To the outside observer, improvements in PC architecture are evolutionary but logical. Processors advance inevitably in speed and performance, in happy accordance with Moore's Law. For Nebojsa Novakovic, a consultant in high-end computing systems, that's hardly the case. The demise of the DEC Alpha processor is a case in point. A performance leader was killed off by corporate whim.
Novakovic argues that despite the turn by AMD to a 64-bit platform, vindicated by the success of the Athlon 64, AMD had better watch out. Intel will be fighting back in 2006 with Merom, Conroe and Woodcrest, as processor design adopts a multicore approach. Even so, life won't be a bed of roses for Intel either. Rumors abound in the industry of the "retirement" of the Intel Itanium, an expensive attempt to break away from the x86 architecture at the high end, and Intel sorely lacks an interconnect that could compete with AMD's HyperTransport.
Novakovic comments on the major technology issues, in this in-depth interview, as Intel and AMD square off for the next rounds of the processor wars.
This is Part III of a five-part interview. Part I appeared on 23 January and Part II on 24 January. Part IV will follow on 26 January.
Q: We're awaiting the arrival of a new-generation core from Intel, with the Merom, Conroe and Woodcrest processors. What scenarios do you anticipate, in terms of performance and the response from end-users and AMD?
A: According to the information that we have to date, the Merom/Conroe/Woodcrest core looks better than the Athlon 64 core, from the point of view of performance and power consumption. However, one thing that that the Merom/Conroe/Woodcrest core will lack is an interconnect that will scale smoothly in high-end systems. For PC systems, where the single-processor approach is normally adopted, unless Intel screws up, in terms of execution, which is always a possibility, I think that Conroe will probably emerge as the performance leader on the desktop in the fourth quarter of next year. If Conroe appears on schedule, and it has a clock speed of at least 3.0GHz, and I believe the clock speed will be at least 10% above that for top-end parts, we think it will have quite a performance advantage over the current Athlon 64 scaled up to above 3.0GHz, if it is still available at that time. This is assuming that AMD does not miraculously decide to have a surprise launch of a new core before then.
My belief is that Intel will regain much of its performance leadership on the desktop and probably also in small servers. And where servers or workstations have a chipset supporting two front-side buses (FSBs) in parallel, as well as a large memory system, the next-generation Woodcrest could come to have the performance leadership for this server segment. And with the right chipset support, it's possible they could also compete fairly well against dual-socket Opterons using DDR2 memory, and they should be capable of reasonably equivalent performance.
In the quad-socket area, I think AMD will continue to lead there, even if Intel introduces complex chipsets supporting four FSBs simultaneously or tries similar tricks. If Intel tries that, it could just make the system overly complicated or too expensive. Certainly in single-socket systems, Intel should, hopefully, be shipping a pretty good platform in a year's time. And that should be a warning to AMD, because a new platform could signal renewed dominance of the x86 platform – a strategy quite different from the attempted dominance of the Itanium and IA-64 – and that could signal a serious threat to the performance leadership of the current Athlon 64 core. So just amending the socket architectures, including that of DDR2, and making some other minor adjustments may not be enough for AMD, in this case.
AMD will have to find ways to maintain market share, and one way of doing that would be drastically expanding the multicore approach, and that would mean not waiting until 2007 for a quad-core approach or the introduction of a new, faster core.
The alternative, for AMD, would be to improve current performance by, for example, having better integer throughput and better cache memory. Another possibility – and this could apply even more to Intel than AMD – would be to add hardware-accelerated functionality in the Southbridge, enabling high-definition (HD) sound, RAID 5, Gigabit Ethernet, a firewall and wireless capability. These functions could be handled by an internal 32-bit multithreaded embedded processor within the Southbridge, with hardware acceleration as well as DSP and DMA support. This would avoid burdening the CPU with a continuous stream of interrupts, and it would not add much cost if all high-end Southbridges for the desktop adopted this approach by default.
Alternatively, AMD could speed up their current floating-point (FP) performance, by, for example, having two SSE2 (FP) instructions in each cycle, rather than every two clock cycles. But at the end of the day, rather than these simple fixes, you need a new core on the AMD side as well. That's particularly the case when you realize that the current Athlon 64 core is essentially a 64-bit version of the old Athlon (32-bit) core. Except for the added support for the SSE2/3 instructions, there is not that much difference between the two cores, so in a sense, the time is right for a new core.
In fact, AMD would have to continuously reassess many aspects of the Athlon 64 architecture, and even then they need to be careful, with Merom, Conroe and Woodcrest due from Intel later this year. Faced with these new cores from Intel, AMD is now more vulnerable to mistakes and could make Intel's mistake of casting around for direction. AMD is far more vulnerable to such mishaps, but they cannot allow themselves such mistakes. In a sense, AMD has always been rather reminiscent of Alpha, in that they have often had good technology, but execution that doesn't match. This even seems to apply at a managerial level, where sometimes I sense there is no mind-set to push the architecture in a big way and really grasp the opportunity that they have – and win. I mean, Intel is not standing still, and the expectation is that Merom/Conroe/Woodcrest and later iterations of this core could tip the balance quite strongly in Intel's favor.
Q: So the corporate culture does have quite a strong influence, at the end of the day?
A: Well, look what happened to Alpha! It's question of what you do with technology, whether it's Alpha, MIPS or an Intel CPU. I mean a boardroom decision, for whatever reason, possibly for the benefit of the individuals involved, or the shareholders, which I doubt, was made to kill Alpha. This was not a decision made by engineers. I mean it's very simple. If you want to strip off a company's assets before you sell it, you take the best assets on offer – you either sell them or you kill them.
People have finally walked out of those companies with big bonuses and pay-offs, basically throwing money away and leaving the casualties lying on the floor. That's what happened, and that's what continues to happen, and not just in the IT industry.
Q: In terms of corporate culture, it seems a pity that Intel remains such an incommunicative entity.
A: Well, I would say that five years ago that was definitely the case. Things have changed substantially in the meantime. These days, I think Intel is way more open and productive, and sometimes AMD can be more open and productive. It depends, both on the situation and what you are talking about. If you step on the wrong foot, AMD can be just as self-protective as Intel is, if not more so.
Generally, I think Intel has modified its attitude and is more communicative. After all, they now have to fight to keep their leadership position in the industry. That is the reality of the situation, right now. And if you ask about this question in the distribution channel, I think you'll hear that as AMD has grown in strength, the strong-arming of the channel has now appeared on the AMD side also. It's the same old corporate story. When you are weak, you need to gain friends. When you are strong, those friends now appear disposable. Intel has now lost some of its position as an industry leader, so now it has to turn around and fix the problem.
Both sides, both AMD and Intel, have their pluses and minuses, but the current situation is one where neither side has made any dramatic improvements to their cores over the past 18 months. That tells you that despite advances in the technology from 0.13-micron to 90- and now 65-nanometer nodes, we've seen very little per-core performance improvement. And it's no secret that both companies are now adopting a multicore approach to performance improvement and have been for quite some time. The multicore approach is not really anything new. Five or six years ago we saw POWER4 adopting a multicore approach, and there were similar plans for MIPS and Alpha.
The important point about AMD64 is that the Opteron and Athlon 64 were planned from Day 1 to be multicore processors. This was partly the influence of Alpha, where these questions had been under consideration for some time. When the Athlon 64 and Opteron platforms were first presented at the Microprocessor Forum, they were shown with the second core present. The presentation indicated there was provision for the second core from Day 1. The plan had always been that improved performance would be gained by having the two cores communicating internally, at full speed. That is the advantage that AMD has, with its current generation of processors – they had a second core ready, and they could extract better performance, using the second core. So, with its current generation of processors, the AMD solution is much more elegant, while the Intel solution has been something of a last-minute fix. Essentially, Intel simply bolted together two cores for the Intel Extreme Edition Pentium.
Even so, while Intel's solution was architecturally clumsy, it did work, and it turned in record performance. I have tested both solutions, and while AMD has the performance lead in most areas, there are a few areas where Intel maintains the lead. That’s particularly the case if you take into account the wider margins for stable overclocking on Intel’s latest CPUs. And of course you have to bear in mind that while AMD does have that performance advantage right now, it's not forever. By the time Merom, Conroe and Woodcrest have appeared, AMD's performance lead will have evaporated. This is because the new Intel chip has been completely re-designed to take advantage of the dual-core approach, and AMD will no longer enjoy the advantage of internal communication between the cores. (The Yonah Centrino Duo is already using full-speed inter-core communication!) In fact, communication between the cores will be faster on the Intel platform because now they will have faster L1 caches. In addition, Intel is throwing in four instructions per cycle per core. Now, it's time for AMD to wake up and start to figure out what its response is going to be.
This is Part III of a five-part interview. Part I appeared on 23 January and Part II on 24 January. Part IV will follow on 26 January.
Nebojsa Novakovic is a Singapore-based consultant for high-end computing as well as maglev transportation systems. He has been active in various projects throughout the Asia Pacific for over 10 years, the most recent being high-end technical computing clusters using top-end Intel and AMD platforms in combination with a Quadrics high-speed interconnect. His IT commentaries, covering high-end computing issues in particular, have appeared in numerous publications, including Singapore's The Straits Times, and he is a frequent contributor to the well known www.theinquirer.net website.
Photo: Nebojsa Novakovic
Article edited by Chris Hall
