kevin G - Monday, January 13, 2020 - link
Not entirely sure AMD has given up on the concept. Leveraging a shared, common memory address space makes sense given the target market. The removes some CPU-GPU communication bottlenecks.The big bottleneck is memory bandwidth and I would have thought we'd be seeing some high end mobile part with an HBM stack to resolve that by now. Just going with a shared 16 GB of HBM in the package for mobile would not only solve the bandwidth issue for mobile but also save on board space. Costs are still seemingly too high for this and HBM is a bit of a power hog for mobile vs. LPDDR4.
jOHEI - Monday, January 13, 2020 - link
AMD did some arch changes to VEGA for Renoir. each CU has 60% more performance than last iteration, so Renoir has the equivilant to 18 Picasso VEGA coresnandnandnand - Monday, January 13, 2020 - link
"Looks like AMD downsized their GPU performance in favour of cpu aggregate raw power."GPU performance is higher despite less CUs, and higher than Ice Lake.
If Tiger Lake comes out later in 2020, it will compete only briefly against Zen 2 APUs, then Zen 3 APUs will come out.
Rudde - Tuesday, January 14, 2020 - link
Aka repeat of Ice Lake & Renoir release.Korguz - Monday, January 13, 2020 - link
ahhh more speculation and personal option from gondalf ....Spunjji - Tuesday, January 14, 2020 - link
Their GPU *performance* has gone up, not down. The number of CUs went down because they don't need as many of the newer CU designs to reach the same point of being memory bandwidth limited.edzieba - Monday, January 13, 2020 - link
Moar Cores! is like the Moar Megapixels! of yesteryear. Additional cores are a benefit to highly threaded tasks, which also tend to be large batch tasks. There, you end up thermally limited in very short order in a laptop, so the theoretical advantage may not actually appear in real world testing.tipoo - Monday, January 13, 2020 - link
It takes more power to scale frequency than core counts, so even in a thermally limited environment, all things the same more cores of equal quality will outperform less in aggregate performance.yeeeeman - Monday, January 13, 2020 - link
Your assumption is correct only if the task you are talking about is capable of using more than 8 threads. General usage of thin and light notebooks is not content creation or very multi threaded so tigerlake will do just fine. It won't win benchmarks but as a sense of speed it will be better than Zen 2 because of much higher IPCSpunjji - Tuesday, January 14, 2020 - link
Your assumption is only correct if if has much higher IPC. All indications are that if it's higher, it won't be by a long way...tipoo - Monday, May 25, 2020 - link
What if having 8 high performance cores expands the use of thin and light devices...It does for me, my work is on a biggish Thinkpad P1 but Renoir brings much smaller laptops right up against its performance.Targon - Monday, January 13, 2020 - link
You go on the idea that people use a laptop the way they use a tablet or phone, only doing one thing at a time. Remember in 2017 when Intel was still pushing the 4 core/8 thread 7700k vs. the AMD 6 and 8 core chips? Two years later, anything less than six cores is seen as entry level garbage on the desktop.For laptops, AMD moves to 6 and 8 core chips, and you are saying that more than four cores isn't important, except for in select situations. The "real world testing" that many Intel lovers keep doing is all about how quickly you can run a single application without anything else running at the same time.
Rudde - Tuesday, January 14, 2020 - link
Phones are funny. Almost all have 8 cores.Spunjji - Tuesday, January 14, 2020 - link
Highly threaded tasks *or* multiple single-threaded tasks. That has almost nothing to do with thermal limitations, though - you're oversimplifying. If it were just more threads = more heat then Intel would be running their 4 TGL cores within a 9W thermal envelope, instead of struggling to get 4 cores to run within the same TDP that AMD will be running 8. :)yeeeeman - Monday, January 13, 2020 - link
I can tell you that the laptops in which these CPUs will end up won't be used for content creation, just in very rare cases. So what it does matter for the target users is single core performance which tigerlake will have. Actually according to preliminary figures it will be some 20% better at same frequency compared to zen 2 and given turbo speeds on zen stop at 4.2ghz we can safely assume tgl will exceed that. From leaks it is said to be 4.3ghz. so all in all this will translate in a much better experience for a general user, that is faster loading web pages, faster light gaming, faster short bursty tasks. I don't really see a problem yet with having only 4 cores in 15wDaironhorse - Tuesday, January 14, 2020 - link
Intel seems to disagree as they have a 6c/12t u series 10th gen which performs 30 percent better than last years 4 core in multithread but just as good in single coreKorguz - Tuesday, January 14, 2020 - link
yea ok... sure...Spunjji - Tuesday, January 14, 2020 - link
Please, in future just post "no matter what AMD do Intel will be better" once and leave it at that, instead of cluttering up the thread with this ill-informed dreck. Your 20% number is nonsense and we still have no idea what the clock speeds of TGL will be.yeeeeman - Monday, January 13, 2020 - link
As for 45w CPUs, yes, Intel can compete only with comet lake h, that is 8 cores or maybe 10 cores (I hope).JayNor - Monday, January 13, 2020 - link
The SIMD GPU operations are interesting. Is that implemented in the NVDA and AMD GPUs?tipoo - Monday, January 13, 2020 - link
Curiously there's a Sony patent filed by Mark Cerny himself for this, for the PS5 GPU. RDNA 2 seems likely to have the idea.https://twitter.com/LiquidTitan/status/12160399096...
maroon1 - Monday, January 13, 2020 - link
96 EU should easily beat AMD low power APU even without even counting the architecture that Xe brings. Also, Tiger lake is rumored to use LPDDR5 memoryI feel like Tiger lake will beat AMD new APU in GPU and per core performance. While AMD APU is going to win in multi-core performance (intel IPC advantage and higher clocks will not be enough to offset the 8 core advantage that AMD have)
AMD new APU focus more CPU performance than GPU. THey are using 512 vega cores which is less than today APU, but clock speed will be higher. It will be slightly faster than current APU in GPU performance.
Gondalf - Monday, January 13, 2020 - link
Honestly i can't see Renoir H beating actual Coffee Lake cpu line in commanding a discrete GPU. Without a steady turbo clock speed you can't do much.So basically AMD is in a segment of the market with not so high GPU needs and some multithread desires without much performance in single thread. I don't know if they will penetrate much in Intel mobile dominium.
Pretty strange choice.
arashi - Monday, January 13, 2020 - link
2020 product vs 2021 product.Go back into your cave Gondaft.
alufan - Monday, January 13, 2020 - link
just lol so AMD has an 8 core 7nm chip with a newer GPU coming in the form of a new Navi implementation lets see how they stack up, in the meantime we can buy AMDs current product that beats intels current product, when exactly will this be on the shelf?like I always say competitive products benefit us all however predicting a product thats probably at least 6-7 months as a minimum off release is the best and therefore the world is blue ....well thats just daft
phoenix_rizzen - Monday, January 13, 2020 - link
Ryzen Mobile 4000 APUs are still using Vega GPU cores. They're tweaked, updated Vega cores, but still Vega.yeeeeman - Monday, January 13, 2020 - link
Given thermals on Zen 2 h series will probably be better than coffee lake h, I expect Zen 2 to close the frequency gap and maybe even beat coffee lake h products in gaming given they can sustain turbo for longertipoo - Monday, January 13, 2020 - link
>without much performance in single threadIt should be matching ICL and launching earlier than Tiger.
yeeeeman - Monday, January 13, 2020 - link
Yep. Gpu in tgl will be faster. Cpu in single core will be some 20% better which is huge given that even in the multi core era, speed of usage is still dictated by IPC. So I am sure tgl will do just fine if efficiency is better or the same as ice lake.Spunjji - Tuesday, January 14, 2020 - link
"Easily beat" might be taking it a bit far, given their TDP issues in Ice Lake. Increasing the width of their GPU will improve performance if they can hold clocks steady, but we don't yet know that. New architecture also requires new drivers, and the last lot still don't perform as well in games as in synthetic benchmarks.LPDDR5 = rumoured by whom? Last I knew, even standard DDR5 isn't due to arrive in servers until Q4 2020 at the earliest.
Foliage - Monday, January 13, 2020 - link
Will this mean there will be no 6/8 core 45W Tiger Lake H parts?drothgery - Monday, January 13, 2020 - link
I'd be shocked if that was the case. Intel's very rarely (if ever) used the same die for U and H series parts, and they've had more cores on H-series than U-series since Skylake.lefty2 - Monday, January 13, 2020 - link
Isn't Intel's 64 EU iGPU already bandwidth constricted in most games? So, having 96 EUs isn't going to make it any faster. Meanwhile AMD reduces the number for CU, but increases the clock frequency giving a smaller die for the same pref.Gondalf - Monday, January 13, 2020 - link
Increasing the GPU clock speed you eat the TDP slowing down the cpu clock speed.This is not Gen 11, it is a brand new solution named Xe. You can't do comparisons without knowing the amount of cache inside the GPU to mitigate the limits of bandwidth. Apple to Orange.
Korguz - Monday, January 13, 2020 - link
you also cant do comparions on unreleased products with NO performance numbers.. fact vs fiction.. come on gondalf.. your intel fanboyism is really showing the last few days..IntelUser2000 - Monday, January 13, 2020 - link
Au contraire, it seems the LPDDR4x-3733 might be little more than required for Gen 11 GT2(64EU).Of course there has to be drastic architectural improvements because 50% increase in EUs itself won't improve performance by 50%.
HardwareDufus - Monday, January 13, 2020 - link
Ian likes selfies....Arnulf - Monday, January 13, 2020 - link
"slightly hardware for the next 5"Too much automated spell checking can be a bad thing ;-)
IntelUser2000 - Monday, January 13, 2020 - link
"Now we know already from Intel’s disclosures that an Xe graphics unit is different to a Gen graphics unit, with an Xe unit capable of doing SIMT work (working on data on its own) individually or SIMD work (wider vector units) collectively by switching modes through software."Ian, I'm not sure if they'll do this for the client. They gave up that approach to reduce EU size in the Gen 9 generation.
DigitalFreak - Monday, January 13, 2020 - link
You must have some strange oral silicon fixation. I keep seeing pictures of you putting those things in your mouth.Ian Cutress - Monday, January 13, 2020 - link
Don't check my twitter then.MrCommunistGen - Monday, January 13, 2020 - link
Ian just really likes chips(pun intended)
Fataliity - Monday, January 13, 2020 - link
So for the same die size, they fit 4 cores + new graphicson 14nm they have 4 cores + old graphics.
Both are around 150mm2
Fataliity - Monday, January 13, 2020 - link
In 10nm, 2 cores is about 25mm. so 50mm for the cores. 50mm for graphics in ICL, and about 25mm for Thunderbolt etc.In 10nm+, 2 cores is about 25mm.so 50mm for the cores. 75mm for graphics. and about 25mm for Thunderbolt etc.
In 14nm, 4 cores + 32 EU's was also 150mm. So about 50mm cores, 75mm EU's. 25mm other.
ICL cores have 80 million more transistors per core than SKYL, so 320m more per 50mm area. So an increase of 6.4m transistors per mm about.
I dont think they hit their 2.8x density target of original 10nm
Fataliity - Monday, January 13, 2020 - link
14nm was 220m transistors per core. New core is 300m. 440m per 50mm = 17.6m per mm.10nm increases it by 6.4m per mm. So 24m per mm. 25% increase in density.
Fataliity - Monday, January 13, 2020 - link
440m per 25mm** typo.IntelUser2000 - Monday, January 13, 2020 - link
Bzzzt! Your analysis is flawed.Skylake is 122mm2. That's the 14nm "old" CPU and GPU. The CPU portion is 50mm2, and GPU is slightly smaller than 45mm2.
Icelake is 31mm2 for the 4 CPU cores. The greatly expanded GPU is slightly smaller at 41mm2.Thunderbolt is only 13mm2.
Tigerlake is 38mm2 for the 4 CPU cores. The yet again greatly expanded GPU is 46-48mm2.
The GPU portion does decrease in size by 2.6x or so, which is pretty much their claim.
Fataliity - Tuesday, January 14, 2020 - link
Okay, so I'm going to trust your numbers for a second. At your numbers, that means the transistor density per mm2 on the CPU is 38.70m transistors. Versus 17.6mm. That's 2.19x better. Not 2.8x which was their original 10nm claim.Fataliity - Tuesday, January 14, 2020 - link
Isn't 7nm HPC getting about 60-69M T/mm2 on TSMC?Fataliity - Tuesday, January 14, 2020 - link
Correction (Edit) Zen2 is 52M T/mm2. Navi is 41M T/mm2 on the 5700, 5500 is 40.5.M T/mm2. Ice Lake CPU is 38.70m T/mm2.outsideloop - Monday, January 13, 2020 - link
Imagine the efficiency gains at 5nm in two years. AMD may be able to power a 16-core desktop Ryzen at 25W in a laptop with incredible performance per watt.deathBOB - Monday, January 13, 2020 - link
What did it taste like?BerenApJiriki - Monday, January 13, 2020 - link
Is there die imagery of 14nm chips where ti is possible to identify the thunderbolt segment? It might give some indication as to transistor density changes...This is an interesting approach from Intel - they are releasing their newest generation chips directly into the mobile platform first - and they have been doing that for some time... admittedly because they seem to be hitting power limits preventing them from making it a full desktop product... but they are really well suited to the mobility solution they are being sold for. Typically AMD is releasing product to the mobile segment that is a half step or full generation behind their desktop solutions - with their latest products still using the Vega arch.
These will go up against AMDs next generation of APU in 2021 - which will be based on RDNA 2 GPU cores and Ryzen 3 CPU Cores on a monolithic 7nm+ die out of TSMC. It will be the first time we will be able to directly compare Ryzen at 7nm+ with Intel on 10nm+. Both companies are touting large IPC gains... which based on the core size of these makes a lot of sense... I wonder if they might be thinking about going for more than 3 threads per core?
dsplover - Monday, January 13, 2020 - link
Never thought AnandTech would be glorious again as he left big shoes to fill, but I’m glad to say reviews have been great and your CES contributions are excellent. Thank YouTechtree101 - Monday, January 13, 2020 - link
Tiger Lake is supposed to have an 8-core 16-thread H-series CPU as well for end-2020, so I presume that these kind of differences (4-core vs 8-core vs xx-core) can be die specific and we see here the 4-core die.albertmamama - Monday, January 13, 2020 - link
Great Heist Dr. Cutress!Rezurecta - Monday, January 13, 2020 - link
I got this burn guys!Nice Intel. So 10nm yields are so bad that they are giving away useless wafers!
/s
Nice article Ian!
Korguz - Monday, January 13, 2020 - link
damn intel NEEDS to change its code names.... have no idea what cpu is what.. or when its going to be released any more...watersb - Monday, January 13, 2020 - link
The little kid in the Twitter video: "When I grow up, I want to be a tech industry analyst!"wr3zzz - Monday, January 13, 2020 - link
I so want that as a coffee table...tamalero - Tuesday, January 14, 2020 - link
*inserts jokes about the horrible yields on intel's latest gen and how that wafer is worth 0*Speedfriend - Tuesday, January 14, 2020 - link
IF this is 2022, when are we going to see 7nm using EUV. I thought that 7nm plans had not been impacted by the 10nm debacle and would follow on quickly from 10nm.Gonemad - Tuesday, January 14, 2020 - link
Left-side question here:Why don't they call the chip "Ice Lake" or "Tiger Lake" instead of "Core i10 11980 Xe-what-the-freak-ever" and add some nice shots of frozen lakes, and large Cats walking by said frozen lakes on the cover of the boxes...?
Some wasted marketing opportunity here. Take the Epson route, with beautiful landscapes on their cartridge covers.
davidefreeman - Tuesday, January 21, 2020 - link
@gonemad And Cannon Lake would just be a hole in the ice...Powervano - Wednesday, January 15, 2020 - link
Cannon Lake was a 10nm- process.peevee - Wednesday, January 22, 2020 - link
Run off with a former Intel employee and ask them what happened to 10nm and Intel culture in general, in detail.Machinus - Friday, February 7, 2020 - link
How are you associating a process version with a core number? I don't understand why you say Tiger Lake = 4 cores when Intel always releases chips at different size and price levels for each "optimization." This is especially the case for more important and widely marketed process versions, which you have suggested Tiger Lake will be. How can you conclude that they will not make dies of other sizes and configurations just based on the 4+IGPU one they decided to show the public?