AMD's Future Chips & SoC's: News, Info & Rumours.

[aldaia]

No. The K7 and K8 were derived from Alpha technology. AMD acquired virtually the whole Alpha team for the K7.

K7 and K8 where certainly influenced by Alpha technology. In particular K7 FSB was based on DEC's Alpha bus and Hypertransport was also based on Alpha technology. However, despite what wikipedia may say, AMD it's not the company the benefitted the most from Alpha technology and its workforce. It's actually Intel who acquired virtually the whole Alpha team.

Around 1998 DEC/COMPAQ laid-off a good chunk of the Alpha team, but not the majority, an even bigger chunk remained at DEC and continued with alpha development for a few years. Most, but not all, of those laid-off, were hired by AMD, that is far from "virtually the whole Alpha team".

Also, in 1997 intel and DEC sued eachother for patent infringment. An agreement was signed in 1998 to cross-licence patents for 10 years. Intel obtained the Hudson fab and the rights to make StrongArm, while Digital got $700 million. Simultaneously, Intel hired a number of the Digital engineers and gradually began to incorporate the company's technology into its own product lines.This granted Intel access to Alpha technology at about the same time as AMD.

When HP bought compaq in 2002, practically all of the remaining of the Alpha team where hired by intel. There are cases where full research LABS including not only the people, but also the installations where transfered from COMPAQ/HP to Intel. So it's Intel who actually ended-up acquiring most of the Alpha team and processor technology from DEC. For instance research performed by Digital's Alpha team in the '90s on multithreading strongly influenced HyperThreading.

Incidentally, the Intel Core 2 comeback after the Pentium 4 fiasco hapened a few years after Intel virtually acquired the whole Alpha team.

 

[juanrga]

You do realize that CPU design is not really 100% original IP from anyone? Intel, AMD, Qualcomm, Samsung and a long list of designers and manufacturers buy IP, improve on top of it and put their own products? AMD buying a design for a CPU, then modifying it and producing its own is fair game.

But no one said anything about fair or unfair! The point was that the only times that AMD has been on pair or even above Intel was when AMD got the needed IP and engineerings from outside.

what you basically say is "FX 83xx was a good chip but AMD engineers did a wrong forecast about their chip hitting 5-6Ghz on air. So they fall behind Intel".

I remember those times. Some magazines assume P3-P4 (not sure) would hit 10GHz and will be the market leader. But it did not. So those products was not GOOD ones just like FX83xx series.

This is the rule of the market. You forecast the technological advances. You design a product that will be using that technology which you assume would be ready by the time you relise your product. and if your assumption is wrong you loose. Inte lost.

Than what intel did to save its ground? make illegal agreements with main OEM's about not to use AMD chips. Which caused AMD to suffer financially. Yes intel had to pay alot of money to AMD but by the time they pay it, AMD was already deep in shit.

There is a huge difference between designing a chip to hit 10GHz and not be able to get them hit 4GHz, and
designing a chip to hit 5GHz and be able to get them hit 4GHz. Bulldozer was behind targets by 20% or so. The P4 was behind targets by 150% or so.

The impact of Intel illegal practices on AMD finances is overstated. Even former AMD guys claim that the problem wasn't Intel: "it wouldn’t have mattered” he said “[Sanders] took his shot, and the game’s been played.” The problem was the engineering and financial mistakes made by AMD during Sanders' epoch:

"The trouble in the entire economic model was that AMD did not have enough capital to be able to fund fabs the way they were funding fabs," Raza said. "The point at which I had my final conflict was that [Sanders] started the process of building a new fab with borrowed money prematurely. We didn't need a fab for at least another year. If we had done it a year later, we would have accumulated enough profits to afford the fab in Germany. He laid the foundation for a fundamentally inefficient capital structure that AMD never recovered from. I told him: don't do it. I put the [purchase orders] on hold. He didn't tell me and accelerated the entire process."

 

[-Fran-]

But no one said anything about fair or unfair! The point was that the only times that AMD has been on pair or even above Intel was when AMD got the needed IP and engineerings from outside.

Sigh... Yeah, because the original people that started the Company is still in it, all the way from the 50s... Because when they've taken/bought an IP, they've done absolutely nothing to improve on top of it and put out something better. Everything AMD has ever done to be on par and put out good products has been buying something from the outside and re-badge it. No added value what-so-ever in their products, none.

I'll just stop here; you won't realize your own shortfalls in your argument nor realize how much cynicism is in them.

 

[daerohn]

No. The K7 and K8 were derived from Alpha technology. AMD acquired virtually the whole Alpha team for the K7.

K7 and K8 where certainly influenced by Alpha technology. In particular K7 FSB was based on DEC's Alpha bus and Hypertransport was also based on Alpha technology. However, despite what wikipedia may say, AMD it's not the company the benefitted the most from Alpha technology and its workforce. It's actually Intel who acquired virtually the whole Alpha team.

Around 1998 DEC/COMPAQ laid-off a good chunk of the Alpha team, but not the majority, an even bigger chunk remained at DEC and continued with alpha development for a few years. Most, but not all, of those laid-off, were hired by AMD, that is far from "virtually the whole Alpha team".

Also, in 1997 intel and DEC sued eachother for patent infringment. An agreement was signed in 1998 to cross-licence patents for 10 years. Intel obtained the Hudson fab and the rights to make StrongArm, while Digital got $700 million. Simultaneously, Intel hired a number of the Digital engineers and gradually began to incorporate the company's technology into its own product lines.This granted Intel access to Alpha technology at about the same time as AMD.

When HP bought compaq in 2002, practically all of the remaining of the Alpha team where hired by intel. There are cases where full research LABS including not only the people, but also the installations where transfered from COMPAQ/HP to Intel. So it's Intel who actually ended-up acquiring most of the Alpha team and processor technology from DEC. For instance research performed by Digital's Alpha team in the '90s on multithreading strongly influenced HyperThreading.

Incidentally, the Intel Core 2 comeback after the Pentium 4 fiasco hapened a few years after Intel virtually acquired the whole Alpha team.

You do realize that CPU design is not really 100% original IP from anyone? Intel, AMD, Qualcomm, Samsung and a long list of designers and manufacturers buy IP, improve on top of it and put their own products? AMD buying a design for a CPU, then modifying it and producing its own is fair game.

But no one said anything about fair or unfair! The point was that the only times that AMD has been on pair or even above Intel was when AMD got the needed IP and engineerings from outside.

what you basically say is "FX 83xx was a good chip but AMD engineers did a wrong forecast about their chip hitting 5-6Ghz on air. So they fall behind Intel".

I remember those times. Some magazines assume P3-P4 (not sure) would hit 10GHz and will be the market leader. But it did not. So those products was not GOOD ones just like FX83xx series.

This is the rule of the market. You forecast the technological advances. You design a product that will be using that technology which you assume would be ready by the time you relise your product. and if your assumption is wrong you loose. Inte lost.

Than what intel did to save its ground? make illegal agreements with main OEM's about not to use AMD chips. Which caused AMD to suffer financially. Yes intel had to pay alot of money to AMD but by the time they pay it, AMD was already deep in shit.

There is a huge difference between designing a chip to hit 10GHz and not be able to get them hit 4GHz, and
designing a chip to hit 5GHz and be able to get them hit 4GHz. Bulldozer was behind targets by 20% or so. The P4 was behind targets by 150% or so.

The impact of Intel illegal practices on AMD finances is overstated. Even former AMD guys claim that the problem wasn't Intel: "it wouldn’t have mattered” he said “[Sanders] took his shot, and the game’s been played.” The problem was the engineering and financial mistakes made by AMD during Sanders' epoch:

"The trouble in the entire economic model was that AMD did not have enough capital to be able to fund fabs the way they were funding fabs," Raza said. "The point at which I had my final conflict was that [Sanders] started the process of building a new fab with borrowed money prematurely. We didn't need a fab for at least another year. If we had done it a year later, we would have accumulated enough profits to afford the fab in Germany. He laid the foundation for a fundamentally inefficient capital structure that AMD never recovered from. I told him: don't do it. I put the [purchase orders] on hold. He didn't tell me and accelerated the entire process."

what I basically say is yes FX83xx was a bad idea, an engineering forecast and desing problem and Intel P4 is also the same.

For the illegal aggreements intel done with OEM's,this aggreement has nothing to defend. Yes AMD made terrible mistakes which in the end cost them a lot of sales and financial crisis yet Intel has proven an unethical company in my eyes. So this effects my purchasing decisins alot. Well in fact this is why I purchased an FX8350 instead of an I5 back than.

Right now the competition is back on track and this will also force intel to hasten their development efforts. and again when you compare two companies, AMD did a great job with Ryzen. and in the coming years I hope to see more from AMD.

 

[juanrga]

Sigh... Yeah, because the original people that started the Company is still in it, all the way from the 50s...

Ironic response, when the argument was that AMD could pull a "rabbit out of their hats" tomorrow, because AMD did it in the past with the K7.

This is not about people. Technological designs require resources and AMD lacks those resources. There is not a magic team of geniuses that will break the laws of physics and economy and will do a superchip that will kill everything else.

I have been reading "that the next product will be the game changer" since Bulldozer. Do I need to link to older posts with people stating that Steamroller was going to outperform Haswell? Do I need to link to older posts about how Zen was the new K8 and killing Intel?

Sandy Bridge i7 was about 50% ahead of Bulldozer FX8150 so 4-core Intel could keep with 8-core AMD in multi-thread.

CofeeLake i7 is about 40% ahead of Zen 1800X so 6-core Intel can keep with 8-core AMD in multi-thread.

Has AMD advanced? Sure. Will Pinnacle Ridge do better? No doubt. But one would also keep in mind that AMD is competing with plan-B from Intel. The main plan was Icelake, which was delayed due to problems with 10nm. Kabylake and CoffeeLake are plan-B from Intel.

what I basically say is yes FX83xx was a bad idea, an engineering forecast and desing problem and Intel P4 is also the same.

For the illegal aggreements intel done with OEM's,this aggreement has nothing to defend. Yes AMD made terrible mistakes which in the end cost them a lot of sales and financial crisis yet Intel has proven an unethical company in my eyes. So this effects my purchasing decisins alot. Well in fact this is why I purchased an FX8350 instead of an I5 back than.

Right now the competition is back on track and this will also force intel to hasten their development efforts. and again when you compare two companies, AMD did a great job with Ryzen. and in the coming years I hope to see more from AMD.

Both Bulldozer and P4 were bad ideas, but P4 was much worse because it missed the performance target by ~150%, whereas Bulldozer only by ~20%.

No one is defending Intel illegal practices. What Intel did is not excusable. The point was other. The point was that former AMD guys claim that the problem wasn't Intel. Former AMD guys claim that the financial problems of AMD are basically a consequence of mistakes made by former AMD managers which "laid the foundation for a fundamentally inefficient capital structure that AMD never recovered from".

So, if AMD former guys admit that the real problem were themselves...

 

[daerohn]

What you basically say is K6-K7 was not too good, Intel's product was worse. So with same logic we can say i7 was not so good, FX series were worse.

Nobody expects AMD to pull a rabbit out of somewhere. However everybody hopes AMD would close the gap and develop a product that will trade blows with Intel counterparts. With ryzen they get closer to this. Ryzen 2 hope they will close the gap further.

AMD has been well behind Intel in the last decade and everybody knows it would be impossible to close the gap with a single step. How ever with Ryzen the gap has been closed a lot more than I expected. If only they were able to industrialize it 1 year earlier it would have been better. But with Ryzen 2 can increase the performance %10-%15 which I think is possible, the gap ill be even more narrow. And with Zen2-3 I hope AMD can finally catch Intel. A %10 increase in performance means around 4600 points in geekbench while %15 means around 4900. And when we check the performance of Ryzen Apu's this scores seems highly achievable.

 

[-Fran-]

https://techreport.com/news/33336/rumor-benchmarks-of-upcoming-ryzen-7-cpu-surface

Improvements all around. Looks like this mid-gen upgrade is going to be interesting, even without "significant" (debatable, I guess?) modifications.

As usual, salt is highly recommended.

Cheers!

 

[aldaia]

 

[-Fran-]

Oooooh. So it means the 2700X will be top dog until further notice. That is interesting. Will they be pulling an Intel with the 2600K and 2700K? Well, they've already done it with the FX8370 and FX9K, so...

Cheers!

 

[jdwii]

Those last slides are fake

 

[-Fran-]

Awww. Welp, it was fun to think about it at least.

Cheers!

 

[juanrga]

What you basically say is K6-K7 was not too good, Intel's product was worse.

No, That isn't what I said. K7--K8 were good chips. What I said is that most of their commercial success was due to P4 being a fiasco.

 

[juanrga]

The slides aren't fake...

 

[jdwii]

The slides aren't fake...

Maybe not but some sites seem to think so they aren't confirmed yet

https://www.pcgamesn.com/amd-ryzen-2-dodgy-slides

lusterfrak that is the XFR 2, XFR 2 Enhanced, and Precision Boost Overdrive stuff. If AMD really want to complicate the positioning of their new CPUs then that would be a good way to go about it...

However if true it seems like it could be a nice improvement

https://www.dsogaming.com/news/first-gaming-benchmarks-specifications-and-prices-for-amds-2nd-generation-ryzen-cpus/

10% boost in fallout 4 for example with a 2700x vs 1800x this could also in part be due to XFR 2.0 allowing more then 2 cores to turbo at once.

 

[juanrga]

The slides aren't fake...

Maybe not but some sites seem to think so they aren't confirmed yet

https://www.pcgamesn.com/amd-ryzen-2-dodgy-slides

lusterfrak that is the XFR 2, XFR 2 Enhanced, and Precision Boost Overdrive stuff. If AMD really want to complicate the positioning of their new CPUs then that would be a good way to go about it...

However if true it seems like it could be a nice improvement

https://www.dsogaming.com/news/first-gaming-benchmarks-specifications-and-prices-for-amds-2nd-generation-ryzen-cpus/

10% boost in fallout 4 for example with a 2700x vs 1800x this could also in part be due to XFR 2.0 allowing more then 2 cores to turbo at once.

The slides are edited, because the source doesn't want to have legal problems by breaking NDA.

 

[8350rocks]

IWhen 7nn hit's AMD will be well ahead in IPC, really looking forward to that.

There is absolutely nothing supporting such claim.

If AMD keeps gaining IPC, and Intel holds tradition for their last 3 generations...then it is an inevitability, unless you can provide a hard source that shows otherwise.

There is no hard source showing that AMD will keep gaining IPC until surpassing Intel. So the claim that AMD will be "ahead" by 7nm isn't supported by anything.

Also "last 3 generations" aren't Intel's tradition, but an exception. The original plan was Icelake to replace Skylake. Then 10nm got delayed and Intel had to release Kaby and CoffeeLake as stopgaps.

3 generations is enough to form a pattern, that becomes a trend, which becomes the norm.

The norm for Intel right now is "look we can boost clocks on the same uarch"...until something comes along to show otherwise, 5 years of history show nothing but milking consumers for the same processors, *while asking them to buy new boards on top of that to keep current*.

 

[goldstone77]

Eh, Sandy Bridge doesn't overclock that well. Maybe a very lucky chip? An 8700K that does 5.6GHz at lower than 1.5V and passes a Cinebench test exists, so there's that.

Eh, I doubt it will go any higher than 4.4GHz. Maayybe 4.5GHz. We'll see.

Tests with Sandy Bridge hitting 5GHz exists! Anandtech has Sandybridge hit 4.8GHz. I have personally overclocked the 4.7GHz stable on a Sandy bridge with Evo 212, but I don't like my processor hitting 80 degrees. I've built multitudes of 2500K and 2600K systems that are still in operation today with evo 212 coolers and 4.6GHz overclock. The trick to overclocking is knowing how to do it. Being able to maintain a constant voltage is the hardest obstacle. Don't expect to get good results just raising clocks without modifying anything else or expect 4.4GHZ. Just in case you are wondering I use Noctua NT-H1. Under load it's operates within 1 Degree of the the most expensive pastes, and you don't have to worry about it hardening. It's smooth and playable after years of continuous usage.

https://www.anandtech.com/show/4083/the-sandy-bridge-review-intel-core-i7-2600k-i5-2500k-core-i3-2100-tested/3

 

[Gon Freecss]

Eh, Sandy Bridge doesn't overclock that well. Maybe a very lucky chip? An 8700K that does 5.6GHz at lower than 1.5V and passes a Cinebench test exists, so there's that.

Eh, I doubt it will go any higher than 4.4GHz. Maayybe 4.5GHz. We'll see.

Well, I do have a 2700K, so you could argue it is closer to a "golden" sample than regular 2600K, but still a Sandy and widely sold at the time.

I'm also not expecting 5Ghz on water and lidded from Zen "v1.5" (or "v1 rev B", whatever you wanna call it), but we just have some specs in a leak that mention the CPU itself could clock up to 4.3-4.4 Ghz on its own (projecting to the 2800X given the 2700X info). XFR, much like Turbo, is meant to really use all the available headroom in the CPUs, which is something I kind of like. That saves me the trouble of OC and tweaking to get 90%+ of the available CPU performance. All Intel CPUs have a lot of headroom, but they go outside of TDP and you have to account for that. Ironic that I'm complaining about it, but if you think a bit, it's actually good for mainstream.

Also, yeah, Intel might be able to clock higher, but they still have 2 cores less with similar power consumption. Differences in performance not-withstanding, it is impressive for the process AMD is using and it is an advantage, like it or not.

Cheers!

Sandy Bridge doesn't overclock as well as delidded Kaby Lake nor Coffee Lake.

Eh, the 8700K consumes less power than the 1700X while performing better, so it doesn't matter anyway.

Eh, Sandy Bridge doesn't overclock that well. Maybe a very lucky chip? An 8700K that does 5.6GHz at lower than 1.5V and passes a Cinebench test exists, so there's that.

Eh, I doubt it will go any higher than 4.4GHz. Maayybe 4.5GHz. We'll see.

Tests with Sandy Bridge hitting 5GHz exists! Anandtech has Sandybridge hit 4.8GHz. I have personally overclocked the 4.7GHz stable on a Sandy bridge with Evo 212, but I don't like my processor hitting 80 degrees. I've built multitudes of 2500K and 2600K systems that are still in operation today with evo 212 coolers and 4.6GHz overclock. The trick to overclocking is knowing how to do it. Being able to maintain a constant voltage is the hardest obstacle. Don't expect to get good results just raising clocks without modifying anything else or expect 4.4GHZ. Just in case you are wondering I use Noctua NT-H1. Under load it's operates within 1 Degree of the the most expensive pastes, and you don't have to worry about it hardening. It's smooth and playable after years of continuous usage.

I]https://www.anandtech.com/show/4083/the-sandy-bridge-review-intel-core-i7-2600k-i5-2500k-core-i3-2100-tested/3

I know 5GHz is possible on Sandy Bridge. I just meant it's not as common as on Kaby Lake and Coffee Lake. Sandy Bridge hits over 4.5GHz easily.

I said I don't see Pinnacle Ridge getting past 4.4GHz. Maybe the best Threadripper bins might be able to reach 4.5GHz, but we have to wait and see.

 

[-Fran-]

Sandy Bridge doesn't overclock as well as delidded Kaby Lake nor Coffee Lake.

Eh, the 8700K consumes less power than the 1700X while performing better, so it doesn't matter anyway.

Considering the amount of process generations apart, I'd expect them to be way further separated, but they're not. Plus, you have to friggin' delid them to get those speeds! Sandy might be ancient and all, but it's been hands down the best CPU for a lot of generations, if not "ever", from Intel and AMD.

And it does matter, since you were touting process prowess and not the combination of both for a CPU. If you go by transistor count and density, AMD is not far away from Intel in most metrics; absolutely within striking distance. That is what I meant with "performance not withstanding".

Cheers!

 

[goldstone77]

Following Significant Technology, Capacity and Expansion Milestones, GLOBALFOUNDRIES’ Sanjay Jha to Pass Baton to Industry Veteran Tom Caulfield
Mar 09, 2018
https://www.globalfoundries.com/news-events/press-releases/following-significant-technology-capacity-and-expansion-milestones

 

[juanrga]

3 generations is enough to form a pattern, that becomes a trend, which becomes the norm.

The norm for Intel right now is "look we can boost clocks on the same uarch"...until something comes along to show otherwise, 5 years of history show nothing but milking consumers for the same processors, *while asking them to buy new boards on top of that to keep current*.

I already mentioned that the original plan was Skylake--> Icelake, and that Kabylake and CoffeeLake refreshes are only plan-B due to Icelake delays.

Moreover there was more changes from Skylake to Kabylake than there is from Summit Ridge to Pinnacle Ridge. And CoffeeLake did bring moar cores as well. So it is funny that Intel is being accused of milking users...

 

[-Fran-]

Funny how a "plan B" has been the biggest step forward for Intel since Sandy Bridge in the mainstream consumer market.

Cheers!

 

[Kulasko]

Moreover there was more changes from Skylake to Kabylake than there is from Summit Ridge to Pinnacle Ridge.

Like what? There was an update to the video engine, what else? Pinnacle Ridge brings the completely new turbo system along improved memory and cache latency across the board. There also is the 12nm shrink. And yes, 12nm is not 12nm and most of the stuff is reused from 14nm, but libaries changed from 9T to 7.5T, resulting in a physical shrink.

 

[juanrga]

Funny how a "plan B" has been the biggest step forward for Intel since Sandy Bridge in the mainstream consumer market.

Plan A was a much bigger step forward...

Moreover there was more changes from Skylake to Kabylake than there is from Summit Ridge to Pinnacle Ridge.

Like what? There was an update to the video engine, what else? Pinnacle Ridge brings the completely new turbo system along improved memory and cache latency across the board. There also is the 12nm shrink. And yes, 12nm is not 12nm and most of the stuff is reused from 14nm, but libaries changed from 9T to 7.5T, resulting in a physical shrink.

Differences between Kabylake and Skylake processors are a 14nm+ process with different pitch, Speed Shift v2, improved graphics cores, Optane, fixed-hardware decoders, OPI 3.0, and higher core clocks.

Pinnacle Ridge brings a 14nm+ process (relabeled to 12nm, but doesn't even listed at Glofo site), Precision Boost 2, faster IMC (which reduces cache/IF latencies), and higher core clocks.

 

[daerohn]

This latest Coffee Lake cpu's from intel was a quick reply to AMD Ryzen. in my opinion Intel never, ever planned to supply this many cores for a long time. They decided to take things slowly as they are lacking serious competition and enjoy the money income. This is why they did not do anything to improve the IPC for clock to clock. The main advantage of coffee lake is having more cores and more speed which seems to be the result of the process improvements.

And you claim intel is not miilking their customers. Intel would harvest every drop of oil out of a fly if they could.