AMD vs. Intel and Nvidia - The Next-Gen GPU War is ON!

[BoredSysAdmin]

I bought a whole buncha amd stock as soon as i heard Lisa Su was taking over AMD. I have known that high iq nerd from a long time ago...bwaaahaha

Now, i watch gleefully as she sinks the intel/nvidia low iq creatures and makes money for me! Long live Lisa! Long live AMD!

high IQ nerd - that's the key right there. or in other words, Very senior engineer. That's right, Intel has forgotten it and put the bean counter in charge.

 

[BoredSysAdmin]

As soon as I read about the NVDA interest in ARM I didn't see the synergy. I'm with @BoredSysAdmin, I think buying Ampere is a much better choice.

As for Intel, IMO the most important technology they're working on is oneAPI, which unifies the programming model for heterogeneous computing. If they build a decent GPU and they succeed with oneAPI they could successfully challenge NVDA and AMD. Cuda is not exactly programmer friendly unless you're a SIMD programming expert. Software enabling is an important factor in how Intel has succeeded beyond just being chosen for the IBM PC.

oneAPI: A New Era of Heterogeneous Computing

Remove proprietary code barriers with a single, open, standards-based programming model for heterogeneous computing—CPUs, GPUs, FPGAs, and other accelerators.

software.intel.com

I recently read an interview with Lisa Pierce from Intel. (another High IQ Nerd

Intel Xe Graphics: An Interview with VP Lisa Pearce

www.anandtech.com

Very much refreshing that Intel now seems quite serious about software. This is very interesting since Intel drivers weren't historically great. Nor AMD's for that matter.

 

[Irvrobinson]

I recently read an interview with Lisa Pierce from Intel. (another High IQ Nerd

Intel Xe Graphics: An Interview with VP Lisa Pearce

www.anandtech.com

Very much refreshing that Intel now seems quite serious about software. This is very interesting since Intel drivers weren't historically great. Nor AMD's for that matter.

Intel has always seemed serious about software to me. While their CPUs have often been less than SOTA, their software ecosystem work has IMO been an important factor in defending Intel's market share. Intel has always had many thousands of software developers, invested in things like compilers (I would guess Intel is the largest private investor in compilers in world), Linux (there's a reason why the Linux Foundation is headquartered in Beaverton, OR), networking and storage software (DPDK and SPDK were both originally developed at Intel), and lately persistent memory software (PMDK). Drivers... well, doesn't everyone hate drivers? Often huge, buggy, but the only way for IC guys to externalize their special features. And for every OS (including VMMs) you need a different version of a given driver for each, especially Windows. I always thought driver development was not really just software development, it was punishment for sins committed in a previous life.

Intel's biggest software challenge IMO is that they typically don't sell software; software only plays a supporting role. There are some exceptions I'm aware of, like VTune, and some of their HPC solution software, but the very large majority of Intel software I'm aware of is just given away to support chip sales. Until software is a first class citizen at chip companies, meaning it generates direct revenue, I think concerns about quality and ease of use will continue indefinitely. Intel is not the only transgressor; I can't think of a chip company that thinks software is as important as it should be.

 

[BoredSysAdmin]

Intel has always seemed serious about software to me. While their CPUs have often been less than SOTA, their software ecosystem work has IMO been an important factor in defending Intel's market share. Intel has always had many thousands of software developers, invested in things like compilers (I would guess Intel is the largest private investor in compilers in world), Linux (there's a reason why the Linux Foundation is headquartered in Beaverton, OR), networking and storage software (DPDK and SPDK were both originally developed at Intel), and lately persistent memory software (PMDK). Drivers... well, doesn't everyone hate drivers? Often huge, buggy, but the only way for IC guys to externalize their special features. And for every OS (including VMMs) you need a different version of a given driver for each, especially Windows. I always thought driver development was not really just software development, it was punishment for sins committed in a previous life.

Intel's biggest software challenge IMO is that they typically don't sell software; software only plays a supporting role. There are some exceptions I'm aware of, like VTune, and some of their HPC solution software, but the very large majority of Intel software I'm aware of is just given away to support chip sales. Until software is a first class citizen at chip companies, meaning it generates direct revenue, I think concerns about quality and ease of use will continue indefinitely. Intel is not the only transgressor; I can't think of a chip company that thinks software is as important as it should be.

One piece of titbit of info you're missing, since I haven't shared it yet, is that I used to work for Intel in a major R&D center and had few good friends who were [few still are] software developers working for intel. So I am well aware that Intel is employing thousands of software developers. You're correct that my grief with Intel is right on the money - is less than polished drivers. Anecdotical evidence that a Google search for "intel video drivers bugs" results in 11million pages is a clear indication of an ongoing and major problem. I also get that devs "hate" writing drivers, but it's the drivers is what consumer see/feel/experience. They (and me) don't care how well this specific chip has computed units, cache, or memory pipe. If I play a video and it's choppy on intel's video card - most people would blame Intel's video card, though, in reality, it could be down to a driver bug.

If you've read my above-linked interview, this is exactly what she is claiming - for the first time ever, Intel will treat Software first. This is a major change and hopefully one of the changes which could help turn this Itanic ship away from its current course with collusion with an iceberg,

 

[Irvrobinson]

high IQ nerd - that's the key right there. or in other words, Very senior engineer. That's right, Intel has forgotten it and put the bean counter in charge.

I think a lot of people forget that Intel had a non-engineer as CEO for many years, Paul Otellini, who was a marketing and sales guy with an MBA. Before him CEOs were engineers, but they weren't computer engineers, they were more into chip fabrication. That was true for the Intel CEO after Otellini too, who was another fabrication specialist. The current CEO uniquely a finance background, but I'm not sure that's the fundamental issue. In a company as diversified as Intel, with fabs and chip design, I'm not aware of anyone who could be expert in both. Intel's problems are below the CEO level.

 

[BoredSysAdmin]

I think a lot of people forget that Intel had a non-engineer as CEO for many years, Paul Otellini, who was a marketing and sales guys with an MBA. Before him CEOs were engineers, but they weren't computer engineers, they were more into chip fabrication. That was true for the Intel CEO after Otellini too, who was another fabrication specialist. The current CEO uniquely a finance background, but I'm not sure that's the fundamental issue. In a company as diversified as Intel, with fabs and chip design, I'm not aware of anyone who could be expert in both. Intel's problems are below the CEO level.

You just making my point for me. Paul Otellini, another bean counter, was put in charge of the group responsible for processors, a chipset for desktop, mobile, and servers between 1998 to 2002. Care to guess which one of the worst intel chips were created on his watch? Answer: https://en.wikipedia.org/wiki/Pentium_4

The Worst CPUs Ever Made

Most of the time at ExtremeTech, we celebrate the best of technology. Today, we're hailing the worst.

www.extremetech.com

2002 he became CEO, right around the time I was working for Intel as well.
What happened next is what saved the company from a very bad position. A tiny group of engineers in Intel's Israel R&D center saw what P4 (Prescott) is utter garbage. They went back to P-III chip and improved it with much better power efficiency and a few modern features. What resulted later in Banias, Dothan - both highly successful mobile chips (of Centrino push which btw hugely responsible for making Wifi as a standard feature on laptops), which later one used an iteration of that design in an even more successful foundation of current Intel Core chips.
In other words, if not for few very smart folks, and a bit of luck - Intel might not even be in the position it is today.

Now, As to how Brian Krzanich, in past process engineer, completely mocked up Intel's 10nm process Fab migration, what Is very interesting. Maybe it was WAY too ambitious. I don't know.

 

[stalag2005]

I happen to work for a company called Chemical Abstracts Service (website is www.cas.org) and I have seen a number of deposition patents from various manufacturers. Suffice to say as you change nodes, the chemistry and masking used has to change as well. It is apparent to me that the success of TSMC's 7 nm vs. Intel's 10 nm which are roughly the same sizes as indicated by a number of industry insider's observations, the problem lays in the deposition parameters and chemistry of the films used to build the IC chips. Intel has been struggling for the past 5 years from their initial announcement in 2016 for 10 nm until now to get the conditions right for large scale production of 10 nm. Now recently they have indicated their 7, 5, and 3 nm future nodes are similarly delayed for similar reasons to Intel's 10. Until Intel can work out the problems, they will continue to lag behind TSMC which has improved dramatically under the partnership with AMD. Not to say that TSMC has also benefitted from partnership with Apple. Between these three companies, TSMC as a fab is currently the industry leader. Of the companies that have major fab include TSMC, Intel, Global Foundaries, Micron, SK Hynix, Samsung, among others. Given this, Intel has to invest time and money to fix their fab issues or they will either play second fiddle, or forced to go fabless like AMD and Nvidia to survive. At this point Intel needs to fix their fundamental production issues to move forward.

 

[stalag2005]

I ought to add that Intel also desperately needs to redesign their processors due to security issues. At the moment, AMD has a significant lead on this score. Intel processors are insecure due the use of ring bus technology first used in the original Pentium processors. Until they redesign this architectural design, they will continue to need security patches. This has significantly hit performance of Intel processors in both Linux and Windows. Intel complains of the developers using bad programming practice, but ultimately a hardware flaw is the responsibility of one company, the designers of the product. If you are concerned about security, use AMD. Current tech is still susceptible from Intel and it is due to a flawed fundamental design. Even their latest are still susceptible due to this flaw. And to add insult to injury, AMD in their Zen 3 based processors are better than Intel chips core for core both in IPC (instructions per cycle) as well as multicore.

 

[Wayde Robson]

I bought a whole buncha amd stock as soon as i heard Lisa Su was taking over AMD.

Good timing, and in retrospect would have been a risky time to do so. I'm certain there were those who "believed" they were about to really make some great advances, but it certainly wouldn't have been obvious. Even considering the engineering talent Su had hired on.

How do you think Intel will do in the GPU market? They literally poached much of the talent that built today's AMD to assemble its Xe graphics division. At the very least I expect good things, but it's no guarantee it will ever be anything more than a 3rd their.

 

[Wayde Robson]

I ought to add that Intel also desperately needs to redesign their processors due to security issues. At the moment, AMD has a significant lead on this score. Intel processors are insecure due the use of ring bus technology first used in the original Pentium processors. Until they redesign this architectural design, they will continue to need security patches. This has significantly hit performance of Intel processors in both Linux and Windows. Intel complains of the developers using bad programming practice, but ultimately a hardware flaw is the responsibility of one company, the designers of the product. If you are concerned about security, use AMD. Current tech is still susceptible from Intel and it is due to a flawed fundamental design. Even their latest are still susceptible due to this flaw. And to add insult to injury, AMD in their Zen 3 based processors are better than Intel chips core for core both in IPC (instructions per cycle) as well as multicore.

Thanks for your insights. Security is a big deal and I hadn't even considered that when researching this story. I don't think I saw one mention of it in anyone's rendition of the AMD story, but that's not to say it wasn't there, it just never clicked with me.

But it's a huge deal today, especially in the US where Chinese technology is concerned. I saw a statistic that showed the three biggest "hacking" nations are China, Russia and India. I know from doing analytics for different companies, it's never a good sign to see your company websites receiving an inordinate amount of traffic originating out of either of those three nations, especially if your market is exclusive to North America. Them bots are always probing.

 

[Irvrobinson]

I ought to add that Intel also desperately needs to redesign their processors due to security issues. At the moment, AMD has a significant lead on this score. Intel processors are insecure due the use of ring bus technology first used in the original Pentium processors. Until they redesign this architectural design, they will continue to need security patches. This has significantly hit performance of Intel processors in both Linux and Windows. Intel complains of the developers using bad programming practice, but ultimately a hardware flaw is the responsibility of one company, the designers of the product. If you are concerned about security, use AMD. Current tech is still susceptible from Intel and it is due to a flawed fundamental design. Even their latest are still susceptible due to this flaw. And to add insult to injury, AMD in their Zen 3 based processors are better than Intel chips core for core both in IPC (instructions per cycle) as well as multicore.

How does the Ring-Bus, currently used only on client CPUs, reduce security?

 

[stalag2005]

How does the Ring-Bus, currently used only on client CPUs, reduce security?

Ring bus architecture is susceptible to exposure of the cache processor ram contents. Even if encrypted elsewhere in the system, in the cache the data is in the clear. Intel cannot fix this with out a fundamental change in how the processor works. Anyone who can read this data can find encryption keys, passwords, etc. that can be used against the computer as well as the people who use the computer.

 

[Irvrobinson]

Ring bus architecture is susceptible to exposure of the cache processor ram contents. Even if encrypted elsewhere in the system, in the cache the data is in the clear. Intel cannot fix this with out a fundamental change in how the processor works. Anyone who can read this data can find encryption keys, passwords, etc. that can be used against the computer as well as the people who use the computer.

Do you have references to this Ring-Bus problem? I did a couple of searches and could find none. It is difficult to believe an on-die interconnect is the root cause of security problems. More likely are the usual suspects, out of order execution, hardware multi-threading, transactional memory, lack of main memory encryption, and protection ring issues. (Protection rings shouldn't be confused with Ring-Bus. They are orthogonal.) Cache-resident data and instructions are unencrypted in every CPU architecture I'm aware of.

 

[Alex2507]

Do you have references to this Ring-Bus problem? I did a couple of searches and could find none. It is difficult to believe an on-die interconnect is the root cause of security problems. More likely are the usual suspects, out of order execution, hardware multi-threading, transactional memory, lack of main memory encryption, and protection ring issues. (Protection rings shouldn't be confused with Ring-Bus. They are orthogonal.) Cache-resident data and instructions are unencrypted in every CPU architecture I'm aware of.

As far as I'm concerned you may now refer to yourself as bilingual.

 

[Irvrobinson]

As far as I'm concerned you may now refer to yourself as bilingual.

You should hear me talking about a subject I’m really expert in, like bathroom cleaning.

 

[stalag2005]

https://hal.inria.fr/hal-01762803/document

http://www.cs.binghamton.edu/~secarch/asplos18.pdf

Request Rejected

Happy munching.

 

[BoredSysAdmin]

You should hear me talking about a subject I’m really expert in, like bathroom cleaning.

So how do you clean moldy caulk?

 

[Irvrobinson]

https://hal.inria.fr/hal-01762803/document

http://www.cs.binghamton.edu/~secarch/asplos18.pdf

Request Rejected

Happy munching.

These issues are not Ring-Bus related.

 

[Irvrobinson]

So how do you clean moldy caulk?

Try Clorox Tilex.

 

[stalag2005]

These issues are not Ring-Bus related.

Actually they are. AMD handles caches differently than the ring bus architecture. The way AMD is secure is due to their architecture. Intel noted the possibility using the ring bus architecture of the side channel attacks back in the late 90’s and ignored the finding. Intel doesn’t secure the cache hoping ring permissions would be enough. It was not. Now they have a mess due to lack of development of a much more secure processor needed to stop the side channel attacks of which there is a plethora of now.