Posts tagged with Performance
AMD Opteron Processor – Delivering True Value, Not Just Benchmarks
Posted by John Fruehe in 4:39 pm
(10 votes, average: 4.80 out of 5)
Loading ...I left the office yesterday at 5:00, headed to the trail for my weekly ride. Because of the recent rain, we shifted to City Park, one of the most technical mountain bike trails in Austin. If you just looked at us and our bikes in the parking lot, you’d notice that the older guy with the borrowed bike (because I finally broke my frame after 4 years) would probably not fare well against better riders, younger riders or newer bikes.
But looks can be deceiving. While I didn’t have the flashy skills or the newest ride, I did have a dedication and a lack of technical problems that actually allowed me to finish the ride ahead of others. Sometimes the outcome isn’t always as obvious.
Too many people get caught up in the benchmarks and think that is the end of the story. Benchmarks can be important as an indicator, but they don’t tell the whole story. As a matter of fact, they really only tell a portion of PART of the story. In my opinion, making a decision based on benchmark alone is like buying a car because you like the color. That is great if all you care about is a red car, but so few of us really think that way in the real world.
One of the things that I have been telling customers is that they need to step out of the “socket” world and start thinking more about the work that needs to get done when assessing the platforms. Sometimes the answer is not obvious.
The non-obvious thread for today started this morning when I went to check out the latest review on AnandTech.com of the AMD Opteron processor vs. our competitor’s processor (which my colleague Margaret Lewis also discusses here). We sent them a review system recently and during their testing, they came up with some interesting comments.
Comparing a 2 socket Intel server to a 4 socket AMD server, you start to see that if you step outside of the raw performance area, there are plenty of other places to be looking, like threading and efficiency, not just clock speed. Here is what AnandTech had to say when comparing our six-core 2435 to a higher clock speed quad core:
“If your application scales well, two 2.6GHz Opteron 2435 will offer 15% better (and sometimes more) performance than a 2.9GHz Opteron 2389 with the same power consumption.”
More performance, at the same power envelope and lower clock speed. Not always the most obvious answer, but clearly a great choice for threaded applications. (And don’t think that applications will be less threaded in the future.) Memory is another area less than obvious. Here is what AnandTech had to say:
“Using relatively ‘old’ technology such as DDR2, the hex-core Opteron based servers are very affordable, especially if you compare them with similar Xeon servers.”
Yeah, I was the old guy on the trail, but that didn’t mean I was the last. I can think of a couple of younger guys finishing after me – so sometimes “newer” doesn’t me “better”. Again, not the obvious answer, but until the prices of DDR-3 come in line with DDR-2, this is our secret weapon. And in a tough economy, who doesn’t want to be smart with their IT budgets.
We have tremendous consistency, but people expect that. And they expect that all processors in the family will have the same features. That is obvious, right? Here’s how AnandTech saw it.
“If you chose the Xeon platform, you should be aware of the fact that Intel’s low end is much less interesting: the best Xeon 55xx CPUs have a clock speed between 2.26 and 2.93GHz. The low end models, the 5504 and 5506 are pretty crippled, with no Hyper-Threading, no Turbo Boost, and only half as much L3 cache (4MB). These crippled CPUs can keep up with the quad-core Opterons at about 2.5GHz, but they are the worst Xeons when you look at idle and full load power. The performance per Watt of the Xeon EE550x is pretty bad compared to the more expensive parts.”
You’ve heard me talk a lot about “no compromises” in our products, and this is what we are talking about. Just because you can’t afford to buy the most expensive processor in the stack shouldn’t mean that you have to compromise so much. Cutting the cache in half? That is not obvious to the typical customer.
When you step back and take a look at all of this in full, you see what was obvious before – 2P servers are the best value for most workloads – is becoming somewhat less obvious. And as we get into 2010, the value that AMD will bring to market will blur the lines even more. It’s time to stop looking at the world of sockets and start looking at the workload, the power consumption and the cost. Only then, does everything become obvious.
Otherwise you’ll just be like that young guy tonight, standing on the trail as the old guy blew past him. Sometimes the obvious answer isn’t always the only answer.
John Fruehe is the Director of Business Development for Server/Workstation products at AMD. His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.
Is a Cloud a Cluster or is a Cluster a Cloud?
Posted by Guest Blogger in 10:55 pm
With the recent publication of several articles about using cloud computing concepts for High Performance Computing applications (here and here), I’ve been wondering about the applications for high-performance, low-power processors. There’s been quite a bit written here already about cloud computing and energy efficiency, but what about applications where high performance and energy efficiency are both important?
At this time last year, the most energy efficient AMD OpteronTM processor-based server (based on the SPECpowerTM_ssj benchmark) could achieve only a score of 203 overall ssj_ops/watt (95,853 ssj_ops & 276W @ 100% target load) and consumed 164W at Active Idle. A server using the newest Six-Core AMD Opteron HE processor achieved a score of 1228 overall ssj_ops/watt (419,277 ssj_ops & 221W @ 100% target load) and consumed only 120W at Active Idle1. That’s more than 6x the performance-per-watt AND more than a 25% drop in Active Idle power.
AMD technology-based servers help increase performance-per-watt and decrease power consumption at the same time by using a suite of features we call AMD-P. AMD-P is supported by the Six-Core AMD Opteron 2400 and 8400 Series processors as well as the Quad-Core AMD Opteron 2300 and 8300 Series processors. This suite of features and the large number or processors that support them enable customers to build energy efficient two-socket, four-socket, and eight-socket servers which can efficiently meet the needs of almost any server application.
When we compare servers using the newest Six-Core AMD Opteron 2400 Series HE processors to servers using existing AMD Opteron processors, we find that a server based on the AMD Opteron 2400 Series HE processor is able to achieve 18% higher performance-per-watt than a server using Quad-Core AMD Opteron 2300 Series HE processors2 and also consumes 18% lower platform-level power than a server using Six-Core AMD Opteron 2400 Series processors3.
That’s a pretty big improvement over a processor that was released just six months ago!
In addition to lowering server Active Idle power and boosting server performance-per-watt, these new Six-Core AMD Opteron HE processors are designed to provide significantly more processing performance than prior low-power AMD Opteron processors. Servers using these processors are able to achieve up to 50% higher performance than servers using Quad-Core AMD Opteron 2300 Series HE and 8300 Series HE processors in the same power and thermal envelope. That’s like getting the performance of V6-powered Ford Mustang and the fuel efficiency of a four-cylinder Ford Fusion in the same car.
Whether they’re being used in a cloud cluster or a High Performance Computing cluster, the newest AMD Opteron HE processors provide plenty of performance for only a few watts.
What do you think – is a cloud a cluster or is a cluster a cloud?
Andy Parma is a Product Marketing Manager for Server/Workstation products at AMD. His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.
1Configuration Information: 2 x Six-Core AMD Opteron™ processors (“Istanbul”) Model 2425 HE in Supermicro 1021M-UR+ server, 16GB (4×4GB DDR2-800) memory, 500GB SATA disk drive, Coldwatt CWA2-0650-10-SM01 power supply, Microsoft® Windows Server® 2008 Enterprise Edition SP1 64-bit
2Six-Core AMD Opteron™ processor Model 2425 HE [SPECpower_ssj™2008 1228 overall ssj_ops/watt, 419,277 ssj_ops, 221W @ 100% target load] compared to Quad-Core AMD Opteron™ processor Model 2376 HE [SPECpower_ssj™2008 1044 overall ssj_ops/watt, 346,326 ssj_ops, 210W @ 100% target load]
3Six-Core AMD Opteron™ processor Model 2425 HE [SPECpower_ssj™2008 1228 overall ssj_ops/watt, 419,277 ssj_ops, 221W @ 100% target load] compared to Six-Core AMD Opteron™ processor Model 2435 [SPECpower_ssj™2008 1228 overall ssj_ops/watt, 487, 764 ssj_ops, 270W @ 100% target load]
SPEC and the benchmark name SPECpower_ssj are trademarks of the Standard Performance Evaluation Corporation. For the latest SPECpower_ssj2008 benchmark results, visit http://www.spec.org/power_ssj2008.
Quad-Core AMD Opteron™ Processor Codenamed “Suzuka”
Posted by John Fruehe in 6:30 am
Brad Pitt’s brother isn’t the only one to live in the shadows of a more famous sibling. The latest AMD OpteronTM 1000 Series processor, codenamed “Suzuka”, was launched in the shadow of its 6-core bigger brother, the Six-Core AMD Opteron processor codenamed “Istanbul.”
The AMD Opteron 1000 Series processor is designed for applications that are driven by cost or power concerns more than scalability. In the past, this meant a single core in a single socket, but in today’s multi-core world, this means four high performance cores in a single socket.
Typically, these processors are used in web servers, small business servers, workstations and even cloud computing. The flexibility of four cores and a low cost infrastructure gives customers an edge when designing for a cost-effective or power efficient platform.
With speeds of 2.5GHz, 2.7GHz and 2.9GHz, these single-socket processors pack a punch. It’s based on the same core as the Quad-Core AMD Opteron processor codenamed “Shanghai,” so all the great features that you find in “Shanghai” are also in “Suzuka.”
One of the best parts about my job is that I have access to all the processors that I could ever want for testing. My server at home migrated from a dual-core AMD Opteron Model 185 processor to a new system board with a quad-core “Budapest,” which is a 2.3GHz processor based on the same core as “Barcelona.”
Recently, I upgraded that server to a pre-production “Suzuka.” I went from a 2.3GHz quad-core with a 2MB cache to a 2.9GHz quad-core processor with a 6MB cache. I saw an increase in performance, but amazingly, because Suzuka is on AMD’s 45nm process, I actually saw a decrease in total server power draw of about 10 watts. A performance increase, a big drop in power consumption – that makes the performance per watt story even better.
All I needed was a BIOS flash – the processor just dropped right into the same AM2 socket. The entire process was 5 minutes, and that included digging around under the sink to find some rubbing alcohol to clean the thermal paste off of the old processor. This incidentally, was the same process that you would have to go through in updating from a Rev F dual core processor to a “Budapest.”
Many may not know that our code names are derived from Formula 1 race tracks. Suzuka is a race course in Japan; recently it had been renovated and was re-opened in April of this year, just about the same time that we were finishing up the final touches and releasing our own “Suzuka.”
One of the biggest races at Suzuka is the 1000Km endurance race. A 1000 kilometer race? Sounds like the perfect compliment for an AMD Opteron 1000 Series processor. Both are designed for the long haul.
John Fruehe is the Director of Business Development for Server/Workstation products at AMD. His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.
Building Blocks
Posted by John Fruehe in 7:45 am
How many times have we heard that a pocket calculator today has the computing power that put a man on the moon in 1969? I can remember my father, who was an engineer, showing me how to use a slide rule when I was growing up (sorry dad, I cheat these days and use a computer.)
Not that long ago, supercomputers were large room-sized behemoths that could require hundreds of millions of dollars of investment and could crack complicated problems. Then everything changed. I blame Linux, but you can choose your own hero (or villain). Suddenly the world of supercomputing went, almost overnight, from a very expensive proprietary and exclusionary world, to an open environment where people use industry-standard hardware and open source software components to construct massive supercomputers at a fraction of their previous cost.
With these supercomputers, companies, universities or governments can take a large problem, like where to drill that hole in the ground to find oil, break it up into thousands of tasks, disperse them across all the computing nodes and then compile the answer. When it can cost up to $1M US to put that hole in the ground, a supercomputer is money well spent.
I am in Tokyo, on vacation this week, but I remember a very vivid meeting here back in the early 90’s at a Japanese auto manufacturer. They were trying to figure out how to drive down the cost of crash simulation. Apparently it is a lot cheaper if you don’t have to build a car and then drive it into a wall. Today crash simulation is primarily done with computers. You can crash more cars in a morning with an HPC cluster than in a year’s worth of playing bumper cars on the Dan Ryan Expressway. Those of you from Chicago know why I picked the Ryan – it was notorious for accidents.
It is currently the rainy season in Japan, and every time I turn on the TV to see how wet we will get today, I am reminded about the accuracy of weather forecasts (insert your own joke here), another area where HPC clusters and supercomputing technology are having major impact.
The ability to lash hundreds or even thousands of low-cost x86 servers together into a supercomputer is presenting some pretty amazing results. In the most recent TOP500 Supercomputers, AMD continues to be prominently featured as a groundbreaking leader.
With the top two overall supercomputers on the www.top500.org list based on AMD technology, and 9 out of the top 20, it is clear that customers are very interested in AMD Opteron™ processors for building high performance supercomputers and for good reason.
The chief concerns for most supercomputer customers these days, believe it or not, are generally not raw performance numbers. When you are putting thousands of processors together, a few percentage points here or there become meaningless. The factors that do drive a lot of the decisions are price, power consumption and scalability.
Think about the task. You are building out thousands of servers, each with multiple processors. For every dollar that you save per processor, you might be saving tens of thousands of dollars in total cost. Many supercomputing sites are frankly operating in tough budget constraints, especially when it’s an academic institution, for example. And power can’t be overlooked. The density of these deployments, along with the networking, can consume huge amounts of power. Scalability is a given, with the large number of pieces that you are breaking a problem into in order to solve.
What makes AMD Opteron processors perfect for supercomputing? Well we excel in these three areas: power, price and scalability. And, only AMD can give you the same 6-core processor architecture in 2P, 4P and 8P configurations, helping you achieve greater scalability. And too, when you want to talk about HPC performance, you can’t ignore that throughput and memory performance are key. Those are also areas where Direct Connect Architecture has and continues to excel.
That is why you see us all over the TOP500 list. And with our 6-core “Istanbul” product now in the market, who knows what November’s list will look like?
John Fruehe is the Director of Business Development for Server/Workstation products at AMD. His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.
That Five-letter Dirty Word
Posted by John Fruehe in 9:24 am
Value. There, I said it. Someone once threatened to wash my PowerPoint slides out with soap for using it.
But customers care about it. Today, more than ever; but even in the crazy dot-com 90’s, people still gravitated to value. Who doesn’t want to get the best value for their IT investments? In the business world, people just don’t burn money; it never makes sense.
On the consumer side, people sometimes associate “value” with “cheap.” But that might be because consumers tend to buy one computer at a time. They research the purchase, they obsess, they compare. Nobody wants to go “cheap” because it is their one shot for the next few years.
However, in the enterprise world, server purchases are a constant fact of life, not a single point in time. Most companies, even in today’s environment, continue to deploy servers and look for the best value that they can get, usually with a price/performance or performance/watt metric. Rarely is the decision made just on raw performance.
I have been a critic of performance benchmarks as much as I have been a fan of them. One of the problems that we see in so many benchmarks is that they fail to comprehend the value of the solution – they only measure one vector. Typically that is raw performance.
Take VMmark for instance. If you take the results at face value, it shows you the approximate performance of different systems. But, it is showing the performance for a specific configuration. Perhaps a configuration that you may never actually deploy in real life.
We recently introduced our new Six-Core AMD Opteron™ processors, and along with those processors there were several new benchmark results introduced, including a VMmark benchmark.
Just looking at the raw benchmarks, one might conclude that a 2P Nehalem-based system is going to be a better choice because of the higher performance. However, Collin MacMillan points out in his Solution Oriented Blog that if you look at only one vector, raw performance, then you might miss the big picture. The reason is that, depending on configuration, the Six-Core AMD Opteron processor-based HP DL385 can be priced almost 2/3 less than the Nehalem-based HP DL380 server.
Just over one third of the cost. That is just stunning. Especially when you consider that the typical customer may be loading 5-10 virtual machines on a single 2P server.
Maybe I am going out on a limb here, but if I was trying to justify a virtualization project, I think that telling the CIO to replace 5-10 physical servers with a single server that costs nearly 3X as much as a competing product is simply a losing proposition.
My colleague, Margaret Lewis, takes a closer look at configurations and pricing of Six-Core AMD Opteron processor-based systems (“Istanbul”) and Quad-Core Intel Xeon processor-based systems (“Gainestown”) that have posted top VMmark scores.
In the enterprise world, value might not have the same stigma that it has in the consumer world, and that is why enterprise customers don’t seem to be afraid of it.
When you are lining expensive data centers with row after row of computing devices, someone is bound to ask you what all this is costing and where is the benefit. If you are buying servers based on our new Six-Core AMD Opteron processors you can rest assured that you are filling your data centers with a superior value – and that speaks volumes.
John Fruehe is the Director of Business Development for Server/Workstation products at AMD. His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.
Six-Core AMD Opteron™ processor codenamed “Istanbul” – It’s Finally Here
Posted by John Fruehe in 6:28 pm
It’s odd to think that I would be using “finally” to talk about our Six-Core AMD Opteron™ processor codenamed “Istanbul.” We were supposed to be launching in October, but the health of the silicon and the fact that the initial version was deemed production ready by our industry partners meant we are looking at “Istanbul” in June instead.
We started shipping production units in May and today we are rolling out the red carpet to introduce “Istanbul” to the world. Over the next few weeks you’ll see our industry partners launching their “Istanbul”-based platforms.
So where do customers want to use these two extra cores? They want them in their databases, in their virtual environments and in their HPC/technical applications. Six cores will allow an application to break up problems into smaller chunks for a more parallelized completion of tasks.
We have been rapidly moving from a world of serial tasks to a world of parallel tasks. And as this happens, having more cores helps applications run more efficiently. Elegant design has won out over brute force in software programming. Not too long ago a single core processor delivered all of the performance that a customer needed, but now, as operating systems and applications have become more threaded, the need for more cores is becoming far more important.
So, this is the death of the Quad-Core AMD Opteron processor codenamed “Shanghai,” right? Hardly.
There are still plenty of applications that are less threaded and will enjoy the higher clock speed of our four core processors. The two products can live side by side in the product line, allowing customers to meet a wider range of application needs, all with a common platform underneath. Best of all, we offer both the four-core and the six-core for a range of platforms, from 2P to 4P and even 8P. No other company on earth offers the same commonality across that range of configurations. Only AMD delivers consistent four-core and six-core configurations across all of the different form factors, from the low-end 2P 8-core systems to the high end 8P 48-core systems. There is choice and flexibility across a highly scalable family of platforms from the leading server vendors.
With “Istanbul” we take the first important step towards the “high-low strategy” that we laid out in April. “Istanbul” meets the needs for high performance/highly scalable applications that crave more cores and more parallel processing. “Shanghai” fills the bill for those applications that need fewer cores and are focused more on energy efficiency or cost-effectiveness.
As the market continues to move in these two directions, we plan to be there with the product that customers need to handle their critical applications.
Now that Istanbul is “finally” out the door my wife asked me if things would quiet down and could I relax a little more. How do I break it to her that things are just getting started?
John Fruehe is the Director of Business Development for Server/Workstation products at AMD. His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.
Playing “Hi-Lo”
Posted by John Fruehe in 12:58 pm
You’d think being here in Austin, we’d be playing Texas Hold ‘Em, but with our recent announcement, it’s clear that Hi-Lo is the winning game.
Take a look at the server market today; 70% of the volume is in the dual socket space[i]. And it’s growing. Willie Sutton was once quoted (incorrectly) with saying that he robbed banks because “that’s where the money is at.” So why are we driving so hard in the two socket space? Because that’s where the volume is at. And you can quote me, correctly, on that one.
With our recent server roadmap update announcement, we are introducing the world to a new strategy for addressing the 2P market – G34 platforms on the high end, and C32 on the low end. Hi-low. It’s a big enough market that you would have to be crazy to think that it can’t be addressed by two different platforms. Two different platforms that are more the same than fraternal twins, but we’ll get to that in a minute.
Think for a moment about where the 2P market is going. In the past few years we have been adding more cores, more expandability and more RAS features to our products – because customers have been demanding it for virtualization and other resource-hungry applications. But at the same time, a new class of application need is being driven by the low end. Cloud computing, the growth of SMB applications and sprawling network infrastructure are creating a demand for lower power and lower priced, reliable servers. The same density and business needs that drove AMD to add the AMD Opteron™ EE processors to our product line are driving a new platform that we expect will give you everything you need for most applications, but at a low cost, with very low power consumption.
G34 platforms, under the platform name of “Maranello”, are designed for expandability and performance. We plan to offer four memory channels per socket and 8 or 12 core processors perfectly suited to handle the most demanding 2P environments. As an added bonus, these processors and platforms are expected to be able to scale easily to 4P. No longer will you have different models to choose from (2000 or 8000), we expect the new G34 processors to be able to easily handle both markets, simplifying the lives of those building servers as well as those maintaining servers.
C32 platforms, which we plan to bring to market under the platform name “San Marino”, are designed for the needs of the other half of the stack. Scaling 1P and 2P solutions, the two memory channels and 4 or 6 cores are planned to perfectly match the hundreds of thousands of applications where 12 cores and 4 memory channels may be overkill. Especially when it comes to electricity. These processors sip electricity, they don’t gulp it. They deliver the right level of performance for small/medium businesses, for cloud environments, and for network infrastructure needs. By utilizing a simple design, with fewer memory channels, AMD expects to deliver a relatively lower infrastructure cost. And with planned power envelopes below the total range of current AMD Opteron processors, we expect that fewer fans, smaller heatsinks and smaller power supplies can help you achieve greater energy efficiency, lower noise, and of course lower cost.
So what about that twins comment? What do these processors have in common? Plenty. They are expected to utilize the same core. The same chipset design. The same BIOS base. We expect the C32 designs to be able to span 1P to 2P and G34 to span 2P to 4P. They overlap is at that “meat of the market” intersection where the highest volumes live, for plenty of coverage in all directions. We expect that as a system designer, it will be a trivial feat to take a C32 design and change it to G34. More commonality across a vendor’s products is a good thing. As a customer this can help you drive down your long term cost of management.
Of course the question “so, you’re abandoning the 1P and 4P space” naturally comes up. Nothing is further from the truth. The C32 is expected to allow for a much more flexible 1P. Low cost and simplicity to meet the needs of the 1P market, but with scale up capabilities to take it to 2P. Think of it as 1P on steroids (the good kind, not the baseball kind).
And 4P? What can help turn the tide in a market that is slowly being encroached by 2P? How about a platform that combines the best of both and allows a company to deliver a low-cost entry-level 4P server? Talk about breathing new life into a market that sorely needs to adjust to the new realities of business in the post Y2K work. The time is right for this hybrid approach.
We’re very excited about how we plan to help our technology partners bring these new designs to market. Without naming names, let’s just say that there is a renewed enthusiasm amongst our industry partners who have made comments about this truly being “game changing.” One even likened this strategy to the launch of the original AMD Opteron processor, an event arguably that did more to change the dynamics of the x86 server market than anything else.
We’ve laid our cards on the table now, and it it’s clear that in 2010, we expect AMD to be holding the winning hand. Let the games begin.
John Fruehe is the Director of Business Development for Server/Workstation products at AMD. His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.
[i] IDC Q4 2008 Worldwide Quarterly Server Tracker, February 2009
Simply Spectacular Virtualization
Posted by Margaret Lewis in 8:25 pm
Seems like the industry wants to treat virtualization like a “high performance computing” workload. There is a growing obsession with hardware vendors (including AMD) to tout top VMmark benchmark scores. The truth is any analysis of virtualization performance needs to consider more than just “raw performance.” So let’s go “beyond the score” and take a closer look at the systems posting some of the top VMmark scores.
First, a short bit on VMmark . VMmark is a consolidation benchmark that generates an aggregate score of individual VMs for a given number of tiles. A tile is six VMs running common load-generation tools that represent typical workloads: web server, file server, mail server, database, java server as well as an idle VM. One client computer is used to generate the load for one tile.
Looking at top VMmark scores you find systems that can run over 100 VMs per server. And there is a lot of chatter about how a 2 Socket Intel Xeon “Gainestown” processor-based server can run 16 tiles (or 96 VMs). However, there is no reference to the cost of the systems posting scores. VMmark documents the system configuration for the benchmark so you can take a stab at pricing these configurations on-line at the hardware vendor sites. (See slides 3 and 4 of presentation) In doing so we found that some of the top VMmark Intel Xeon “Gainestown” processor-based server configurations price out at about 175% and in some cases even higher than the top performing AMD Opteron™ processor (“Shanghai”) configurations (based on April 16, 2009 prices). Even in the performance-oriented high performance computing world this would turn heads. Cost does matter.
Going a step further, we now have the information to evaluate the price/performance of some of these systems by taking the estimated system cost and dividing it by number of VMs achieved during the VMmark run. While large number of VMs might be impressive – most IT professionals in today’s economy are focused on balancing performance and price—looking at the cost per VM helps to better understand the cost of putting the system in action. What you find is the system with the top VMmark score is not the system that gives you the best cost per VM.
Now that we have looked at the VMmark systems configurations, what type of virtualization configurations are customers really running? When looking at customer case studies posted on hardware and software vendors’ sites we find servers configurations ranging from 16GB to 64G of memory as more of the norm. We also don’t find many data centers pushing 100 VMs on a system. Responses to SearchDataCenter.com’s 2008 Purchasing Intentions Survey reveals that only 5% of respondents are running more than 25 VMs on a server – 61% are running less than 10 VMs per server and 33% are running 10 to 25 VMs per server. And since many customers are implementing virtualization as a cost saving strategy – we don’t see many of these customers buying the top bin “performance” processor models, which by design tend to consume the most power.
We did the pricing on systems configurations using energy efficient processors and more typical memory configurations for virtualization (again based on April 16, 2009 pricing), comparing both system pricing and cost per VMs (see slide 5 of presentation). Take a look for yourself. We think you will agree considering performance and price can give you a better view of its overall value.
So the question remains: how do *you* define “simply spectacular” virtualization? Is it in terms of raw performance or is it price/performance? Hopefully after reading this post, you have a different answer than when you started.
Margaret Lewis (@margaretjlewis) is a Product Marketing Director at AMD. Her postings are her own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.
“Istanbul” – Right Around the Corner
Posted by John Fruehe in 1:34 pm
I am sitting in Berlin, Germany right now, and Istanbul is right around the corner, geographically speaking.
But literally speaking, the processor is right around the corner as well, because this week AMD announced that “Istanbul,” the code name for our upcoming 6-core AMD Opteron™ processor, is planned for launch in Q2.
How are we able to pull off such a feat? Well, to begin with, Istanbul is based on the highly successful “Shanghai” design. With a highly leveraged core design and a very well-behaved 45nm process, making the leap from 4-core to 6-core was a snap. In addition to the two extra cores added, we also plan to include a new feature called HT Assist, something that I have blogged about in the past.
HT Assist is expected to provide significant memory and I/O performance increases by reducing the overhead of cache lookups. Think rifle shot instead of shotgun, cutting down on a lot of the inner-chip communications.
So starting with a robust design is half of the challenge, the other half is executing well. And we have been executing quite well. Shanghai was delivered to market ahead of schedule, and now Istanbul is expected to follow in its footsteps. Normally, when you design any new processor, you allow for several revisions of the silicon before you get to the final production silicon. When designing in a computer, everything looks fine, but until you start hammering away on actual silicon, you don’t really know how things are going to turn out. Chemistry, physics and math can be funny that way.
But if you do a great job on the design, and you nail it the first time, you can shave months, or quarters, off of the project schedule. Our current schedules are a testament to the abilities of our engineering teams around the world. Our global design centers in Austin, Sunnyvale, India and Boston, toiled around the clock to make sure that Istanbul was a robust design. Later this quarter, we expect you to be able to reap the rewards of their work.
So what can a 6-core processor mean for you? If you have well-threaded applications or you are running a multi-threaded environment like virtualization, we expect that they will be able to eat up those extra cores with a smile on their face. Greater levels of parallelism can help more work get done simultaneously for greater efficiency – and we expect it to all be possible within the same power and thermal levels that you see today with “Shanghai.” As a matter of fact, we are designing these processors to fit into the same socket 1207 architecture as Shanghai, so you should be seeing Istanbul in all of the old familiar places. Now is a good time to ask your server vendor when they expect to be shipping Istanbul processor-based systems. (I am guessing that you will like the answer.)
Istanbul is expected to bring 24-core computing to the 4-socket space with true interconnected and scalable performance (not 24 cores on a front-side bus). And in the 2-way space we expect that you’ll see 12-core systems; providing the perfect balance of performance and power for scale-out virtualization. I’m sure Margaret will have a thing or two to say about that, so keep an eye on our virtualization blog.
Istanbul is the city that lies at the gateway of the world, with one foot in Europe and one foot in Asia. If your applications are craving the greater scalability only additional physical cores can deliver, then look no further than AMD’s planned six-core Istanbul processor– designed to give you one foot in the world of greater scalability and the other one solidly planted this quarter.
John Fruehe is the Director of Business Development for Server/Workstation products at AMD. His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.
Return on Hype
Posted by John Fruehe in 2:34 pm
(10 votes, average: 3.80 out of 5)Recently, our competitor claimed an amazing Return on Investment (ROI) statistic – replace 9 older single core servers with 1 new multi-core one and repay that investment in less than a year. “The cost savings from energy alone will pay for new servers in about eight months[1].”
Having been around IT planning teams for the past 15 years, hearing any ROI statistic always sets my radar off. This is no exception. To me, this claim feels very unusual and there are only 2 explanations that I can think of; Either they are trying to oversimplify a very complex calculation by only looking at one factor (power); or they simply don’t understand the complexity of enterprise applications. Either way they risk doing a major disservice to customers.
The argument that a company can pay off the investment in a new multi-core server by retiring 9 older single core ones is akin to buying a new hybrid car and raving about how much money you are saving every time you fill the tank, ignoring that you had to purchase a car in the process. Return on investment should encompass all of the costs of a solution; otherwise it risks overstating the return.
Let’s take a look at retiring 9 single core servers by consolidating them down to one multi-core server. Simplistically you are going to incur the following costs:
· Consolidation prep – you have to actually do all of the planning and prototyping of the system, mapping data, etc., this is not a simple “copy and paste” exercise. Let’s not forget the data center planning piece of this exercise. You are going to have to remove all of the systems and install a new one.
· Migration of the data – this includes the actual movement of the data. Maybe you get lucky because all 9 servers magically had the exact same data structures and can all coexist happily with each other. Or not. I’m going to bet on “not”, I’ve seen enough of these projects.
· Security – You had 9 separate servers with 9 separate ACLs or security profiles set up to manage who could – and more importantly – could not access the data. Whenever you start consolidation of systems, it is important to make sure that the Marketing Department can’t see the Payroll Department’s files.
· Testing – once you have the new servers in the rack, you don’t actually just flip a switch. You are going to have to touch all the applications that touch that server. Including middleware, backup, security, and network infrastructure. One incorrect MAC address can result in a bunch of troubleshooting if you can’t quickly diagnose the problem.
· Unplanned consequences – Did you ever add a new user and find another suddenly can’t print? Most project managers I’ve worked with include some measure of “overage” to the project to help compensate for having to track down the “stragglers” of any project.
· Licensing changes – Well, 9 servers running 9 copies of the old program might be a sunk cost in ROI, but I am betting that as you consolidate these servers you may end up needing to upgrade to the newest version of the software in order to handle the complexity of the new environment.
· Disposal – you will need to get rid of the old systems, let’s not forget that you can’t just leave them in the dumpster (don’t forget to take the time to truly destroy the hard drives…)
And this is all just the tip of the iceberg, I’m sure that each one of you can provide your own list of hidden costs in trying to do a project. There is a human cost, and with the typical cost of ~$65/hour (the fully burdened cost estimate from the last project I worked on a few years ago) the human costs will likely dwarf the hardware purchase. If you don’t comprehend these costs, you can’t accurately assess ROI.
I’m not naïve in thinking that projects like this happen every day. But it is a bit naïve to think that power costs alone can determine ROI.
Looking at the typical server deployment, you can rest assured that the hardware is the lowest cost of the project by far. So if you want to do yourself a favor, don’t let your company fall for the “pays for itself in 8 months” hype that’s out there – do your own research and get the full story with all the costs revealed. Otherwise you’ll be the one explaining things to the CFO.
John Fruehe is the Director of Business Development for Server/Workstation products at AMD. His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.
[1] http://download.intel.com/products/processor/xeon/dc55kprodbrief.pdf Intel footnote – Source: Intel. March 2009. Compares replacing nine four-year-old single-core Intel® Xeon® processor 3.8GHz with 2M cache-based servers with one new Intel Xeon processor X5570-based server. Results have been estimated based on internal Intel analysis and are provided for information purposes only.