BluesFanUK:laugh:

What's funny? you thought they'll just sell a 8 cores @ 4 cores price? even the 8700k is more expensive than the 7700k

nothing new...Intel doing Intel things :)

Liviu Cojocarunothing new...Intel doing Intel things :)

That's the problem, it's not. Intel's thing was tick-tock. When that went down the drain, the woes started. Right after woes started, AMD kicked them in the nuts. For bonus points :D

I don't know what you guys expected.
Only in 2019, with 10nm from Intel and 7nm from TSMC, we can and should expect significant improvements, from Intel and AMD!
Until then, it's just light tuning and more clocks, just like AMD did in Zen+.

dwadeRIP AMD. Discounting their products is all they can do.

lol, what?

Same core counts, 200MHz more frequency, and the FOURTH iteration of the same architecture (So no IPC improvement) means "RIP AMD"?

I really hope I'm just missing some sarcasm, because this honestly might be Intel's weakest refresh ever. It's certainly the least compelling to me.

kingsI don't know what you guys expected.
Only in 2019, with 10nm from Intel and 7nm from TSMC, we can and should expect significant improvements, from Intel and AMD!
Until then, it's just light tuning and more clocks, just like AMD did in Zen+.

I'm not holding my breath for 2019 either. Ice Lake may be something good, but judging by how tight-lipped Intel is about it, it probably isn't. And AMD will milk Zen for years to come, meaning they'll probably mimic what Intel has done with Core - small, incremental improvements.

kingsI don't know what you guys expected.
Only in 2019, with 10nm from Intel and 7nm from TSMC, we can and should expect significant improvements, from Intel and AMD!
Until then, it's just light tuning and more clocks, just like AMD did in Zen+.

Unfortunately it won't be until LATE 2019/Early 2020 for Intel's 10nm... AMD on the other hand will be in the first half of 2019 probably around late March early April for the Ryzen 5/7 3x00 chips. So Intel's 9x00 series is honestly going to compete against Zen 2 based Ryzen chips and this is going to be a joke. Where AMD is probably going to give more cores, IPC, lower power consumption, and higher clockspeeds. Intel is giving a small clockspeed bump of less than 5%. So Intel's first 10nm chips IF they get released late 2019 will be put up against AMD's Zen 3 chips using the 7nm+ or the Ryzen 4x00 series of chips which is expected to be more than just Zen+ to the Zen 2 architecture. For their sake they better hope their 10nm problems are resolved soon...

EDIT: I originally wrote 2020 for the 3x00 Ryzen chips this is wrong and I meant 2020 so I changed it for future readers.. As this is when these will be released 2020 would be the 4x00 Ryzen Zen 3 chips.

CaqdeUnfortunately it won't be until LATE 2019/Early 2020 for Intel's 10nm... AMD on the other hand will be in the first half of 2020 probably around late March early April for the Ryzen 5/7 3x00 chips. So Intel's 9x00 series is honestly going to compete against Zen 2 based Ryzen chips and this is going to be a joke. Where AMD is probably going to give more cores, IPC, lower power consumption, and higher clockspeeds. Intel is giving a small clockspeed bump of less than 5%. So Intel's first 10nm chips IF they get released late 2019 will be put up against AMD's Zen 3 chips using the 7nm+ or the Ryzen 4x00 series of chips which is expected to be more than just Zen+ to the Zen 2 architecture. For their sake they better hope their 10nm problems are resolved soon...

Congrats. Rarely have I read a more documented, full of references piece of opinion. Keep it up. :wtf:

bugCongrats. Rarely have I read a more documented, full of references piece of opinion. Keep it up. :wtf:

I could post plenty of references if you'd like and most of what I wrote isn't much of an opinion. Besides what changes Zen 2 is to make on AMD's Zen chips but yeah..

But for fun here are a few recent relevant ones that were used as a basis for my educated opinion.
Intel delay timeline-> www.anandtech.com/show/12693/intel-delays-mass-production-of-10-nm-cpus-to-2019
AMD Zen roadmap-> www.anandtech.com/show/12233/amd-tech-day-at-ces-2018-roadmap-revealed-with-ryzen-apus-zen-on-12nm-vega-on-7nm

Honestly what hope does Intel have against the Zen 2 chips with the 9700K/9900K? A very hot 14nm chip based on a slightly optimized 4 year old architecture vs a high performance 7nm chip on a new Architecture. If AMD can pull 4.6-4.7Ghz singlecore boost and a 10 - 15% IPC improvement on average Intel will be in trouble for the rest of 2019.

bugAnd AMD will milk Zen for years to come, meaning they'll probably mimic what Intel has done with Core - small, incremental improvements.

Yes, unfortunately AMD seem to plan to do at least five iterations of Zen. They better start planning the successor now, because designing a new major architecture takes >3 years. I'm worried after seeing what they've done with GCN, and knowing their primary focus is custom designs like gaming consoles, I'm worried that the improvements we saw in Zen will be a "one time thing", just like GCN.

It's also important to remember that even though Zen have done some good design choices, most of the improvements over Bulldozer was either correcting "design faults" of Bulldozer or "low hanging fruit". To close the gap further, they need to make a much more advanced front-end/prefetcher, which is much more costly both in terms of development time and die size than just adding another ALU. AMD will also have to step up their game in AVX support, which may not me difficult, but will require a lot of die space, and eventual support for AVX-512 might require cache changes.

CaqdeUnfortunately it won't be until LATE 2019/Early 2020 for Intel's 10nm... AMD on the other hand will be in the first half of 2020 probably around late March early April for the Ryzen 5/7 3x00 chips. So Intel's 9x00 series is honestly going to compete against Zen 2 based Ryzen chips and this is going to be a joke. Where AMD is probably going to give more cores, IPC, lower power consumption, and higher clockspeeds. Intel is giving a small clockspeed bump of less than 5%. So Intel's first 10nm chips IF they get released late 2019 will be put up against AMD's Zen 3 chips using the 7nm+ or the Ryzen 4x00 series of chips which is expected to be more than just Zen+ to the Zen 2 architecture. For their sake they better hope their 10nm problems are resolved soon...

The first half of 2020 is >1.5 years away, a lot can happen by then.
Your scenario of AMD crushing Intel in 2020 relies on a lot of highly speculative assumptions:
- Intel doing nothing for the next >1.5 years, and Intel 10 nm still be as "bad" as today.
- "TSMC 7nm" to be fantastic and for AMD to have all the production capacity they want.
- AMD Zen 3 to match or exceed Intel's IPC, without degrading energy efficiency.
- AMD Zen 3 to be scalable at higher clock speeds.

It will only take one of these to be slightly off for you to be wrong in your estimate.

CaqdeHonestly what hope does Intel have against the Zen 2 chips with the 9700K/9900K? A very hot 14nm chip based on a slightly optimized 4 year old architecture vs a high performance 7nm chip on a new Architecture. If AMD can pull 4.6-4.7Ghz singlecore boost and a 10 - 15% IPC improvement on average Intel will be in trouble for the rest of 2019.

Skylake might be older than Zen, but still outperforms it by a good margin.
Even if "TSMC 7nm" is extremely good and have no major problems, we'll be looking at late Q4 2019 or Q1 2020 before it can match the clock speeds of the current mature node, and even longer to exceed it.
AMD will also need a bit more than 4.7 GHz boost and 10-15% IPC improvement to exceed Skylake.

efikkanYes, unfortunately AMD seem to plan to do at least five iterations of Zen. They better start planning the successor now, because designing a new major architecture takes >3 years. I'm worried after seeing what they've done with GCN, and knowing their primary focus is custom designs like gaming consoles, I'm worried that the improvements we saw in Zen will be a "one time thing", just like GCN.

It's also important to remember that even though Zen have done some good design choices, most of the improvements over Bulldozer was either correcting "design faults" of Bulldozer or "low hanging fruit". To close the gap further, they need to make a much more advanced front-end/prefetcher, which is much more costly both in terms of development time and die size than just adding another ALU. AMD will also have to step up their game in AVX support, which may not me difficult, but will require a lot of die space, and eventual support for AVX-512 might require cache changes.

Hey, that's how the game is played. It's a huge investment, one simply cannot afford to come up with a new architecture every couple of years. However, all is not lost. With Core, Intel was able to extract much more performance from a second (or third, depending on how you're counting) iteration, by going to a lower nose and increasing clock speeds dramatically. Maybe AMD can play a trump card like that with Zen, too.

bugHey, that's how the game is played. It's a huge investment, one simply cannot afford to come up with a new architecture every couple of years. However, all is not lost. With Core, Intel was able to extract much more performance from a second (or third, depending on how you're counting) iteration, by going to a lower nose and increasing clock speeds dramatically. Maybe AMD can play a trump card like that with Zen, too.

Sure, but the "Core" series from Intel have featured some pretty different architectures.
Both Nehalem and Sandy Bridge was major redesigns over their predecessors, Haswell and Skylake a little less so (except for X/SP models).

If we compare this to my GCN analog;
Pitcairn -> Hawaii -> Fiji -> Polaris -> Vega
Kepler -> Maxwell -> Pascal -> Volta
There is a huge difference in how these companies evolve their products. GCN did initially close most of the gap with Nvidia, but has fallen further and further behind in every iteration since then. AMD must be careful not to do the same mistake in the CPU market. Tweaking Zen for a couple of years is fine, but then it's time to move on, or they will fall further behind again.

It's kind of strange how many people keep claiming that Intel's innovation have stagnated, while praising AMD who plan to recycle Zen for nearly a decade.:confused:

efikkanIt's kind of strange how many people keep claiming that Intel's innovation have stagnated, while praising AMD who plan to recycle Zen for nearly a decade.:confused:

We praise AMD for being competitive once again (specially considering the company's balance sheet was covered in negative numbers for years), while we criticize Intel for getting lazy (also considering their R&D division probably has more money and funds than AMD is worth). However, the idea of recycling Zen for a decade is something that has not yet happened (I mean, come on, Zen has been, what, barely a year on the market?), and as such, while it is something that we'll watch carefully for, we are not yet going to blast them with criticism.

efikkanThe first half of 2020 is >1.5 years away, a lot can happen by then.
Your scenario of AMD crushing Intel in 2020 relies on a lot of highly speculative assumptions:
- Intel doing nothing for the next >1.5 years, and Intel 10 nm still be as "bad" as today.
- "TSMC 7nm" to be fantastic and for AMD to have all the production capacity they want.
- AMD Zen 3 to match or exceed Intel's IPC, without degrading energy efficiency.
- AMD Zen 3 to be scalable at higher clock speeds.

It will only take one of these to be slightly off for you to be wrong in your estimate.

Skylake might be older than Zen, but still outperforms it by a good margin.
Even if "TSMC 7nm" is extremely good and have no major problems, we'll be looking at late Q4 2019 or Q1 2020 before it can match the clock speeds of the current mature node, and even longer to exceed it.
AMD will also need a bit more than 4.7 GHz boost and 10-15% IPC improvement to exceed Skylake.

I mean 2019 in that post for the timeframe in one spot and was meaning to mainly focus on that issue. Which as the sources note and Intel themselves note they have a production issue. As far as TSMC goes it should be noted that AMD also has the ability to use 7nm chips from Samsung and Global Foundries and we do know they are also leveraging Global Foundries for 7nm and that at least for now Yields of TSMC 7nm today are much better than Intels 10nm considering Intel themselves have said it won't be until 2019 when they are able to start mass production with no idea when in 2019 they are going to be starting it (We would hope during the 1st half for their sake).

Skylake does not outperform Zen by a good margin at least not as much as most think it does. Most assumptions that Intel is beating AMD by a good margin are due to Intel chips having a ~25% clockspeed headroom over AMD's first gen Ryzen chips (4Ghz vs 5Ghz), but IPC wise they are very close to each-other. If AMD hits 4.7Ghz clockspeed Intel would only have a 6-10% clockspeed advantage when overclocked and the IPC would be behind AMD so that wouldn't be much of an advantage at all. IPC is only off by 3-4% on average with major losses of up to 15% and Wins of up to 8%. So yes if AMD gets 4.7Ghz and 10-15% IPC Average they will be Winning most of the time against Coffeelake.

IPC -> www.techspot.com/article/1616-4ghz-ryzen-2nd-gen-vs-core-8th-gen/

efikkanSure, but the "Core" series from Intel have featured some pretty different architectures.
Both Nehalem and Sandy Bridge was major redesigns over their predecessors, Haswell and Skylake a little less so (except for X/SP models).

If we compare this to my GCN analog;
Pitcairn -> Hawaii -> Fiji -> Polaris -> Vega
Kepler -> Maxwell -> Pascal -> Volta
There is a huge difference in how these companies evolve their products. GCN did initially close most of the gap with Nvidia, but has fallen further and further behind in every iteration since then. AMD must be careful not to do the same mistake in the CPU market. Tweaking Zen for a couple of years is fine, but then it's time to move on, or they will fall further behind again.

It's kind of strange how many people keep claiming that Intel's innovation have stagnated, while praising AMD who plan to recycle Zen for nearly a decade.:confused:

Fair enough. I am just trying not to get my hopes up. Great way to avoid disappointment further down the road ;)

As for Intel's stagnation, I've said it countless times: it's true, Intel did stagnate in IPC (and raw clockspeed, thus overall performance, in the past few years). That much is true and it provides ammo for some people. However it also just one piece of the puzzle as Intel have been making good progress in pretty much every other aspect.

Caqdeand that at least for now Yields of TSMC 7nm today are much better than Intels 10nm considering Intel themselves have said it won't be until 2019 when they are able to start mass production with no idea when in 2019 they are going to be starting it (We would hope during the 1st half for their sake).

Are they? All we know is that test production on "TSMC 7nm" looks good so far, and they are starting mass production now, which may enter the market around Q1 2019, which is when we'll know how good yields and volume they get of production samples. And remember, even if "TSMC 7nm" in Q1 2019 have better yields than any of the previous nodes, we will still be looking at relatively limited volume production which will gradually improve throughout 2019. It will not be perfect from day one, and it will take some time for the new node to allow higher clocks than the current mature node. But when "TSMC 7nm+" arrives, that's when we'll see the full potential of the node.

And remember, Intel's 10nm is in (low) volume production right now. It can improve a lot by the time "TSMC 7nm" arrives.

CaqdeSkylake does not outperform Zen by a good margin at least not as much as most think it does. Most assumptions that Intel is beating AMD by a good margin are due to Intel chips having a ~25% clockspeed headroom over AMD's first gen Ryzen chips (4Ghz vs 5Ghz), but IPC wise they are very close to each-other.

No, Intel does not have a 25% clockspeed headroom.
Take for instance i9 7900X vs Threadripper 1950X, very comparable average performance, yet Intel manages it with just 10 cores vs. 16, and at similar clocks 3.3 GHz-4.0 GHz vs. 3.4 GHz-3.7 GHz. And you'll see this throughout the lineup; Intel performs better per core in average.

And there isn't much clockspeed headroom for either of them going forward. As we all know, the overclocking headroom today is practically nothing compared to 5+ years ago, and the new nodes will only give marginally better headroom, and only towards the end of the node's lifecycle. The scaling potential in the next decade will be in IPC. As mentioned, AMD did mostly pick the "low-hanging fruit" in Zen1 (which is the right place to start), but now they have to tackle the harder challenges, and every improvement pushing their IPC closer to Intel will only get more costly. Intel's front-end/prefetcher is comparatively several generations ahead, and even if we assume Ice Lake will be a lousy 5% better than Skylake, it will still be much harder for AMD to close the remaining gap in Zen2, and doing so would become their greatest achievement ever.

bugFair enough. I am just trying not to get my hopes up. Great way to avoid disappointment further down the road ;)

Yes, that's my point; we need to be realistic here. The AMD hype is extreme, people claim they will outperform Intel in 2019 and 2020. Even getting on par with Intel's three year old Skylake will require a major redesign and lots of die space, plus there is a new major architecture called Ice Lake looming.

efikkanNo, Intel does not have a 25% clockspeed headroom.
Take for instance i9 7900X vs Threadripper 1950X, very comparable average performance, yet Intel manages it with just 10 cores vs. 16, and at similar clocks 3.3 GHz-4.0 GHz vs. 3.4 GHz-3.7 GHz. And you'll see this throughout the lineup; Intel performs better per core in average.

And there isn't much clockspeed headroom for either of them going forward. As we all know, the overclocking headroom today is practically nothing compared to 5+ years ago, and the new nodes will only give marginally better headroom, and only towards the end of the node's lifecycle. The scaling potential in the next decade will be in IPC. As mentioned, AMD did mostly pick the "low-hanging fruit" in Zen1 (which is the right place to start), but now they have to tackle the harder challenges, and every improvement pushing their IPC closer to Intel will only get more costly. Intel's front-end/prefetcher is comparatively several generations ahead, and even if we assume Ice Lake will be a lousy 5% better than Skylake, it will still be much harder for AMD to close the remaining gap in Zen2, and doing so would become their greatest achievement ever.


Yes, that's my point; we need to be realistic here. The AMD hype is extreme, people claim they will outperform Intel in 2019 and 2020. Even getting on par with Intel's three year old Skylake will require a major redesign and lots of die space, plus there is a new major architecture called Ice Lake looming.

Your second point is not really that good. An understanding of their turbo core is needed to compare these chips. The Second issue is an average based on what? Low Threaded, Heavily Threaded, or a Mixed. And if it is mixed what is the weighting of the style of appilcations. These are HEDT chips you are comparing typically used for Multithreaded applications I'd hope. For instance in Cinibench the 1950X is 40% faster than the 7900X and in 7zip it is 59% faster while decompressing a file (which is basically the core count % difference). I wouldn't call that comparable. BTW get your clockspeeds right Threadripper is 3.4Ghz to 4.0Ghz (3.5 Turbo all core and 2 core 3.5), While the 7900X is 3.3ghz to 4.5Ghz (With a granular descending turbo). But anyways the Threadripper design brings about it's own issues when it comes to comparing Skylake and Zen. The interposer used to bridge the Zen chips increases latencies and is an issue that can be dealt with outside of the Zen chip design by using an Active Interposer design. My previous post has a link to a site that did an IPC comparison as I posted in the previous post the IPC difference is between Ryzen 2nd Gen and Coffeelake is between -15% and +8% depending on the task. This gives Intel a 3-4% IPC advantage this isn't that much. This is different for Threadripper and Skylake-X as their IPC is not the same as Ryzen and Coffeelake IPC. Their designs work differently. Skylake-X especially.

As far as the clockspeed headroom. Yes Intel had a 25% headroom over AMD on first Gen Ryzen. Coffeelake is able to hit typically 5Ghz on an Overclock. Ryzen was able to hit 4.0Ghz. 2nd Gen lowers this to around 20% with it's 4.3-4.35Ghz Clockspeed maximums. I wasn't comparing the clockspeed headroom of Intels Monolithic HCC and LCC chips to AMD's Zen. Ryzen, Threadripper, and EPYC have the same headroom for good reason they are all made using the same chip (barring cooling, PSU, and your motherboards VRM's). Intels Desktop (6 Core), LCC (10 Core), HCC (18 Core), and XCC (28 Core) have different headrooms for clockspeed maximums and it makes sense they are different chips. And also for good reason the larger the Intel Die the lower the Headroom will be an Intel 6 core desktop chip will have more headroom than the 28 core XCC based chip (Hell it needed a Water chiller to hit 5Ghz).

AMD has basically bridged that gap already with IPC people just don't see it because it isn't where they want it to be bridged (mainly in Games). There should be more than just Techspot doing an IPC benchmark if I'm right there is. I know that PCper did one when Zen came out, but that one is not really valid due to the early BIOS performance bugs and fixes between now and then changing the performance in a lot of the areas AMD was weak in the benchmarks they ran.

EDIT: As for what an active Interposer is and what changes it might bring to the table these research documents by AMD and the University of Toronto will shed some light on that and also why AMD moved away from monolithic processors for the Server and HEDT market's. This is also why I will not compare Ryzen's clockspeed headroom to Intel's HEDT chips and instead only use Intel's Desktops for comparison sake. I'm assuming Intel will move to this kind of design within the next 4-5 years.

www.eecg.toronto.edu/~enright/micro14-interposer.pdf
www.eecg.toronto.edu/~enright/Kannan_MICRO48.pdf