Posted on 06/20/2003 2:33:04 PM PDT by Ernest_at_the_Beach
AMD's Opteron: Does it measure up?
Is the mainstream IT media in Intel's pocket?
By : Thursday 19 June 2003, 12:17
When AMD launched Opteron, it was pretty much a server product launch, as there were no workstation boards available. Of course, this will soon change.
Is the mainstream IT media in Intel's pocket?
It has been almost two months since Opteron's launch, and what's really surprised me has been the poor response from the mainstream IT media, especially when one sees how Opteron performs in independent benchmarks. Hardware web sites have justly reported on Opteron's outstanding performance, but this news event hasn't transferred to the IT mainstream as headline grabbing reports. Don't get me wrong, there has been reporting on Opteron's performance, but nothing that really nails the sail to the mast. It's worse than the dumbing down that Tyan meted out to its own Opteron motherboard. It smells of leveraged censorship.
Google "Opteron performance" and you'll find plenty about Opteron before and after launch day. Post this event, the lack of forthright reporting on Opteron's performance has been extraordinary. Opteron's performance doesn't require investigative journalism to report upon, so why the silence of the lambs? Did these IT bureaux also receive that fateful telephone call as well? Which may have asked in these "difficult" economic times how much they "really valued" their advertising revenue.
Even with the May 22 official confirmation and posting of Opteron's SPEC results, which confirmed Opteron as the best performing two- way and four-way servers in the world, the mainstream IT media have again only offered lip service to Opteron's record breaking feats. I could only find one mainstream report on that story, which was just a copy and paste job of AMD's press release.
The poster child for this issue happened just recently. In a piece about Itanium performance, Itanium gets publicity and praise for only "one" benchmark result, which wasn't even a new benchmark record. Opteron has broken a heap of benchmark records, yet you'd be hard pressed to find a mainstream IT news story that has reported commensurably on that event. Why?
First, let's look at the Itanium's history. Itanium was stillborn. It was great practitioner skill that revived it back to life. Since that time it's had to exist on life-support. Only recently has Itanium been able to stand unaided, but it's still confined to isolation. If Itanium's immune system is exposed to any more "bugs", death may follow rapidly. So this still leaves Itanium as the sick man of the IT world. In the almost decade of development and the billions of dollars invested, Itanium has not yet returned a profitable dime. If Itanium ever hits the ground running, how many years before the project moves into the black? So financially speaking, Itanium is on its deathbed - a prognosis that could prove to be terminal, or one that may take years to recover from. With all of these shortcomings, why the disproportionate exposure?
It's interesting to note that the Aberdeen Group had predicted broad deployment of Itanium in the first half of 2002, and by 2005, Itanium would garner 42% of all worldwide server revenue. Well, we're now in 2003, and I wouldn't call 0.003% for this year's projected server processor share as "broadly deployed". That percentage number is generous, as it's based on the total estimated output for all Itanium processors, which includes those for workstations. The Aberdeen Group was not only looking through rose tinted spectacles when it extrapolated its forecast, it had added over glasses that were tinted with a very bluish hue.
One can highlight Itanium's eon in development and its commensurate dollar engulfing black hole. But more importantly, if it's taken this long for the experts - we're talking about Intel and HP - to bring Itanium to utility, even if it's not yet profitable, how long will it take the "non-experts" - the end users who will use this technology, to become fully conversant with Itanium's programming model? Len Tsai, who was chief architect at NEC, dropped the bombshell that probably cost him his job: "Learning EPIC instructions will take time. It will take 15 years to educate people about it. That's half a generation, and I am getting old." Using ball park figures, if it takes 15 years to educate people about the instruction set, the Itanium project will be 25 years old before it's ready for prime time. If Itanium reaches that age, then maybe the Aberdeen Group's forecast about broad Itanium deployment will be realized.
Second, let's look at Itanium's relevance. AMD has forecast that this year's worldwide server processor market to be around eight million units (page 11 of PDF). This year's total Itanium volume is estimated at less than 23,000 processors (page four of PDF). AMD64 output is planned to be in the millions this year, which Opteron is part, and tens of millions next year (page four of PDF). With Itanium's paltry contribution to server sales, why the disproportionate exposure?
It's not just the dumbing down of Opteron's performance that concerns me. Athlon related technology has the same problem.
Nvidia's announcement for its dollar saving multi-PC image, which the INQUIRER reported on, received virtual zero coverage from the IT mainstream. A week later, Intel's Stable Image Platform Program (SIPP) was widely reported. The contrast between these news events is stark. Intel's announcement pledges to keep platform configurations stable for 12 months, which "may reduce" IT support costs. Nvidia's press release announced a multi-PC image based on its nForce and UDA technologies, which is something uncommon in this industry. Northeast Utilities, a Fortune 500 company, was able to use one system image for two different nForce-based PCs, with perfect compatibility to boot. The press release also linked the company's testimonial, which gave a detailed account of how the company's multi-PC image "did reduce" IT support costs. Again, why the disproportionate exposure?
Can all of this be traced to the fear of one company? In a piece about Intel strong- arming, the author described vendors at the Opteron launch as being "scared to death". It doesn't take a genius to figure out which company's products garners the highest exposure on IT web sites, and the consequent dollar return in advertising revenue that is generated. If Intel can exert the fear factor on hardware vendors, then doing the same to IT web sites should be a cakewalk. If that's the way the game is played, then it's no better than using a bagman to influence the media.
If it's just pure ignorance that has confounded the IT mainstream media about Opteron's performance, then now they won't be able to plead stupidity any longer. AMD has released a new PDF - The AMD Difference, which makes Opteron's performance advantages crystal clear. So with AMD's third attempt at getting Opteron's record breaking feats fully into the public domain, and the joint Microsoft and AMD live Webcast today, then maybe the IT mainstream media will now truly listen up and see through its journalistic responsibilities.
It's not all doom and gloom for AMD and its partners. InfoWorld named Newisys CTO Rich Oehler 2003 InfoWorld Innovator. This recognizes the work that Oehler, his company, and AMD have done in bringing Opteron from design to being the best performing product in its class. I'm sure AMD, its partners, and the people behind the technology are thankful to receive such accolades. But I'm sure they would all be far more grateful if Opteron had received the media plaudits that its outstanding performance deserves, and the commensurate publicity that its market outlook demands.
Anyway, I shall redress the balance here. I shall report on the Opteron launch, the benchmarks, the serious challenge that now faces Intel, the future of the Opteron platform, Opteron's upgrade flexibility, and the aftershocks of Opteron's arrival.
Opteron launch
The launch started with a video and a question: "One last question. Where did the idea for creating breakout performance products come from?" Hector Ruiz, AMD president and CEO, responds, "Well, that is quite a story." Jerry Sanders, AMD chairman and company founder, then says, "Whenever I talk to the MIS directors, I always use the same thing. You've got to find a way to simplify the business. The future is 64-bit computing. I've got some ideas how we can 'hammer' that home." The phone interrupts Sanders, so Ruiz continues the story: "I knew we had to create a new class of computing, a breakout performer, but how?" Sanders breaks in: "Well you know how I feel about that. It's always about the customer. You have to listen to what the customer says. I've been saying that since I started the company 34 years ago. Listen, listen, listen." Ruiz adds, "We needed to listen to everyone. Find out what they wanted. Find out what they needed. Listen."
Did AMD listen? Is Opteron really the breakout performance product that AMD claims? The best people to answer that question are from the companies that AMD listened to, so let's hear what they had to say.
AMD VP Marty Seyer gave the background on the first of these companies:
"Speaking of leadership, Computer Associates is the leader in system enterprise management software, providing business critical technology that shapes the way business is conducted around the world. CA's client base runs every type of application imaginable on a variety of platforms. One thing that they have in common though is to lower the total cost of ownership. This is what CA discovered when it tested 32-bit and 64-bit computing on AMD Opteron processors."
Emma McGrattan, VP of R&D for CA:
"The Advantage Ingress development team and AMD have worked closely together over the past 12 months to bring to market a value proposition for the 64-bit database market place that's going to be unbeatable. The beauty of the AMD Opteron port for us was the fact our clients don't need to make a single change to the current applications to benefit from 64-bit computing. The clients can run 32-bit applications accessing a 64-bit server, and what we found because we can access data from cache rather than going to disk - the performance on the database engine is phenomenal." She adds, "Moving from a 32-bit to a 64-bit environment, we're seeing in applications that would normally be disk intensive, we're seeing about a 10:1 performance improvement in going to 64-bit where we can store more data in memory."
She further says, "The Advantage Ingres port to the AMD Opteron processor took us two days to complete the entire port cycle as well as any testing that we needed to do. We were able to just rebuild the product and we didn't hit a single glitch in running through our tests with this. The performance benchmark that we ran as part of the test process was something that we had never seen before, and we had to go back and check that all our calculations were correct in generating the statistics for the benchmark because the performance is unbelievable."
Michael Tiemann, CTO for Red Hat, reiterated Jerry Sanders' point about simplifying the business (begins at 1:50):
"Last year, when I was talking with a major financial services company who's looking at making a major migration from legacy to modern systems, the question the CIO asked was what few things need to be the same so that everything else can be different. And to me, that's a remarkable question because it really opens up the question of architecture and the question of platform. It means that technology becomes a supporting rather than a limiting factor.
In a lot of technical fields, whether its risk analysis, cash management, oil and gas exploration, people are looking for the performance of the 64-bit environment. They're looking for the scalability of a 64-bit environment. These large scale applications demand 64-bit processing, which the AMD Opteron processor delivers."
Robert J. Picciano, Director of Database Technology for IBM:
"Information in the market place today is growing in leaps and bounds at over 250 MB a year per person on the planet. So you need an advanced leading edge information infrastructure to provide the access to that information, to secure that information, and to deliver it no matter what the application requires.
With DB2 on the AMD Opteron, the customer gets to make the choice about when they move from 32-bit to 64-bit. In porting [DB2] to the AMD64 platform I was amazed. My guys came to me and said we got it done in two days. Now you have to look at something here. DB2 is about a ten million line of code product. It's the world's most advanced product written in C++. Now porting isn't even in our lexicon, we enabled and optimized each platform we work on. So when they came to me and said it was done in two days, I was floored. What a phenomenal accomplishment."
Ted Mariner, VP of Advanced Hardware & Software Systems for Veritas DGC:
"Veritas DGC provides geophysical services to the oil and gas industry. We acquire data, we process that data, and we try to produce a accurate sub-surface image of the area that we're exploring to allow oil companies to correctly target oil and gas that is in our sub-surface.
If you can look at the way Veritas uses IT - compare it to a factory where for us the data are our raw materials. We have to feed that data through our IT factory. We need low cost, high performance, and a massive amount of compute capacity.
The cost benefits of AMD's 64-bit solution are that to us it's on the same curve as the 32-bit commodity hardware. All the other 64-bit solutions that we have available to us at the moment in the RISC-based area are really on the RISC price curve, and we're looking at maybe 3 to 4 times the overall cost of a system. So to us it's a no-brainer.
Really, I think AMD have been very, very good at being proactive about wanting to come forward and helping. How can we help? What do we need to do to help you? What do we need to do to provide you with a service? That is a really refreshing thing to have in this industry where most people are saying well, how many do you want? This is how much it is. When will you settle the bill? We're really impressed by that side of it.
If anyone is integrating 64-bit systems for the HPC cluster environment, they will have to make use of AMD Opteron. If they don't, they will not be in that business."
These are just four quotes from a whole list of endorsements that AMD has been able to garner for its AMD64 platform. For balance, here's Intel's endorsement page.
"Unbeatable", "unbelievable", "phenomenal", "it's a no-brainer" are just a few of the superlatives that have been used to describe what Opteron delivers. When Larry Ellison's Oracle gets fully onboard, we'll no doubt have to add his now famous "unbreakable" to that list. One thing should be obvious. These were candid comments from industry leaders who value their reputations. Another is that AMD has been communicating with its customers, but more importantly, it has also been listening. Contrast that to Intel's megaphone diplomacy:
Upgrade or get left behind!
In a briefing before Intel's spring analyst conference in New York, CEO Craig Barrett effectively told the business world in Europe, Japan, and the US: Upgrade your IT infrastructure or get left behind. Intel says that many companies are working with office computers and servers that haven't been replaced for at least four years. At the conference, CEO Paul Otellini showed a slide that said 500 million PCs are below 700 MHz.
If Intel had been communicating with its customers and had delivered on a technological roadmap that made sense, surely there would be no delay or doubt about upgrading to Intel platform infrastructure. It seems clear to me that the reason for the delay is not just economic, but has more to do with platform decisions that still have to be made, be it IA-32, AMD64, IA-64, or something else. Remember, decisions made today will have a lasting impact well into this decade. With the drubbing that Opteron has already meted out to both Xeon and Itanium, many IT decision makers are no doubt re- evaluating the IT landscape as it exists today. Also, with the latest problem that has afflicted Itanium, many may have already concluded that Itanium's staying power will have crumbled before it can make any serious impact in the market place.
During Otellini's conference presentation, he couldn't muster any serious ammo with which he could attack Opteron with, so he ended up firing blanks instead. In comparing Prescott to Opteron, Otellini said that its die size would be half that of Opteron. Has Otellini still got his head buried in the sand? Key northbridge functions are now incorporated on Opteron's die, three HyperTransport links are also on-die, and it's also equipped with 1 MB of level two cache. So naturally, Opteron's die size is going to be larger than previous AMD designs. If one wants to make real estate comparisons, we should be including the die size of an Intel platform northbridge for an apples to apples comparison. And what about Itanium? Its die size is over twice that of Opteron without the northbridge.
Barrett's upgrade or get left behind message is either a plea of desperation on Intel's part to kick-start the upgrade cycle into action, and/or it's a recognition that Opteron has arrived and now has to be part of Intel's enterprise calculus. When Otellini can only make irrational die size comparisons against Opteron, it really says that Intel has no short term answer to Opteron. The benchmarks make that point abundantly clear.
Benchmark highlights
When Opteron's most important performance attribute permeates IT decision makers - that is its ability to scale with hardware, software, and workload, two questions will top all others when other platforms are compared: How does it scale against Opteron? What is its price/performance ratio?
Scalability will be Opteron's calling card when it's evaluated for possible use, be it its ability to scale in benchmarks or its simultaneous ability to run 32-bit and 64-bit code. These are qualities and differentiators that are non-existent or severely lacking in Xeon. Itanium can scale and has an ability to run 32-bit code. Unfortunately, one has to pay through the nose to scale Itanium and its speed at running 32-bit code falls far short of Opteron. Opteron beats Xeon hands down in the 32- bit world; in the 64-bit one, when the cost of Itanium is factored in, Itanium is found to be wanting.
AMD used database, email, processor performance, web serving, and Java application server benchmarks - developed and validated by independent third parties, which shows that Opteron shines. I won't give a blow by blow account, but I will summarize the outstanding results.
The database TPC-C performance benchmark shows that the four-way 1.8 GHz Opteron platform outperformed the four-way Xeon MP, but fell just short of the four-way Itanium. Opteron made up for this by coming first in the TPC-C price/performance comparison. At $2.76, this is the result that really matters, as it left all other contenders a very long way behind. Also, it's the only four- way server in the top ten (4th place to date) of TPC-C's price/performance table, so this tells you were all the other players are. Today's best TPC-C price/performance result is $2.47, but this is for a single processor server. Opteron's outstanding four-way TPC-C price/performance opens up new opportunities to sell four-way servers at unheard of price points.
The email serving benchmark shows Opteron ahead of both the two-way and four-way Xeon-based systems. What's interesting to note here is how Opteron scales from two-way to four-way. For the two-way result, the 1.8 GHz Opteron platform bettered the 2.8 GHz Xeon system by 12%. The four-way result shows the same Opteron processor competing against a 2 GHz Xeon MP system that has over twice the on-die cache, yet it still trails Opteron by 18%. It just goes to prove how efficient Opteron's interconnect infrastructure really is, and it also illustrates again that it's not how much on-die cache or how fast a processor runs, it all boils down to solid overall design.
SPEC's processor performance benchmark results show Xeon was again second best to Opteron, but Itanium proved its prowess in floating point math, even though its integer performance was abysmal. But again it's Opteron's scaling ability which has to be mentioned here. For the SPECfp®_rate2000 results, the four-way Opteron is 144% better than the four-way Xeon MP, and just about equaled the four-way Itanium. For the SPECint®_rate2000 results, the four-way Opteron is 40% better than the four-way Xeon MP, and 60% better than the four-way Itanium. Opteron just gets better as it scales.
The SPECweb®99_ssl (Zeus) web server benchmarks show that the best Opteron system was over 50% better than Xeon in the two- way results, and over twice as good in the four-way results. The best Itanium platform, which had 6x the level 2 cache of Opteron, delivered comparable results. The SPECweb®99 (Red Hat CA2) benchmarks show the two-way Xeon was 16% worse than the best two-way Opteron. But when Opteron is scaled to four-way, the best Opteron system bettered the four-way Xeon MP by 51%.
The SPECjbb® 2000 Java application benchmarks show that a single Opteron is able to outperform a two-way Xeon system by 19%. A two-way Opteron increases that to 91%. For the four-way result, Opteron betters the Xeon MP by 19%.
Benchmark analysis
Here's how AMD summarized Opteron's record breaking SPEC results:
1. Highest recorded 2P and 4P SPECweb® 99 scores, breaking for the first time the 10,000 barrier on a 4P system
2. Highest recorded 2P and 4P SPECint®_rate2000 scores
3. Best scaling 1P to 2P to 4P SPECfp®_rate2000 scores
4. Highest recorded 4P SPECweb 99_SSL scores
5. Highest recorded 1P, 2P, and 4P SPECjbb® 2000 scores
What should be clear, to even the Intel diehards amongst us, is that Opteron rocks. It beats Xeon where it matters in the 32-bit world, and scales even better in the 64-bit one. Opteron also scales magnificently on hardware, but at far less cost than Itanium. If IT decision makers decide to go with Opteron instead of Itanium, serious up-front cost savings will be made. Also, because Opteron scales so much better than Xeon - with hardware, software, and workload, the precious IT dollar will stretch far further when Opteron platforms are purchased.
Xeon MP price/performance - it's
not looking good for Intel
These two-way and four-way systems were configured with 512 MB of RAM and the cheapest SCSI drive on offer. Since these Dell systems have CDROM drives as standard equipment, this Einux option was included. No other options were added. Einux has removed its config page, so bear in mind that its pricing is about a month old. Dell's pricing is newer.
Even though purchase cost is a small factor in the total cost of ownership, in this time of economic restraint, it's still illustrative to compare the different platforms. Controller cards from Einux are substantially less expensive than from Dell, but Round Rock's buying power does bring down the cost of Windows. At the time of the original comparison, the 18 GB and 36 GB U160 (15K RPM) drives came in at the same price.
Server pricing
PowerEdge 6650 dual 1.5GHz Xeons - $10,214
Einux Quad Opteron 840 w/2CPUs - $10,074
Einux saving - $ 140
PowerEdge 6650 quad 1.5GHz Xeons - $13,014
Einux Quad Opteron 840 w/4CPUs - $12,074
Einux saving - $ 940
PowerEdge 6650 dual 1.9GHz Xeons - $11,713
Einux Quad Opteron 842 w/2CPUs - $13,074
Dell saving - $ 1,361
PowerEdge 6650 quad 1.9GHz Xeons - $16,713
Einux Quad Opteron 842 w/4CPUs - $18,074
Dell saving - $ 1,361
PowerEdge 6650 dual 2.0GHz Xeons - $15,713
Einux Quad Opteron 844 w/2CPUs - $15,074
Einux saving - $ 639
PowerEdge 6650 quad 2.0GHz Xeons - $25,713
Einux Quad Opteron 844 w/4CPUs - $22,074
Einux saving - $ 3,639
When these servers are configured for real use, it's the memory outlay that can be the main differentiator of purchase cost between platforms, especially when 16 GB and 32 GB quantities are used. I was surprised at the extra cost that Dell charges for DRAM. Either Xeon MP is picky about memory, or Dell is benefiting from a very healthy mark up.
Memory pricing:
Dell/Einux
4GB - 8x512 - $1,991/$1,192 - Einux saving: $799
8GB - 16x512 - $4,091/$2,384 - Einux saving: $1,707
16GB - 16x1GB - $9,491/$5,584 - Einux saving: $3,907
32GB - 16x2GB - $35,591/$23,984 - Einux saving: $11,607
When the Einux price points are added to the performance advantage that Opteron delivers, Xeon isn't even in the same league, especially when one sees how Opteron scales when server load is increased. Two more benchmarks highlight those attributes:
ExtremeTech's Nile application server benchmark emulates a book ordering transaction processing environment. At 500 users, Xeon took almost twice as long as Opteron to complete the same test. So to be competitive with Opteron, you'd need to be using "two" Xeon servers.
AnandTech had its own database benchmark, which also highlighted the scaling prowess of Opteron. When the data base setting was increased to 20x, Opteron was able to complete 450 more transactions per second. Don't do what I did and calculate the 38% better performance figure. Over one minute, Opteron will process 27,000 more transactions. Over one hour that increases to 1,620,000. Again you'd have to use "two" Xeon servers to better the Opteron result.
It should be noted that these benchmark results are for two- way systems. Knowing how Opteron scales from two-way to four-way, the four-way numbers should surely get better when Xeon MP is in the frame.
The end of Xeon's gravy
train? How will Intel respond?
When Opteron debuted in New York, the aftershocks were soon felt at Intel's HQ in Santa Clara. The chip giant will respond with increased speed grades of Xeon MP at up to 2.8 GHz, which is a 40% increase over the current flagship. Also, the standard up to two-way 3.06 GHz Xeon will have its level 2 cache doubled to 1 MB. These actions indicate that Intel is taking Opteron's challenge very seriously. But these enhancements can only be described as band aids. Xeon is hemorrhaging performance because its northbridge-based architecture is out of date. Opteron's ability to scale has made today's Xeon yesterday's technology. Intel will have to pull the proverbial rabbit out of the hat if it's to rescue Xeon from its predicament, or its server business will suffer erosion as the market votes with its feet.
What has to be most worrying for Intel though is the 240/840 series Opteron, the slowest at 1.4 GHz, which under heavy server load, like the ExtremeTech and AnandTech benchmarks, will outperform or be comparable to the very best Xeons available. In this environment, Xeon performs like it just came out of the stone age, and the price difference is like night and day. How does Intel respond to that level of price/performance? For the time being it can't, and Xeon's performance enhancements won't make a significant impact until the arrival of 90 nm devices.
Future Opteron enhancements
If Intel's 90 nm Netburst products don't pack anything more substantive than more cache and higher frequency, then from where I'm sitting, Opteron's server performance lead could be maintained for at least the next year and through the 90 nm generation. What do I base this on?
Distributors say that AMD has a 2 GHz Opteron part in the works, so it shouldn't be too far off - possibly being reserved for the 100 series launch. More importantly though, they report that the core voltage is 1.45 volts, which is 0.1 volts lower than Opteron parts. At that voltage level, AMD may be able to yield future devices at 2.2, 2.4, and possibly even 2.6 GHz. Let's not forget that Opteron has two extra stages in its pipeline, which should help it to scale higher in frequency than Athlon. Also, with Opteron at 0.2 volts less than Athlon, and the SOI process, I don't think it's a reach to say that these parts should be deliverable on the current process generation.
Even though Opteron's performance has been nothing short of spectacular, I would characterize it as also being raw. Remember how SSE2 turned the P4 from an ugly duckling into a swan? Compiler optimizations that are being done by the Portland Group might change Opteron from being the best x86 server processor to one that looks beyond reach. The beta was released last April, so when it's ready for prime time this month, Intel had better taken out its Xeon and Itanium insurance cover. If the Portland Group delivers, Opteron's aftershocks may just push Santa Clara's enterprise group into the sea. OK! I'm being a little facetious, but you get the point I'm making.
Opteron currently supports up to DDR333, and this piece says that it has no problems with DDR400. So when DDR400 becomes available for the server market, Opteron will no doubt be first out of the blocks with official support. There is also talk that DDR-I will peak at DDR533 before the transition to DDR-II. So after DDR400, Opteron may see two more memory speed bump improvements.
Opteron on 90 nm has to be even more worrying for Intel. AMD should be able to double the level two cache to 2 MB and continue with the 1 MB parts as cheaper entry level devices, which could end up as the natural replacement for Athlon MP. Also, let's not forget Athlon 64, which with little modification, could end up as the bottom dollar server.
Opteron has plenty of areas to scale and develop. Also, the Portland Group compiler optimizations will also improve performance. Let's not forget that there are 32-bit apps that are now hitting the 4 GB memory limit. A 32-bit OS can take up to 2 GB of address space. But using a 64-bit OS, a 32-bit application can gain access to the full 4 GB of memory. So when you take this all together, it's clear to see where Opteron enhancements will be coming from.
It's not just Opteron's performance and its future enhancements that Intel has to worry about; the upgrade flexibility of Opteron is just awesome.
Opteron's upgrade flexibility
Those who buy Opteron platforms will benefit from far superior flexibility when it comes to upgrades. Imagine an IT buyer who needs a two-way platform that supports greater than 4 GB of memory, but his memory outlay restricts his budget to entry level processors. He decides to buy a 240 (1.4 GHz) series Opteron platform with 8 GB of DDR266 DRAM.
A year after the purchase, his workload demands faster memory. He finds out that DDR400 is now available at reasonable cost. His current memory wouldn't be wasted as there is strong demand for it on eBay. After the upgrade, the 50% increase in theoretical memory bandwidth was very noticeable, which would keep up with his workload requirements for at least another year.
When that time came, memory bandwidth was no longer the bottleneck, processing power was. By this time the 250 series Opteron (2.2 GHz) had been around for a while and was very good value. So when he upgraded, the 57% increase in processor speed was also very noticeable. Because enthusiast demand for Opteron product is strong, finding a new home for his superfluous processors would not be a problem.
Another year passed, which would normally mean the replacement of the platform. Unfortunately, there had been a squeeze on the IT budget, which meant the platform change would have to wait for another year. Because the platform required more oomph to see it through its last year, and there was sufficient funding in the budget to meet that need, the go ahead was given to upgrade again. DDR533 was the limit that the platform supported, and the entry level Opterons that supported 2 MB of level two cache were now affordable, so it was decided to upgrade to those parts. Again there was no problem finding new homes for the excess kit.
With this final upgrade, the platform lasted an additional year. The company had been so pleased with the value that they had garnered out of its first generation Opteron, when the decision was made to replace the platform, Opteron2 superseded.
The example given is probably more typical of a workstation environment than a server one. So if we expand this to the server world, the memory footprint could have been expanded to 16 GB. Even though the OS may have started out as 32- bit, the OS and applications could have migrated to 64-bit.
When you factor in the fact that many companies are using servers and desktop PCs that are greater than four years old, when companies plan for those platform replacements, purchases will have to last the long haul. When companies evaluate Opteron and Xeon for this journey, Opteron's credentials would be the only ones that would cost effectively meet that requirement.
Because Itanium and Xeon are hobbled by their northbridge- based architecture, these platforms will never be able offer the upgrade flexibility that's native to Opteron. When this fact dawns on IT decision makers, and they add all of the other AMD64 advantages, Opteron will just be too compelling to ignore.
Opteron: Does it measure up? Did
Hammer live up to the hype?
If you've read this far, then you know that it's an assured yes, but AMD is not yet home and dry. It has delivered on the technology and the benchmarks speak for themselves, but what of Opteron's future and its ability to return on its investment?
If IT decision makers had based their purchases on price/performance alone, then Celeron would never have been the success that it is. This proves that there are other factors which determines the winner of a purchase order. Because the AMD platform has been perceived to be second best, many of its commercial opportunities ended up as Intel wins. But with AMD64, this is about to change.
AMD brings together an evolution of technologies that have been primed for the AMD64 roll out.
So how will Opteron impact the markets it competes in? Its opportunities in the server world should now be plainly obvious, so Iet's discuss its workstation prospects.
Opteron workstation prospects
In terms of processor sales, the workstation market probably accounts for at best two million units a year, which is only 25% of the server market. This market will be harder to crack than the server one, as the top five vendors control 95% of the market. AMD has landed one of these companies, Fujitsu Siemens, to sell such systems in Europe, but its market share was only 3.2% for 3Q02. Of course, if Opteron's performance in workstation benchmarks is anything like what we've already seen, then the white box vendors will shift a lot of this product. If AMD can sell its fair share of Opterons into both markets, then improvement to its ASP should follow.
I see one area of rapid growth for Opteron. That's the dual processor workstation. Many workstations that are in use today have or will soon reach the 4 GB limit. Two-way Opterons can support up to 16 GB of RAM, but not many will justify the $5000 and up outlay that eight 2 GB RAM sticks would cost. Of course, there is an easier path to memory nirvana. Eight 1 GB memory modules can be purchased for less than $2000, so it's obvious to see where memory investments will be made.
When it comes to making full use of that memory, 64-bit Opteron doesn't have to be encumbered by PAE, which it natively supports. Also, Opteron's ability to run 32-bit apps under a 64-bit OS, which can help applications that are memory constrained by freeing up additional RAM, extends their useful lives. If Xeon's flexibility is not improved to be at least the equal of Opteron, then I can't see how long term Xeon investments can be justified.
One-way Opteron workstations will also be entering the market. Where the 4 GB RAM limit is not an issue, applications should realize performance gains from Opteron's low latency design, especially when nVidia's single chip platform processor is used. Because of lower cost, these workstations could also prove to be very popular enthusiast PCs as well. The Asus SKN8 workstation motherboard, which is based on the nVidia nForce3 Pro 150 core logic, is one board that many will want to see perform. If it delivers, it could prove be a serious sales driver for the 100 series Opteron.
Four-way Opteron workstations are also on the cards; now that's some serious big tin to have sitting beside your desk.
Opteron's aftershocks - what does
all of this mean?
It means the AMD Opteron has arrived, and AMD and its partners have changed the server world paradigm for the better. If Intel doesn't get onboard, it'll get left behind. It's no wonder that AMD is promoting Opteron as the Xeon killer, because in truth, that's what it is. In the enterprise space, which ultimately means moving to 64-bit, Xeon is blown clean out of the water. AMD is blazing broadside with 16 inch guns. Intel can only respond with guns that are half the size. If I had been an Intel exec on the launch day of Opteron, I would still be suffering from shock and awe. When we're looking at two- day ports that can yield almost 10:1 performance improvements, simultaneous 32-bit and 64-bit computing, performance and price leadership, and all at the commodity price level of 32-bit computing, then Opteron is the only serious technology contender in town. When you contrast this to Intel's He ath Robinson approach to simultaneous 32-bit and 64-bit computing - Itanium and Xeon racked together, then you know where the smart money will be going.
With this year's projected Itanium output to be less than 23,000 processors (page four of PDF), and AMD64 technology planned to be in the millions, and next year in the tens of millions, then Itanium's continued traction and market acceptance problems are likely to deteriorate exponentially. With the latest quarterly sales figures from IDC, which shows Itanium server sales didn't exceed two thousand units, it just compounds what I've already said. The bookies will have already closed the book on Itanium, as they can't afford to take any more money that says Itanium is going to fail. It's like backing the old nag that's running in the Grand National, you just know it ain't gonna win.
It's interesting to note that a month after Opteron's launch, Intel launched its "real server" campaign. This says only real servers use Xeon, which is an obvious slight on Opteron. When IT decision makers compare Xeon to Opteron, there will be no doubt which platform merits the "real server" label. If Itanium is really the 64-bit future, why is Intel promoting the processor technology of the past? Intel owns 90% of the x86 server market, so why launch this campaign at this time? The simple reason is Opteron. It has frightened the living daylights out of Intel, so it had to respond. But there is more.
Intel has said that many companies are using servers and desktops that are over four years old. When you look beyond what Intel has said, when these companies upgrade their aging infrastructure, IT decision makers are likely to stretch the longevity of their new platforms from over four years towards even ten. When you contemplate that possibility, who in their right mind is going to buy Xeon, with its built-in 32-bit obsolescence, when Opteron doesn't suffer that limitation? This is why Intel's server roadmap is stuck in no man's land. Those who wish to remain Intel customers have very stark choices. Xeon or Itanium, or a mix of both. For many, Itanium just doesn't cut the ice, and for future investment protection, 32-bit Xeon will be out-of- date. So this leaves us with Opteron, which provides the best 32-bit server performance today, and a 64-bit upgrade path for the future. This is what frightens Intel the most because Xeon doesn't measure up to the long haul, and Itanium is too much of an unquantifiable risk, which leaves Opteron to clean up the middle ground.
Let's add some historical perspective to this as well. Microsoft has estimated that 35% of Windows' customers still use NT4, which is now 7 years old. That tells you the mentality of the enterprise IT buyer. They buy for the long haul. Microsoft plainly understands this and the consequent effect on its revenues, which is why it wants to move its business customer licensing model to one that is subscriber-based. If business customers choose that model, it won't matter to Microsoft when they upgrade their software, as its revenue stream will be based on a recurring fee, not the outright purchase of software.
At Intel's spring analyst conference in New York, COO Paul Otellini discussed the challenge of getting companies to replace their aging infrastructure. He shared a 1933 quote by Charles F Kettering, founder of Delco and then Director of R&D for GM: "I believe business will come back when we get some products that people will want to buy." Great quote, but where are the server products? Centrino is making great inroads in the notebook space, Xscale is about to go exponential, and the P4 is still bringing in hearty returns on its investment. But when one looks at Xeon and Itanium, the enterprise world is holding its breath. These aren't the products that the server world really wants to buy. They're looking for something better, and Opteron might just prove to be the deal.
AMD has estimated that this year's worldwide server processor sales will be around eight million units (page 11 of PDF). If we use a $250 ASP, that's a $2 billion market, at $500, it doubles to $4 billion. AMD suffered a total loss of $1.3 billion on sales of $2.7 billion for 2002, and with its market share of server processors at less than 10%, it should be clear to see where the biggest impact on revenue growth can come from. AMD's quickest return to profitability should come from server market penetration. It now has the technology to make that happen. All that it requires is a market that is willing to buy. If Europe, Japan, and the US are about to begin the upgrade cycle of infrastructure that was purchased before Y2K and the dot.com bust, then Opteron's arrival could not have happened at a better time. With the roll-out of the rest of the AMD64 platform still to come, AMD will be very well positioned to reap from its emerging technologies.
One thing we should all know about Intel - it's not stupid. If Itanium has to be folded, Intel will fall back to plan B. If Hans de Vries' Prescott analysis is on the ball, then Yamhill, the 64-bit extensions that makes the Netburst architecture 32-bit and 64-bit compliant, is already in place, so events will determine Yamhill's future. In the race to the 64-bit crown, Intel won't be averse to changing its horse mid-course, especially if it breaks a leg, which means Itanium could be history before it ever comes of age.
With AMD64, AMD's solution for 32-bit and 64-bit computing covers both desktop and server applications. As 64-bit applications become available for AMD64, and 32-bit applications gain the 32-bit performance boost that a 64-bit OS can deliver, if Yamhill is still a mirage, then Intel's IA-32 architecture will become like Xeon, the dinosaur of the industry.
In all of this discussion, no one from the biggest player of all has been quoted. Microsoft, the software behemoth, will support AMD64 with a version of Windows Server 2003 that's been optimized for Opteron; application software will follow. Brian Valentine, Microsoft senior VP Windows division, said at the Opteron launch, "So, 64-bit computing, we think it's the wave of the future." He went on to say, "And it isn't just about scientific computing. It isn't just about database computing. It's about making a Windows server and a Windows desktop run in 64-bit in every workload that any customer might want to run it in. That's what it's all about. And 64-bit is the wave of how we get to that pervasive high-end computing model today, but also will become pervasive in just about every workload in the future as we move from 32-bit to 64-bit in the world. And the Opteron processor is a great step to help us get there." He also said, "We're seeing some great compatibility in running 32-bit applications on top of the 64-bit platform. So we're also optimizing not only 64-bit applications, but we're also optimizing Windows to run 32-bit applications even better than what you can perhaps see in certain memory constrained or processor constrained scenarios on Windows 32. So we'll make 64-bit not only great for 64-bit applications, but we'll make it run 32-bit applications in a great way. I'm not just talking about server applications, I'm talking about desktop applications as well."
AMD has secured IBM as its only tier one vendor that will sell Opteron-based systems. One quarter has described IBM's support as "lukewarm", but one cannot detract against its significance, as it brings carrier grade credibility to Opteron. We should not forget that 29% of all servers shipped in 2002 were from white box vendors, which is just 1.1% less than market leader HP. Even with IBM as the only tier one vendor, it still leaves a large part of the server market open for Opteron's talents.
Intel likes to talk about its "four pillars". These are its R&D investments, its manufacturing capacity, its brand momentum, and Intel capital. Of course, AMD's efforts in these areas would look minuscule by comparison. But if we look at AMD's four "virtual pillars" with respect to Opteron in the server space, then AMD's prospects look very promising indeed. These would be:
1. Customer-centric innovation - one platform for all server needs.
2. Investment protection - best 32-bit performance today, 64- bit at customer's choosing.
3. Tomorrow's technology today - no Heath Robinson computing.
4. Unmatched price/performance - universal appeal, and all at today's 32-bit commodity cost.
With all of these qualities rolled into one platform, Xeon has become the dinosaur of the IT world. Because of incompetence, if Itanium ever had any widespread relevance, it's now passed its birthright to Opteron. When one looks at Xeon and Itanium together, which is Intel's solution to simultaneous 32-bit and 64-bit computing, you have to ask the question - is Intel really serious? Is this the IT joke of the decade? Or will someone put their hands up and say, "OK, we got it wrong"? We can't answer those questions today, but they'll have to be responded to soon.
Because of Opteron's significance, IDC will reference x86-64 data in its next quarterly report. Now that will be one set of numbers that many will be looking forward to seeing.
AMD has the last word
At a recent semiconductor conference, Robert Rivet, AMD SVP and CFO, made an enlightening comment about the "multiple prong effect" that moving into the server space could have. He said, "You know, maybe naively we thought we could work into the enterprise space from the bottom up, but the reality is decisions are made from the top down. Decisions made are based on what servers I have and what desktops and mobil I purchase, versus trying to get a desktop into the market place. So clearly there is a potential wave effect here that will take place as we get servers locked into more companies. We will see continued penetration in the enterprise space, which is our single digit market share area."
At the Opteron launch, AMD VP Marty Seyer discussed the challenges and needs of today's IT decision makers: "Today's technology decision makers, whether enterprise CIOs or small business leaders, have to make very tough investment decisions in one of the most hostile business environments in recent history. If you listen to them, they will tell you that they need a few simple things. For example, help me use IT to make a difference in my business. Let me do more with less. Less cost, less disruption. Why do I have to pay a premium for the next generation of performance? What they're saying is help me to get the most out of today's technology and prepare me for the future. In short, they say, we need your help simplifying our business."
Seyer went on to address how Opteron would address the IT needs of business: "With today's launch, AMD makes one simple promise. We promise the AMD Opteron will simplify business. Simplify business by removing all the barriers to 64- bit computing, by protecting the huge investments that IT departments have made in their hardware, their software, and their IT staff, and by being flexible. The AMD Opteron processor is exactly what our customers are asking for and it's what we're here to deliver. A new class of computing for business that is compatible, reliable, and secure."
AMD has talked a lot about customer-centric innovation - giving the customer what they want, not what they don't need. Opteron truly delivers on that objective, but Hector Ruiz, AMD president and CEO, gave the full flavor of what AMD64 is all about: "With the launch of the AMD Opteron and the AMD Athlon 64 processors, we have the ability to revolutionize computing - from high-end servers, to high-performance desktop systems, to small form factor devices. We have the potential to change the very landscape of computing based on an architecture that represents the essence of customer- centric innovation." Well, AMD has begun the revolutionary change with Opteron, and I believe Opteron will change the server world landscape for the better.
When AMD VP Marty Seyer discussed Opteron and Itanium at the Opteron launch, he asked, "Why will the AMD Opteron succeed? Simply put, we did it right." He also said, "With regard to the 64-bit competition, watch out, even the paranoid may not survive."
It seems only right to close with a quote from Opteron's greatest exponent - Jerry Sanders. Speaking at a strategic decisions conference in June last year, he said, "Hammer changes the game! Opteron is simply the best!" With the facts that are before us today, few would argue with that statement. µ
Related stories by the same author
Nforce and UDA deliver dollar saving multi-PC image
Is AMD ready to impact the enterprise space
Part 1
Part 2
Part 3
Related articles
So what's your oldest piece of kit?
When Is 64 Not Two Times 32?
AMD's Opteron Steps Out. Will the Market Follow?
Workstation 'renaissance' predicted
Sun forced to use Intel chips because AMD Opteron too good
Opteron reviews
Ace's Hardware
AnandTech
Part 1
Part 2
Part 3
Part 4
ExtremeTech
InfoWorld
THG
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.