My Second AMD

A few weeks ago I became a happy owner of a ThinkPad A485, the first ThinkPad (together with the E485 and related variants) to use an AMD CPU. History buffs will know that it’s far from the first ThinkPad with a non-Intel CPU; the very first ThinkPad-branded laptops (ThinkPad 700/720 in 1992) used IBM’s own 486 SLC and SLC2 processors, the 1995 ThinkPad 365 used Cyrix Cx486DX4 processors, and the 1996 ThinkPad 365E sported the IBM-built Cyrix 5×86. But for over 20 years, ThinkPad meant Intel, until the Summer of 2018.

What I just realized is that the A485 is only the second AMD-based machine I ever bought. The first one was a Lenovo IdeaPad Z75 in 2014, a laptop with an AMD A10 APU based on the infamous Bulldozer microarchitecture. Ironically I never had any real complaints about the Z75’s CPU, but the low-res display was absolutely horrible and made the laptop too painful to use very often.

Throughout the years I bought a number of Intel CPUs (386SX, Pentium, Pentium III, Pentium 4, Core Duo, Core 2 Duo, Core i7) but somehow never an AMD. Now, that’s not to say I don’t own any AMD CPUs. I have quite a few, from 286s to Phenoms and Opterons, through Am386DX, Am486DX, 5k86, K6, K6-II, K6-III, and a smattering of Athlons and Durons. But that’s not the same thing as buying a new one.

For several years, an AMD Athlon was my office machine, until its motherboard died. It was a good workhorse, but I never much liked the VIA chipset. In fact that was probably the biggest reason why I never bought AMDs until recently. In the good old days of Socket 7 boards one could pair a rock solid 430HX chipset with an AMD K6 processor. But at that time I used a Cyrix 6×86 and later 6x86MX. When Athlons came around, the most widespread chipsets for AMDs were from VIA, with all kinds of niggling issues. Later there were nForce chipsets, perhaps better engineered but with their own set of problems.

In the early 2000s I had a budget board built around the Intel 440ZX chipset, which was just slightly stripped down 440BX, a rock star among chipsets. Later I bought several Intel boards (865PERL, DG965RY, DQ67OW) and those were all extremely solid motherboards, causing basically zero problems over many years of 24/7 use.

While I believe boards for AMD-based server and workstation boards have been generally solid, my experience with desktop AMD boards was not so great. Several Phenom boards which I used at work (from MSI and ASUS) gave me all sorts of trouble. Random instability, strange NIC behavior, boards outright dying.

Whether it was Intel killing the 3rd-party chipset business or nVidia not wanting to supply chipsets to its arch-rival (once AMD bought ATI), AMD was eventually forced to supply chipsets for all of its CPUs, and I believe that was a good thing. Unfortunately this more or less exactly coincided with the introduction of the Intel Nehalem microarchitecture, and for years AMD flailed around, offering CPUs which for relatively little money produced a lot of heat and not so great performance.

As someone who used mostly Intel CPUs I found that extremely distressing, because the effect on Intel was very obvious. With no competition, Intel was able to charge a premium for each new edition of a slightly warmed over design. Want more than four cores in a desktop? ECC? More than (for years) 32 GB RAM? Forget it. Or buy a super overpriced workstation board with a workstation CPU. Want a cheaper CPU that doesn’t have half the features disabled? No can do.

When AMD Ryzen showed up, Intel suddenly discovered how to produce six-core desktops (and now even laptops), and all sorts of things started happening. Competition can do that. For that reason, I’m glad that Ryzen-based ThinkPads are now available.

Written on a ThinkPad A485 and AMD Ryzen Pro CPU

This entry was posted in AMD, PC history, ThinkPad. Bookmark the permalink.

38 Responses to My Second AMD

  1. Nils says:

    Interesting how different circumstances make people feel different.

    I started using AMD with a Duron 900 I got from my mother approx. 3 years after it came to market. From this point I had a lot of AMDs (Athlon 1400, Athlon XP 1800+ up to 3200+, Sempron and Athlon64). IIRC all of them had the “cheap” VIA chipsets.
    I never had problems with these and as a regular linux user I am used to problems with strange hardware…
    Overclocking was a fun thing (“painting” traces on the CPU to connect pins), worked well and one day I have failed to lock the heatsink properly, it felt off while running and immediately it smelt like burned hardware – the computer shut off. The CPU was colored because of the heat, but works up today (My Celeron 900).

    I also had a few i865/later machines. These were giving me a lot of trouble. The P4s went too hot. The lower-end machines are really the lower end of everything you can find. The price is horrendous. The ones with onboard graphics caused a lot of trouble when plugging in a real video card. Linux drivers were not available and the closed source drivers from intel were a hell of not working bullshit…

    Somewhere around 2009/2010 I bought a Phenom II, this was a total disappointment. The integrated graphics couldn’t handle what the box was claiming (2 screens at full hd), it was so hot in normal use, I had to buy another heatsink and still was too hot. When compiling big packages it shut off and the fake raid on this board drove me into crazyness. But these are all issues not cauesed by the CPU.
    I cant remember which board or chipset it was, I sold it after a few weeks. It was a not-so-cheap PC from one of the bigger german computer-sellers…

    Despite I am using an i5-520m laptop and an i7-4770k desktop PC I still favour AMD over Intel. I got these two machines very cheap, so I bought them. The days where every single mhz counted, one month intel had the faster CPU, the other AMD, are over.
    Also I found a way to be (almost) happy with a 8 years old laptop. Whenever I go out for buying something I’ll go for AMD, it is usually cheaper.
    If Intel shows up a with a CPU the same price as a AMD cpu, I would think “there must be something bad about this CPU, why to hell is it so cheap???”.

    One thing I never have taken into account is power consumption. Always rising energy costs make me think about… The next computer I buy, will be something that is happy with 50W under full load and has >=8GB RAM…

  2. Chris M. says:

    Lenovo has shipped AMD Thinkpads before. The x100e and x120e came with a AMD APUs.

  3. Michal Necasek says:

    I stand corrected — so they did (2010-2011). So I’ll pretend I was talking only about the mainline full-size “workhorse” ThinkPads from the beginning 🙂

  4. Daniel says:

    I suspect the issues you’ve had were all software… on the Linux side, things perform far better. AMD has been quite solid.

    Not that it’s all perfect. AMD’s power management issues in the Thunderbird days (see: s2k bus disconnect) was an embarrassment. I had an Asus/Via board from that era where the Northridge consumed almost as much power as the CPU, but with a tiny heat sink abd no fan, but I quickly decided it was unacceptable. But mostly, AMD systems have been rock solid.

    And there’s no shortage of Intel issues. I had a P4 system with awful graphics that would periodically show glitches, and require rebooting. The same system would not boot with any add-in video cards I tried, so onboard was the only option. And should we talk about early Pentium 3 EeePCs that drained the battery while powered off?

    On both sides, the situation has improved over the years as the industry stabilized and matured.

  5. Yuhong Bao says:

    “More than (for years) 32 GB RAM?”
    That has to do with the use of unbuffered DIMMs, and that the last common DDR3 was 4Gbit. It probably don’t help there is less interest in large die size DIMMs today.

  6. Yuhong Bao says:

    Thinking about it, even with Skylake laptop vendors have to decide the tradeoff between 8Gbit DDR4 support and DDR3 being cheaper in 2016.

  7. random lurker says:

    (This is just personal recollection, I do not claim to be able to back up what I’m saying)

    Personally I was amazed by the performance *and* stability provided by the Thunderbird family of CPUs. Athlon XP was an OK interim measure as they prepared to bring out the Athlon 64 family. However, they absolutely botched that job by hobbling its performance with Socket 754 (no dual channel memory, low bus speeds etc.) and they had to quickly come up with a non-server version of Socket 940 (because that required registered memory etc.) – resulting in Socket 939, which then it turned out that it wasn’t really extensible into the future, so they had to come up with AM2, and the whole idea of socket compatibility forwards and backwards being a strong point in the AMD camp turned out to be tenuous at best, not to mention the unending troubles with the K10 microarchitecture and its successors.

    Fortunately they survived and are now clawing back with Ryzen. I can’t imagine how bad or nonexistent Intel CPU developments would be without AMD.

  8. W. Ricciotti says:

    I think the AMD760 chipset, back in the early 2000s, was pretty solid.

  9. Michal Necasek says:

    I think so too, unfortunately I never had a board with that chipset back then (now I have one or two).

  10. Michal Necasek says:

    Exactly… we all desperately need AMD, whether we use their CPUs or not. The Athlon was a lot of things coming together just right, and the AMD64 architecture was learning from the mistakes all of Intel’s RISC competitors made.

  11. Yuhong Bao says:

    What is fun is that the Athlon XP family lasted until 2005 as “Semprons”, which lacked SSE2 support.

  12. Nils says:

    This is no surprise, as the Athlons also lack SSE2.

  13. Yuhong Bao says:

    What is fun is that I found out how VIA was making slides in 2003 mentioning how the Athlon XP didn’t need dual channel. Of course, try telling that to those who needed more than 1GB of RAM (back then 512Mbit DDR chips was expensive and was typically limited to DDR266).

  14. Nils says:

    Are those slides available somewhere?

  15. Chris M. says:

    Crappy VIA chipsets kept me well away from AMD at the time. The VIA 686A south bridge was notorious for causing all kinds of problems if you had a Soundblaster Live! for example.

    I generally call the entire period from 2000 (or mid-1999 depending on how much you hate RAMBus) to 2006 the “era of crap hardware”. Between the capacitor plague, notoriously low quality control, and the Ghz wars between AMD and Intel (adding a ton of heat to already marginal hardware), its nearly impossible to find a stable machine from this period. I worked as a tech during this period (adding to the fun), and almost all my repairs during this time period were fighting buggy chipsets or intermittent hardware failures… and it was overwhelmingly AMD platforms (way too many nightmares of VIA 4-in-1 drivers). Intel platforms were less trouble-prone, but overheating and the occasional non-Intel chipset boards caused headaches.

    Once the Core2Duo came out on the Intel side and AMD bought ATI and started making their own chipsets, things improved greatly.

  16. Yuhong Bao says:

    Trivia: ALi was the last chipset vendor to include the I/O APIC, with it taking until 2002 before the M1535 southbridge series was replaced.

  17. Michal Necasek says:

    I think those were the last CPUs (at least more or less mainstream ones) without SSE2, right? All AMD64 CPUs had it, all Pentium 4s had it, and Pentium M had SSE2 as well.

    Although what most Pentium Ms didn’t have was PAE/NX.

  18. Yuhong Bao says:

    Yep, that is what I mean.

  19. Michal Necasek says:

    I used a VIA 686B based Athlon board which was somewhat unstable until it completely died of bad capacitors. Can’t blame the capacitors on VIA really, and I don’t know if the instability was caused by bad caps too, or if it was something else.

    On the Intel side I bought a 3.2 GHz Northwood P4 sometime in 2003 (with Intel chipset and board) and I can say that it was hot, loud, reasonably fast for the time, and very stable. In fact the machine still worked last time I checked on it, less than six months ago.

    I suspect Intel focused a lot on the Itanic in that era, and I don’t know if the P4 was a product of them just not caring or consciously damaging the x86 architecture. But AMD64 and Microsoft put an end to that.

    There are fascinating rumors that Intel implemented the AMD64 architecture in silicon before AMD did, working from AMD’s public architecture specs. It might explain some of the Intel vs. AMD architectural discrepancies where Intel follows ancient AMD documentation but not actual AMD CPUs.

    I also wonder if AMD underestimated just how well their K5 and K6 CPUs worked with Intel chipsets; they traded that off for a much better Athlon performance, but the cost in stability and support was significant. A 430HX board paired with a K6-III can get quite respectable performance.

  20. Yuhong Bao says:

    Of course, the funny thing is that even Intel had to do it for a while after the Rambus debacle. They were lucky that VIA rushed to add SMP support to the Apollo Pro 133A chipset just as the problems of the Intel 840 chipset was becoming apparent, with a stepping of the VT82C686A that added the I/O APIC.

  21. hobbes says:

    “Although what most Pentium Ms didn’t have was PAE/NX.”

    That’s not true, *all* Pentium Ms have fully functional PAE, in early models it’s just not exposed in CPUID due to a CPUID bug, so not unlike the *opposite* of sysenter/sysexit in Pentium Pro. Later models fixed that issue and also added support for NX.

  22. Michal Necasek says:

    PAE/NX without NX is not PAE/NX anymore. Or did all Pentium Ms have NX? That sounds unlikely.

    I really don’t think it was “buggy”, it was clearly deliberate. For that reason you can’t say that all Pentium Ms have fully functional PAE, only that the ones where PAE was forced seemed to work. At least I couldn’t find any information about whether it was a purely marketing decision to not show PAE or if there were some technical reasons. It’s Intel, either is possible.

    Anyway, not very relevant for Windows because the OS looks at CPUID. So if PAE isn’t listed there, it’s not there.

  23. Richard Wells says:

    Intel’s webpage for the Pentium M claims that some models have NX though rather inexplicably Intel made an update to many of those models which removed NX bit support. The 7?5 models without NX were released about 3 to 6 months after the similarly performing 7?0 models with NX. Having the NX bit was rather necessary for the impending micro-server market. See for details.

  24. Michal Necasek says:

    The official Intel documentation is confounding, to say the least. In ark, Intel says Pentium M 745 has NX (aka XD), which would also require PAE. I’ve seen suggestions elsewhere that the support might be silicon revision dependent.

    The spec update also talks about 65nm Dothans that no one has apparently seen in the wild.

  25. Yuhong Bao says:

    Intel would not typically do market segmentation based on the NX bit, given that it is an important security feature.

  26. Richard Wells says:

    I can’t say if Intel wanted to market segment on the XD bit but I am fairly sure that Dell at least did. A quick search shows Dell putting the 740 (with XD) in business laptops while the consumer laptops got the 745 (matching CPU without XD).

  27. Michal Necasek says:

    That’s exactly what I’d worry about. The feature must be considered untested, and it’s entirely possible that it works on 99% of Pentium Ms but not all of them. Without at least testing all known S-specs it’s impossible to know.

  28. hobbes says:

    “PAE/NX without NX is not PAE/NX anymore. ”

    Right, but “PAE/NX” can be misleading when they’re two separate features.

    “At least I couldn’t find any information about whether it was a purely marketing decision to not show PAE or if there were some technical reasons. ”

    I suspect market segmentation, their reason being that no laptop would need 4+ GB of RAM at the time. That would neatly avoid Pentium M eating into the high performance and server markets, where Intel was pushing their lackluster NetBurst monstrosity instead, and where Pentium M *could* be used because socket 479 to 478 adapters were a thing.

    I don’t consider technical reasons probably simply because they had *a known working implementation of PAE already*, due to the fact that Pentium M is just an evolution of P6, i.e. Pentium Pro, II and III. In fact, I would be surprised if there are significant silicon differences in PAE between the two: other than bigger and faster L2 (but L2 was changed several times in P6 too) and hardware prefetching (which was already present in Pentium III Tualatin), I don’t know of any differences in the memory subsystem between the two.

    Of course, PAE being undocumented and likely untested on Pentium M, I wouldn’t be surprised if some CPUs had silicon defects in the pertinent areas which caused malfunctions, and still be treated as good parts *because* it’s undocumented.

    “Anyway, not very relevant for Windows because the OS looks at CPUID. So if PAE isn’t listed there, it’s not there.”

    You can always patch that check out 😉

    “I’ve seen suggestions elsewhere that the support might be silicon revision dependent.”

    If I remember correctly, that’s correct for Dothan, the same model could support NX or not depending on the stepping. I think the stepping that added NX was C0. Banias of course never had NX, and Yonah/Sossaman always had it.

    “Intel would not typically do market segmentation based on the NX bit, given that it is an important security feature.”

    Until Goldmont and Skylake (so very recently!) Intel used to segment based on AES-NI, which is arguably also an important security feature: it’s nigh impossible to implement AES without it in a way that both has acceptable performance and is impervious to side channel attacks. At this point I tend to assume that if there’s some twisted way to justify market segmentation, Intel *will* find it.

  29. Michal Necasek says:

    To clarify, I speak of PAE/NX because modern 32-bit operating systems require NX, which in turn requires PAE, so in practice both are needed.

    I absolutely agree that if Intel can come up with a way to segment the market on some random feature, they will do it (thank goodness AMD doesn’t do it that way). It need not make any sense. And I am convinced the PAE removal was a marketing decision, not a technical one.

    So let’s summarize the history: The Pentium M and Opteron (which included NX) came essentially at the same time, in the first half of 2003. The earlier Pentium III-M included PAE, presumably out of inertia. The Pentium M did not, probably a) for market segmentation reasons, and b) because typical Pentium M chipsets never supported more than 2GB RAM. Only that turned out to be a case of really bad timing for Intel, because in 2004 Windows XP started supporting NX (DEP in MSFT-speak) and in the later Dothan models (Jan 2005?), Intel re-enabled PAE and added NX.

    Where it gets really fuzzy is all the Pentium Ms that do not advertise PAE — we know the P6 core had PAE, and apparently it was functional in Pentium Ms even when not reported in CPUID. But not in all of them. Faulty silicon? Broken steppings? We don’t have enough data. Could some of those even support NX? No idea.

    I have seen numerous suggestions that C0 stepping Dothans all had PAE/NX, although finding reliable data is next to impossible.

  30. Richard Wells says:

    The best candidate for finding a Pentium M that supported NX would be the blade servers from a number of companies. Certainly, HP talked a good game about their implementation having full security. This was before the official listing of support for NX by Pentium Ms and before the dedicated server variants were produced. I have my suspicions as to what was involved in those designs but the long descriptive white papers seem to be retired and I don’t have one to test.

  31. Yuhong Bao says:

    AFAIK with Pentium M they enabled NX when they moved to a 533 MHz FSB.

  32. Michal Necasek says:

    The Pentium M datasheet is actually unusually clear: “Pentium M processors with CPU Signature = 06D8h will also include the Execute Disable Bit capability.” The 06D8h signature should correspond to C0 stepping.

    You may be thinking of a document similar to this: I can’t find any mention of NX anywhere, but the models listed there are exactly those that were available (including in FC-PGA) as C0 stepping.

  33. Michal Necasek says:

    Yes, from what I can tell all 533 MHz FSB models include NX, but some 400 MHz FSB models do too in newer steppings (notably the 1.4 and 1.8 GHz models, aka 738 and 745).

  34. Yuhong Bao says:

    Interestingly, ALi does not have a K7 chipset based on the M1563 southbridge that supports the APIC, choosing to focus on the K8 instead. That probably made sense as ALi had a P4 FSB license while VIA’s P4 chipsets was facing legal troubles.

  35. Yuhong Bao says:

    Thinking about Skylake laptops and DDR3/DDR4, I am thinking that it all has to do with the 20nm transition. On 25nm the die size of 8Gbit DDR4 is still pretty large (~100mm^2) and 4Gbit DDR4 die size is larger than 4Gbit DDR3. This presented OEMs with a hard problem.

  36. Yuhong Bao says:

    Of the early AMD K8 chipsets in 2003-2004 BTW, I am thinking that ALi/ULi was probably the best one. Especially since the chipset interconnect used the same HyperTransport bus as the processor itself.

Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.