Monday, June 17th 2019

Intel "Ice Lake" IPC Best-Case a Massive 40% Uplift Over "Skylake," 18% on Average

Intel late-May made its first major disclosure of the per-core CPU performance gains achieved with its "Ice Lake" processor that packs "Sunny Cove" CPU cores. Averaged across a spectrum of benchmarks, Intel claims a best-case scenario IPC (instructions per clock) uplift of a massive 40 percent over "Skylake," and a mean uplift of 18 percent. The worst-case scenario sees its performance negligibly below that of "Skylake." Intel's IPC figures are derived entirely across synthetic benchmarks, which include SPEC 2016, SPEC 2017, SYSMark 2014 SE, WebXprt, and CineBench R15. The comparison to "Skylake" is relevant because Intel has been using essentially the same CPU core in the succeeding three generations that include "Kaby Lake" and "Coffee Lake."

A Chinese tech-forum member with access to an "Ice Lake" 6-core/12-thread sample put the chip through the CPU-Z internal benchmark (test module version 17.01). At a clock-speed of 3.60 GHz, the "Ice Lake" chip allegedly achieved a single-core score of 635 points. To put this number into perspective, a Ryzen 7 3800X "Matisse" supposedly needs to run at 4.70 GHz to match this score, and a Core i7-7700K "Kaby Lake" needs to run at 5.20 GHz. Desktop "Ice Lake" processors are unlikely to launch in 2019. The first "Ice Lake" processors are 4-core/8-thread chips designed for ultraportable notebook platforms, which come out in Q4-2019, and desktop "Ice Lake" parts are expected only in 2020.
Source: WCCFTech
Add your own comment

153 Comments on Intel "Ice Lake" IPC Best-Case a Massive 40% Uplift Over "Skylake," 18% on Average

#101
trparky
Boost speeds are nice and all but only if the boost speeds last longer than a second or two. What many of us need is sustained and consistent performance and not a couple of short boosts in speed when the time is right and the chip is sufficiently cool enough to be able to do it.

And then we have the benchmarks that yeah, they show amazing numbers but how do those numbers translate into real-world performance? It's great if you're the benchmark king but at the end of the day what really matters is how much work you can get done in a certain time frame. We've even seen it with SSDs, sure the sustained throughput numbers look great but do they necessarily translate to faster loading software? Nope, 4K random reads are still semi the same over the last few years.
Posted on Reply
#102
Steevo
yeeeeman, post: 4066713, member: 127591"
What about this benchmark? https://www.notebookcheck.net/Intel-s-Ice-Lake-i7-1065G7-CPU-briefly-pops-up-on-PassMark-and-outstrips-AMD-s-new-Picasso-Ryzen-7-3750H-APU.424636.0.html
This is Passmark. Single thread score at ~4.8 Ghz (short test, might actually run at 4.8Ghz) of 8665U is 2400 points. 1065G7 gets 2625 points, at 3.9 Ghz. If we get the 1065G7 to 4.8 Ghz, we get 3200 points. That would translate into 34% higher IPC. Any thoughts? I was also skeptical about the 40% mentioned in this stupid forum picture, but passmark looks a bit more legit to me.
Hardware acceleration is nothing more than knowing the answer to a question, for example building in the logic to "cascade" a predetermined answer for a specific question. Perhaps Intel is investing in more real world accelerated math on board with its own clock domain. If they are advancing as fast with GPU design and parallel acceleration as they claim it might start bleeding over into their own version of a APU.
Posted on Reply
#103
Reeves81x
FordGT90Concept, post: 4065873, member: 60463"
18% over four years is pretty terrible, especially if you consider how much performance was lost because of security mitigations.
Exactly, in fact i suspect that these are benched against chips with full mitigation, so really who knows what the ipc improvement is over non mitigated "original" chip. Probably just the result of them FINALLY implementing hardware based mitigation. Intel up to its old tricks again.
Posted on Reply
#104
efikkan
Steevo, post: 4066850, member: 19251"
Hardware acceleration is nothing more than knowing the answer to a question, for example building in the logic to "cascade" a predetermined answer for a specific question. Perhaps Intel is investing in more real world accelerated math on board with its own clock domain. If they are advancing as fast with GPU design and parallel acceleration as they claim it might start bleeding over into their own version of a APU.
You have to remember that IPC metrics are averaged across a wide selection of workloads. Even if one show a 5% gain and another shows a 40% gain, the scores can be completely legit, as different benchmarks stresses different parts of the CPU. The point of a benchmark is (hopefully) to create a realistic and representative workload, not to saturate the CPU.

Sunny Cove will increase many of its resources by 50-100%, including >100% in integer multiplication and division, so we should expect substantial gains in some benchmarks.

Hardware acceleration have nothing to do with knowing the answer beforehand. Hardware acceleration is about implementing pars or whole algorithms in hardware, which is much faster than software running on generic computational hardware. Examples of this includes compression, encryption and video codecs, which CPUs have special instructions for. Another type of "acceleration" is SIMD operations like SSE and AVX. In general, software have to be rewritten and recompiled to use new instructions. Exceptions exist, as the CPU's front-end have a very limited ability to optimize when converting x86 to microoperations, but this is only the case for instructions coming in specific patterns. Like e.g., the CPU can unroll small loops and eliminate instructions, if the loops are tight enough.
Posted on Reply
#105
John Naylor
Tech media needs something to write about in order to generate income. Reading these articles is like reading about fuel efficiency in a world w/o friction for a vehicle that has a prototype engine but the chassis / body exists only on drawing boards. Results in a lot of fanboy chest beating but nothing useful to apply to real world usage.
Posted on Reply
#106
ratirt
efikkan, post: 4067090, member: 150226"
Sunny Cove will increase many of its resources by 50-100%, including >100% in integer multiplication and division, so we should expect substantial gains in some benchmarks.
Will increase by 50--100% its resources? Where did you get that information from?
Posted on Reply
#107
londiste
ratirt, post: 4067394, member: 165024"
Will increase by 50--100% its resources? Where did you get that information from?
https://fuse.wikichip.org/news/2371/intel-sunny-cove-core-to-deliver-a-major-improvement-in-single-thread-performance-bigger-improvements-to-follow/
https://en.wikichip.org/wiki/intel/microarchitectures/sunny_cove
Most caches, buffers and queues get sizable increases, additional store pipe(s) and some additional capabilities in main execution pipes (not only AVX-512).
Posted on Reply
#108
ratirt
londiste, post: 4067406, member: 169790"
https://fuse.wikichip.org/news/2371/intel-sunny-cove-core-to-deliver-a-major-improvement-in-single-thread-performance-bigger-improvements-to-follow/
https://en.wikichip.org/wiki/intel/microarchitectures/sunny_cove
Most caches, buffers and queues get sizable increases, additional store pipe(s) and some additional capabilities in main execution pipes (not only AVX-512).
Thanks for this but can we already tell that this will guarantee better performance for about 18% IPC improvement as it says? I mean don't get me guys wrong, I want intel to buckle up and go balls out with the tech but still this is just on a paper as a theory. I'm sure they have (maybe) tested this somehow but as we know Intel's 10nm still isn't ready. I hope Intel can pull this one off and if it does, that would be one of the greatest improvements for Intel in a decade :) Improvement of the architecture yeah but there's still this damn node process they are working on now.
Posted on Reply
#109
londiste
If you look at the original post, first part of it as well as the first image is results from Intel's first-party testing of Sunny Cove core vs Skylake core which is what their 18% claim is based on. The actual results across several benchmarks (or rather parts of benchmarks) vary from what looks like a 1% loss in worst case to 40% improvement in best case with the rest falling inbetween. Geomean is 18%.


10nm is not ready does not mean Intel themselves are unable to get some working CPUs produced. They can and are known to have working 10nm CPUs for 2 years or maybe even more. 10nm is just not ready for mass production, meaning actual products on shelves.

Edit:
Intel says laptops with 10nm mobile CPUs will be on shelves by the end of year so they must have gotten something working. This is supposedly happening on the improved 10+nm process.
There is no guarantee that IPC increase in % will be there for all use cases. In fact, it definitely will not be, as illustrated by Intel's slide.
Posted on Reply
#110
ratirt
londiste, post: 4067442, member: 169790"
If you look at the original post, first part of it as well as the first image is results from Intel's first-party testing of Sunny Cove core vs Skylake core which is what their 18% claim is based on. The actual results across several benchmarks (or rather parts of benchmarks) vary from what looks like a 1% loss in worst case to 40% improvement in best case with the rest falling inbetween. Geomean is 18%.


10nm is not ready does not mean Intel themselves are unable to get some working CPUs produced. They can and are known to have working 10nm CPUs for 2 years or maybe even more. 10nm is just not ready for mass production, meaning actual products on shelves.

Edit:
Intel says laptops with 10nm mobile CPUs will be on shelves by the end of year so they must have gotten something working. This is supposedly happening on the improved 10+nm process.
There is no guarantee that IPC increase in % will be there for all use cases. In fact, it definitely will not be, as illustrated by Intel's slide.
I'm not saying they can't make 10nm but how efficient it will be (and what yields? and of course the price since we all know Intel likes to bump it). I don't have time to read the entire article to tell if the improvement is also based on the die shrink or just architecture. We know this is the first try of the 10nm for Intel and since they are struggling with this for some time, it is not certain it will be much better than Intel's current 14nm. It will be better for sure but for how much. Besides, Intel has reduced the pipeline(I think that's what you guys mean with the pipeline improvement) That would mean less frequency (or since it is a shrink it will compensate the shorter pipeline). I remember Pentium E or D that had the pipeline increased giving a chance for higher clocks but eventually it went down the drain since Intel didn't put into account the temperature constraints. I hope this product is what it says it will be. 18% is massive for intel and that will also give AMD something to think about :)
Posted on Reply
#111
londiste
ratirt, post: 4067457, member: 165024"
I'm not saying they can't make 10nm but how efficient it will be (and what yields? and of course the price since we all know Intel likes to bump it).
Do we really care about manufacturing efficiency or yields? If they can put a product out with acceptable price, it is their problem how they manufacture it :)
ratirt, post: 4067457, member: 165024"
I don't have time to read the entire article to tell if the improvement is also based on the die shrink or just architecture. We know this is the first try of the 10nm for Intel and since they are struggling with this for some time, it is not certain it will be much better than Intel's current 14nm. It will be better for sure but for how much.
Improvement is based on architecture. Intel has been said for a couple years now that die shrink from their 14nm++ to 10nm will result in lower frequencies.
ratirt, post: 4067457, member: 165024"
Besides, Intel has reduced the pipeline(I think that's what you guys mean with the pipeline improvement) That would mean less frequency (or since it is a shrink it will compensate the shorter pipeline).
Pipeline length does not change. They added/improved pipes, meaning more (parallel) units in execution stage, making CPU "wider".
These are commonly called pipes, I suppose technically they are parallel execution units in a superscalar CPU.
ratirt, post: 4067457, member: 165024"
I hope this product is what it says it will be. 18% is massive for intel and that will also give AMD something to think about :)
Architecture is not the problem, manufacturing it is. Despite them announcing laptops on shelves by the end of year Intel is assumed to be still struggling with 10nm. There is a lot unknown about what they are capable of and when but it is pretty clear Ice Lake/Sunny Cove (and subsequent architectures) will not be a player in the desktop space before end of 2020 or 2021 at earliest.

Intel will go for mobile first because 10nm should boost power efficiency and lacking high end of frequencies will not matter there anyway. Then they will go for Xeons for similar reasons. After they get these in order we can hope for something new on desktop.
Posted on Reply
#112
ratirt
londiste, post: 4067460, member: 169790"
Do we really care about manufacturing efficiency or yields? If they can put a product out with acceptable price, it is their problem how they manufacture it :)
Well if you are talking about the price then you should care. Less yields correspond to higher processor cost meaning you (us customers) will have to pay for it. Knowing Intel's greediness and cost of motherboards it may be pretty costly which I'd rather not see. Although chances for something in a reasonable price (for each person it can vary widely) are slim in my understanding.
londiste, post: 4067460, member: 169790"
Architecture is not the problem, manufacturing it is. Despite them announcing laptops on shelves by the end of year Intel is assumed to be still struggling with 10nm. There is a lot unknown about what they are capable of and when but it is pretty clear Ice Lake/Sunny Cove (and subsequent architectures) will not be a player in the desktop space before end of 2020 or 2021 at earliest.

Intel will go for mobile first because 10nm should boost power efficiency and lacking high end of frequencies will not matter there anyway. Then they will go for Xeons for similar reasons. After they get these in order we can hope for something new on desktop.
So I guess we will not see desktop Sunny Cove this year. Hopefully next year but that's still a big unknown.

londiste, post: 4067460, member: 169790"
Improvement is based on architecture. Intel has been said for a couple years now that die shrink from their 14nm++ to 10nm will result in lower frequencies.
Sure Intel said that. On the other hand AMD shrunk it's Ryzen 2 die and get better frequencies so maybe the changes in the Intel's Sunny Cove pipeline does have an impact on clocks anyway. Or simply the node 10nm is average but intel wants to keep up with AMD in the node department as well.
Posted on Reply
#113
londiste
ratirt, post: 4067490, member: 165024"
Sure Intel said that. On the other hand AMD shrunk it's Ryzen 2 die and get better frequencies so maybe the changes in the Intel's Sunny Cove pipeline does have an impact on clocks anyway. Or simply the node 10nm is average but intel wants to keep up with AMD in the node department as well.
Do not expect Intel to always lie. They don't.
Besides, Intel showing that 10nm will net lower frequencies is admitting a negative aspect of their newer process isn't it?

When it comes to frequencies Zen/Zen+ is a rather low bar in comparison. Intel did/does close to 5GHz boost clocks on their 14++nm while TSMC/GF 14/12nm rarely gets over 4.3GHz. I would say this is much more a foundry question than architecture one. While architecture plays a part the manufacturing process seems to be a much bigger factor in maximum clocks. TSMC and their 7nm seems to also be a bit better than Intel's 10nm in frequencies among other things.

The other side of this is that we really do not know much about the frequency behaviour of Zen2 yet. Not that we do know much about Intel and Ice Lake either. 1065 G7 seems to be the fastest rumored 10nm CPU and rumors say it has a boost clock of 3.9GHz. Whiskey Lake on 14nm++ has 4.6 or 4.8GHz.
Posted on Reply
#114
ratirt
londiste, post: 4067498, member: 169790"
Do not expect Intel to always lie. They don't.
Besides, Intel showing that 10nm will net lower frequencies is admitting a negative aspect of their newer process isn't it?

When it comes to frequencies Zen/Zen+ is a rather low bar in comparison. Intel did/does close to 5GHz boost clocks on their 14++nm while TSMC/GF 14/12nm rarely gets over 4.3GHz. I would say this is much more a foundry question than architecture one. While architecture plays a part the manufacturing process seems to be a much bigger factor in maximum clocks. TSMC and their 7nm seems to also be a bit better than Intel's 10nm in frequencies among other things.

The other side of this is that we really do not know much about the frequency behaviour of Zen2 yet. Not that we do know much about Intel and Ice Lake either. 1065 G7 seems to be the fastest rumored 10nm CPU and rumors say it has a boost clock of 3.9GHz. Whiskey Lake on 14nm++ has 4.6 or 4.8GHz.
Never said Intel lies all the time. Not telling the truth is not lying deliberately.
We don't know what the frequencies will be but assuming Zen shrink to 7nm and boosts clocks and yet intel gets 10nm (half of what AMD did) and the clocks are settled at 3.9Ghz. Maybe that is why they start from mobile processors? Later on they will try to up the clocks a bit for desktops. I mean I think that's the case.
It isn't always the node process that drives the frequency. I can only hope the clocks will get better with maturity of the 10nm process.
Posted on Reply
#115
trparky
Their processors already run as hot as the surface of the sun, adding more cores to satisfy enthusiasts is only going to make their chips run even hotter. And as others have said, chip yields are already bad with their eight-core series of chips, you can tell this what with how much the 9900K costs. Believe me, Intel is no charity here folks; they'll pass the extra costs onto the consumer, guaranteed. With that being said their 10nm stuff is ready for notebooks because mobile chips historically have never been high clocked chips so the lower clocks because of their 10nm process isn't really going to hurt that segment of the market but will definitely hurt the desktop market. People here would cry foul big time if all a sudden their desktop chips can't clock to 5 GHz.

Will Intel solve their 10nm process issues? Probably, but not before AMD takes a good portion of their market share and rightfully so.
Posted on Reply
#116
efikkan
ratirt, post: 4067457, member: 165024"
I'm not saying they can't make 10nm but how efficient it will be (and what yields? and of course the price since we all know Intel likes to bump it).
Intel promised just a month ago that Ice Lake-SP will ship in Q2 2020 on 10nm+. All the chips shipping this year are still on the first version of 10nm, so there is a potential for improvements to the node.

trparky, post: 4067556, member: 170376"
Their processors already run as hot as the surface of the sun, adding more cores to satisfy enthusiasts is only going to make their chips run even hotter. And as others have said, chip yields are already bad with their eight-core series of chips, you can tell this what with how much the 9900K costs. Believe me, Intel is no charity here folks…
Coffee Lake refresh struggled initially with yields, plus a large portion of the production capacity were reserved for other stuff like modems. Intel has done a new stepping, and have now more 14nm capacity than ever reserved for CPUs, and yields are good, so they can sell these at $300 now and still make a profit.
Posted on Reply
#117
ratirt
efikkan, post: 4067618, member: 150226"
Intel promised just a month ago that Ice Lake-SP will ship in Q2 2020 on 10nm+. All the chips shipping this year are still on the first version of 10nm, so there is a potential for improvements to the node.
Yes promised and then it might be delayed since they already have problems with yields(that's a fact now) and they are asking Samsung to help them out. The chips shipping this year are for mobile. Keep that in mind when you think of the 10nm. It will take some time till the desktops are out and Q2 2020 is very optimistic.

efikkan, post: 4067618, member: 150226"
Coffee Lake refresh struggled initially with yields, plus a large portion of the production capacity were reserved for other stuff like modems. Intel has done a new stepping, and have now more 14nm capacity than ever reserved for CPUs, and yields are good, so they can sell these at $300 now and still make a profit.
Bull crap my friend. Same thing. Samsung is going to help with the production cause Intel's yields are bad and they can't pull this off. Intel has way more money from CPU's then modems so there is no way it would sacrifice their cash making CPUs for anything else. This 300$ you are talking about? Since I remember Intel has been asking always a lot for it's chips. Now you think it will be 300? You can't be serious. Especially Samsung is going to produce the chips as well and you think that this is an indication of cheaper CPUs because 3rd party is involved in this? Intel always been having problems with yields. Monolithic architecture is what they got. Their processors are huge. No matter how mature the node process can be there's always defects. Not to mention the die size area is increasing without any additional cores added. AMD is pushing for more cores and yet the die for Intel is going bigger fast without adding cores. They will never sell these at 300$ unless they want to lose money and with Intel that is not possible. I will believe it when I will see it no other option.
I'd bet if they were going to lower prices exponentially, Intel would come up with a lame idea of premium processors (basically same but the premium is there) to justify the high price.
Posted on Reply
#118
londiste
efikkan, post: 4067618, member: 150226"
Coffee Lake refresh struggled initially with yields, plus a large portion of the production capacity were reserved for other stuff like modems. Intel has done a new stepping, and have now more 14nm capacity than ever reserved for CPUs, and yields are good, so they can sell these at $300 now and still make a profit.
Coffee Lake Refresh steppings are not about yields. They are about security mitigations. There have not even been hints about yield issues (for desktop CPUs).
ratirt, post: 4067822, member: 165024"
Intel always been having problems with yields. Monolithic architecture is what they got. Their processors are huge. No matter how mature the node process can be there's always defects. Not to mention the die size area is increasing without any additional cores added. AMD is pushing for more cores and yet the die for Intel is going bigger fast without adding cores.
It is probably safe to assume Intel's 14nm yields are in the same ballpark as TSMC/GF's, right?
I see Intel having a lot of room for lowered prices, if they want that to happen.
- Current Intel's CPUs die sizes are 125mm^2 for 4-core (7700K), 150mm^2 for 6-core (8700K) and 175mm^2 for 8-core (9900K).
- Zen/Zen+ 8-core dies are a little above 200mm^2.
- Zen2 cores chiplet is 75mm^2 but IO Die is 120mm^2.
Posted on Reply
#119
ratirt
londiste, post: 4067838, member: 169790"
Coffee Lake Refresh steppings are not about yields. They are about security mitigations. There have not even been hints about yield issues (for desktop CPUs).
It is probably safe to assume Intel's 14nm yields are in the same ballpark as TSMC/GF's, right?
I see Intel having a lot of room for lowered prices, if they want that to happen.
- Current Intel's CPUs die sizes are 125mm^2 for 4-core (7700K), 150mm^2 for 6-core (8700K) and 175mm^2 for 8-core (9900K).
- Zen/Zen+ 8-core dies are a little above 200mm^2.
- Zen2 cores chiplet is 75mm^2 but IO Die is 120mm^2.
Yes. But you have to remember that chiplets are easier to make and cheaper 'cause of the yields, therefore AMD can sell them cheaper. I hope this post isn't for stating that Zen's dies are bigger. You have to remember that this isn't a monolithic chip and that's why the size of it may be bigger in comparison to the monolithic but each chiplet is very small and that boosts yields enormously.
Plus you have the CCX combinations which basically defective chips can be used as well for lower quality products or/and less cores.
Posted on Reply
#120
londiste
ratirt, post: 4067842, member: 165024"
Yes. But you have to remember that chiplets are easier to make and cheaper 'cause of the yields, therefore AMD can sell them cheaper. I hope this post isn't for stating that Zen's dies are bigger. You have to remember that this isn't a monolithic chip and that's why the size of it may be bigger in comparison to the monolithic but each chiplet is very small and that boosts yields enormously.
Plus you have the CCX combinations which basically defective chips can be used as well for lower quality products or/and less cores.
Zen die is clearly bigger. 2700X die is about 20% bigger than 9900K die.

Zen2 has at least 2 dies. For reasonably close comparison, lets look at 8-core Zen2 like 3800X. Without considering any extra aspects, just the die area - 120+75 is still 10% bigger than 9900K. One of these dies is pretty much the same size as 4-core Intel CPU. The second die is indeed small and awesome. However, from all the forums and threads (including people who should have direct knowledge of the industry) 7nm production cost is estimated to be about 60% higher than 12/14nm and defect rate to be about twice as much as a mature process like 12/14nm. Plus some of the packaging challenges AMD shared on their slides.

I would not expect AMD's 8-core CPU production cost to be smaller than Intel's, I would lean towards the opposite.

Edit:
The problem with monolithic chips comes seriously into play with much larger chip sizes. Skylake-X 10-core (LCC) is 322mm^2, 18-core (HCC) is 484mm^2, 28-core XCC is 698mm^2. This is the sector where AMD already benefits from the ~200mm^2 Zen dies and will benefit further from 7nm Zen2 chiplets.

Defective chips are used for other SKUs by all manufacturers. Intel uses Coffee Lake Refresh dies in 9900K (9900), 9700K (9700) and 9600K. In a bit weird way not in anything lower than these. Depending on how you want to take this, it could also indicate that yields are not a problem.
Posted on Reply
#121
trparky
londiste, post: 4067849, member: 169790"
Defective chips are used for other SKUs by all manufacturers.
Understood, however, there are only so many chips you can relegate to the lower end of the product spectrum due to manufacturing defects until you've eaten too far into your profit margins.
londiste, post: 4067849, member: 169790"
For reasonably close comparison, lets look at 8-core Zen2 like 3800X. Without considering any extra aspects, just the die area - 120+75 is still 10% bigger than 9900K. One of these dies is pretty much the same size as 4-core Intel CPU.
It's not all about the size of the die but the complexity of the die as well. Intel, due to the fact that their architecture is a monolithic architecture meaning everything that the processor needs to function is on one piece of silicon, the chances of things going wrong during the manufacturing phase are exponentially higher when compared to AMD.

AMD has the manufacturing advantage of not having to put everything on one piece of silicon thus the complexity of those dies are far lower than Intel. Splitting the processor into pieces and parts, like one die for the actual processor and another die for the I/O is ingenious. They've now cut the complexity of manufacturing the actual processor dies by nearly 50%, that's a huge chunk.

You, @londiste, sound like an Intel fanboy with your posts here. You can't seem to imagine a world where Intel isn't at the top of the heap, the cream of the crop, etc.
Posted on Reply
#122
londiste
trparky, post: 4068041, member: 170376"
You, @londiste, sound like an Intel fanboy with your posts here. You can't seem to imagine a world where Intel isn't at the top of the heap, the cream of the crop, etc.
Just curious, what makes you think that?
Posted on Reply
#123
trparky
When you have comments like this...
londiste, post: 4067849, member: 169790"
Intel uses Coffee Lake Refresh dies in 9900K (9900), 9700K (9700) and 9600K. In a bit weird way not in anything lower than these. Depending on how you want to take this, it could also indicate that yields are not a problem.
How can you say that there are no yield issues when a 9900K costs $500? The only reason why I can think of why it costs so much is that Intel is having issues making those chips and pricing it so high in hopes that people won't be tempted to buy them when they're in short supply. You also mention that there are manufacturing issues with AMD's products while overlooking the fact the Intel is having to turn to Samsung to make some of their chips. If we go by your thinking Intel would have never had to turn to Samsung. Why else would they turn to Samsung, a direct competitor, if they didn't have yield issues? I can't imagine Intel ever giving up and turning to a competitor if they didn't have manufacturing issues, it would be like GM turning to Ford to make their engines while they (GM) still make the body of the car.
Posted on Reply
#124
efikkan
ratirt, post: 4067822, member: 165024"
Yes promised and then it might be delayed since they already have problems with yields(that's a fact now) and they are asking Samsung to help them out. The chips shipping this year are for mobile. Keep that in mind when you think of the 10nm. It will take some time till the desktops are out and Q2 2020 is very optimistic.
Plans are always a bit fluid with these companies. Yet Intel have made repeated claims up until 1 month ago that Ice Lake-SP will ship in Q2 2020. Even Intel don't know the final verdict until the final stepping arrives, and my guess this will happen very late in 2019, so by then we should expect either a lot of performance "leaks" or some disastrous news, and then we'll know. Intel will get some financial troubles if Ice Lake-SP is cancelled or further postponed, their commitments there are much more serious than the desktop market.

Intel have not promised any desktop 10nm products for Q2 2020. An "Ice Lake-X" might be coming, but I would guess that would be later. Intel have driver support for a desktop "495 series" Ice Lake chipset though, we don't know what this is for.

Whatever Samsung is going to help out with, will be coming later.

ratirt, post: 4067822, member: 165024"
[quote=efikkan, post: 4067618, member: 150226"]
Coffee Lake refresh struggled initially with yields, plus a large portion of the production capacity were reserved for other stuff like modems. Intel has done a new stepping, and have now more 14nm capacity than ever reserved for CPUs, and yields are good, so they can sell these at $300 now and still make a profit.
Bull crap my friend. Same thing. Samsung is going to help with the production cause Intel's yields are bad and they can't pull this off. Intel has way more money from CPU's then modems so there is no way it would sacrifice their cash making CPUs for anything else. This 300$ you are talking about? Since I remember Intel has been asking always a lot for it's chips. Now you think it will be 300? You can't be serious. Especially Samsung is going to produce the chips as well and you think that this is an indication of cheaper CPUs because 3rd party is involved in this? Intel always been having problems with yields. Monolithic architecture is what they got. Their processors are huge. No matter how mature the node process can be there's always defects. Not to mention the die size area is increasing without any additional cores added. AMD is pushing for more cores and yet the die for Intel is going bigger fast without adding cores. They will never sell these at 300$ unless they want to lose money and with Intel that is not possible. I will believe it when I will see it no other option.

I'd bet if they were going to lower prices exponentially, Intel would come up with a lame idea of premium processors (basically same but the premium is there) to justify the high price.[/quote]No, I'm just stating the facts, I never said they will lower the prices that much. i9-9900K will be cheaper to make than 8-core Zen 2 for now.

Intel will lower their prices (or not) depending on how well they are selling.

ratirt, post: 4067842, member: 165024"
[quote=londiste, post: 4067838, member: 169790"]
I see Intel having a lot of room for lowered prices, if they want that to happen.
- Current Intel's CPUs die sizes are 125mm^2 for 4-core (7700K), 150mm^2 for 6-core (8700K) and 175mm^2 for 8-core (9900K).
- Zen/Zen+ 8-core dies are a little above 200mm^2.
- Zen2 cores chiplet is 75mm^2 but IO Die is 120mm^2.
Yes. But you have to remember that chiplets are easier to make and cheaper 'cause of the yields, therefore AMD can sell them cheaper…[/quote]TSMC 7nm is expected to become about twice as expensive per chip size vs. 14nm, but that's when 7nm approaches optimal yields. Currently an 8-core Zen 2 would cost more to make than an i9-9900K, and that's even with factoring in the benefits of the small chiplets.
Posted on Reply
#125
trparky
efikkan, post: 4068075, member: 150226"
TSMC 7nm is expected to become about twice as expensive per chip size vs. 14nm, but that's when 7nm approaches optimal yields. Currently an 8-core Zen 2 would cost more to make than an i9-9900K, and that's even with factoring in the benefits of the small chiplets.
Then how is AMD's chips so much cheaper than Intel's chips? I can't imagine AMD eating the cost, they're not a charity here folks.
Posted on Reply
Add your own comment