Is Mario Rodrigues paid by AMD?

MARIO RODRIGUES' little series on AMD has whipped up a bit of interest. You'll find a bundle of feedback on that topic below. We kick off, though, with an appraisal of 'kicking' Pat Gelsinger's crystal balls.

Subject: Desktop chips 'to hit 15GHz by 2010'

Hello,

Just a general comment on these future predictions of processor speeds by Intel, AMD, IBM, etc.

It seems to me that a lot of people tend to get either quite excited ("That's Incredible!") or pessimistic ("That's impossible!") when chip companies talk about the future speeds they expect to reach with their CPUs, GPUs, network chips, etc. This seems rather strange since the "future" of microprocessors can be fairly well guesstimated by just using Moore's "Law".

If Moore's Law continues for the foreseeable future and most scientists believe it will, then microprocessor speeds will increase by 40-60% per year at a compounding rate. It is approximately 40% if the speed doubles every 24 months and approximately 60% if it doubles every 18 months. It is this compounding rate that gives microprocessors their incredible speed increases. It is this failure to recognize the compounding rate that confuses people when it comes to chips and also investing their money wisely.

So Pat "Kicking" Gelsinger says that we will be able to buy 15GHz Intel chips by the end of this decade. Doing a back-of-the-envelope calculation of what we should expect chip speeds to be 8 years from now shows that they should be anywhere from 16 to 40 times faster. Assuming that Intel is still using the same Pentium 4 architecture in 2010 (doubtful), then the P4 should by clocking anywhere from 44.8GHz to 112GHz. This now makes 15GHz seems very slow.

Therefore, if 15GHz is Intel's target for 2010, then there will need to be a much more efficient Intel chip running at that speed. Maybe they can stuff 4 CPUs inside one chip? That will make it just about right, otherwise Intel will definitely not be keeping up with Moore's Law.

I guess that I just whipped out my crystal balls, too. Let's try to not make a habit of this.

Cheers,
Michael Shandony

Subject: Is AMD now ready to impact the enterprise space?

Come on, are you paid by AMD to write this article?

Hitler was bad, there is no arguing about that, but should you now hate every German because of their history?

Of course Intel did make bad mistakes and had major drawbacks during their existence but did you once had a look at the current market since the beginning of this year?

AMD is making losses for more than a year now because they aren't able to deliever their CPU's to the market, engineering problems, delays, overclocking problems, bad stability and performance of their Athlon MP's and paper releases. After confusing the consumers with their PR rating they now again change the PR ratings rules on the fly (2400+ - 2800+) so that even people who knew how their system works before, now no longer can't follow it.

You were talking about the 1GHz race Intel lost and AMD was so happy about. What about the 2GHz or even 3GHz race this year. Of course MHz isn't everything but why did AMD not claim that when they were the first to deliever a CPU running at 1GHz?

What about AMD's claim to fool wall street moneymakers that AMD will make next quarter 10% more profit? What's 10% if you are currently making 50% more loss then profit...? Then you'll still make 45% more loss then profit. But I don't think that those stock buyers aren't aware of that or of the fact that AMD still won't be able to produce Athlon XP's in any quantity they need to do to achieve those 10% or whatever more profits...

You certainly have seen my point by now... Intel is not all good, of course, but AMD in any case isn't either...

Actually I found your article really funny, I mean, AMD is at the moment going through their greatest decline (hopefully the hammer will save them if it arrives at time) and Intel CPU's are getting better everytime a new one is released (0.13 micron, 512KB L2 Cache, more IPC, HT, DCDDR and so on) but you are writing about how great AMD is and how bad Intel did in the past few years :) I like that.

P.S.: Do you know how important stability is in the server or renderstation domain and why 90% of the companies choose Xeon CPUs instead of Athlon MPs? There are known stability problems with those chips and the companies are aware of it...

Cheers,
James Hinkley

Subject: Is AMD now ready to impact the enterprise space?

I would like to thank you for the wealth of information provided in your article - things I forgot like caminogate (the power of intel pr ;)) and things I never really knew anything about (FDIV) coz I only started following the scene in 1998.

I know of 2 'instability issues' at AMD however - but not nearly as significant as a recall. The first is the 'broken feature' in it's 750 chipset at introduction. Memory Bypass I believe it was called. It didn't prevent stable operation whatsoever, but once fixed (5th revision?) it provided a substantial performance boost (early adopters were the 'victims' of this). However, a similar thing happened with the introduction of the 2.8ghz PIV (the new stepping fixing something that should have been working at introduction in the first place).

The second instability is just a personal observation. AMD can't seem to decide properly what to introduce when anymore. Their roadmap has changed so many times that I don't even know anymore what Barton exactly is/will be, let alone when it will be introduced. etc. etc.

greetings,
Erik

Subject: Is AMD now ready to impact the enterprise space?

Hi there,

According to a recent discussion in comp.arch the AMD hypertransport bus does not use ECC. There is some secondary error detection through CRC, but as indicated in the thread the CRC data only arrives after 300 ns or so, at which point the data may already have been processed. Solving this in a graceful way is non-trivial, and if proper ECC is already too much effort, I doubt that this has been solved in the Opteron chip.

So is AMD/Opteron enterprise ready? I doubt it, and I strongly suspect Intel marketing will have a field day with this.

Regards,
Andries Thijssen

Subject: Is AMD now ready to impact the enterprise space?

Hi Paul, I read through this article and lets just say it leaves out many positive contributions Intel has and is making to further Corporate and Enterprise computing………but I guess that's what I'd expect from an article promoting one of our competitors…………….

But to call this an "Analysis" really does not hit the mark, certainly call it an "opinion"………………. Everyone has one of those

Graham Palmer, Intel UK PR

Subject: Is AMD now ready to impact the enterprise space?

"If a motherboard is not stable it doesn't work, end of story. PC hardware has to meet a minimum stability requirement; anything beyond that is going into the realm of diminishing returns."

I take issue with this comment. Obviously you have never had to diagnose PC problems before. There are many cases where a "flaky" (technical term) chip, RAM, motherboard or PSU can cause a PC to work fine for hours then suddenly freeze, or lock up when a certain application is launched, etc etc. Windows XP has copped a lot of flak from some people who are trying to run it on old, sub-standard (ie cheap) hardware and then blame stability problems on XP itself. Once I had a PC which would give a VXD error on startup even after reformats. Worked fine with a different keyboard (strange I know but true).

Anyway to gist of your article seems correct - I have not noticed any differences in stablity between AMD and Intel, although I take issue with many of VIAs products (which most AMD systems use) after been burned several times in the past with them - they may be better now, but I am not about to find out.

Damo

Subject: Is AMD now ready to impact the enterprise space?

Hi Mario,

You did an excellent job in part 1. of your evaluation on AMD vs. Intel and the bogus myths that prevail. I have stated the same exact info. time and again over the past several years. AMD has delivered rock solid since the Athlon debuted and they kicked Intel in the knee repeatedly along the way. Intel has all of it's Spinmeisters (a.k.a. talking heads), out in force at the moment to discredit AMD and the coming Hammer series, because they can not answer the challenge and they need to buy time. If Intel can falsely discredit AMD with FUD then they might be able to get some of their R & D hardware in the hands of reviewers and do another paper launch as they have done many times in the past. Naturally AMD is being grilled for this right now but if you look at the past record Intel has done this more often than AMD but this is the Bull Intel FUDsmen use in an attempt to discredit AMD. At least when AMD's products reach the marketplace they function properly. No Athlon AMD Mobos/chipsets, etc. have ever been recalled. I've lost count of the Intel recalls which even extend into the Xeon Mobo category.

Keep the "facts" coming. You're gonna take some flames from the Intel Spinmeisters but sooner or later most analysts and consumers who lack industry facts will take the blinders off and see Intel for what they are. The times, they are a changin' for the better!

Randy Hubbard
Race-Tech Engineering, Inc.

Subject: Is AMD now ready to impact the enterprise space?

Yeah, a bug AMD still has not fixed since its creation on Palomino.

The bug where speculative cache-loads mark the cache line as dirty and needing to be written if they are replaced. The problem is that the cache line was never altered and when written can destroy the correct data.

Linux and Windows make this bug go from a slight performance drain to a system crasher by marking shared video addresses as cacheable pages. That was justified by the performance gain - the same reason AMD justifies their bug.

The Windows hack to make these bugs not crash the box is to decrease the page size from 4MB to 4KB. The problem is that reduces performance (for me) by 8% due to the too-small TLB cache. Intel put generous ones in the P-III and later processors. AMD never learned that lesson, fully.

Palamino only gained a few more entries over Thunderbird.

Making Linux not crash is/was harder. I just use PCI video cards on Linux boxes (along with W2k servers where I don't want the performance hit). If you remember the first dual AMD motherboard (Tyan Thunder), it had on-board ATI Rage video, on the PCI bus and not via AGP. That board was targeted straight at the server market. AMD obviously wanted stability.

That said...the AMD MPX chipset makes for cheap and fast servers. Intel's E7500 solution costs $300+ more for the motherboard and then a little extra for the registered memory. Its only worth it for the added I/O bandwidth of multiple PCI-X busses, which I do need for multiple gigabit Ethernet cards. I'm very disappointed with E7500 memory bandwidth. Like the Nforce chipset, one might expect to see nearly 2x memory bandwidth by having two memory controllers, but the reality is that its only slightly more than one. I hope the E7501 and/or E7505 fix that problem.

One last bit on stability. More recent K7 motherboards have had much improved power supply components in order to work at ever higher clockrates. As a result, more modest speeds are more stable too. I've even been very successful upgrading old P-II systems WITHOUT upgrading the power supplies (ECS K7S5A + XP1800). That's lots cheaper than any P4.

Mark

Subject: Is AMD now ready to impact the enterprise space?

I mostly agree with what you say, but let's not forget some recents defaults of AMD CPU:
- fragile cores
- no good thermic protection
While these problems had mostly impacted the DIY market, it does not bode very well for AMD's reputation..

Another problem has been the Linux problems with AGP and AMD, some weird caching problems apparently, I never had a really a good explanation of who was the culprit: AMD or Linux's developers ? Not good for the trust factor, anyway..

I think than x86-64 CPU will do very well within the enterprise: 64 bit workstations and servers at an affordable price should be very well received after the initial "it's new, be careful" phase.

For the home market, the x86-64 CPUS will compete only on their price/performance ratio: except for hype reason, 64bitness should not matter in the home market, at least for two or three years *I think*.

I'm writing this on a PC with 512MB of RAM, and games are already starting to need more than 256MB of RAM, but games memory greed is limited by the installed parc..

Name and email address supplied

Subject: Is AMD now ready to impact the enterprise space?

Re your article on AMD 10/21: AMD's biggest problem standing in the way of major acceptance in the enterprise is, IMHO, THE CHIPSET. Until AMD -can/decide to/are able to- produce chipsets at a rate to compliment proccessor deliveries the quality/stability/suitability question will remain.

Via et al just ain't makin it.

Intel agrees.

Cheers
Name and email address supplied