The UltraSparc IIIi is pathetic compared to the top of the line x86. They are 3 years behind at least. Their processor architecture is an ancient IN-ORDER design. Their peak point to point bandwidth is 3.2 GB/sec and DRAM bandwidth is 4.2 GB/sec. The Opteron is an out-of-order design with almost 2x the bandwidth. The SPECint2000 and SPECfp2000 scores will be double that of Ultra III. The Opteron offers 2x more performance and better scalability.
Just add 64 bit addressing and some high bandwidth point to point links and now you have a scalable x86 server solution. This is what Opteron brings to the table.
I can't believe Sun still wants to compete in the microprocessor market.
I totally agree that Sun cannot compete. How hard is it for a system vendor to add RAS features like hot-swap? Given the big fat margins that servers offer, it's only a matter of time before Dell et al. develop some killer servers to eat Sun's lunch.
Sun has to pay for OS development and processor development. Dell can just use linux and commodity processors from Intel or AMD and just foot the bill for the system development. Sure Linux isn't ready for the enterprise today, but what about 2 or 3 years from now?
Sun's vertical business model is obsolete. When common server/big-iron reference designs are developed, there is no way Sun can compete on the multiple fronts of OS, processor, and system design. It costs billions to develop competitive processors and the massive money and power of Microsoft is just barely able compete with Linux on the server. How is Sun ever going to be competitive? They're too inefficient.
Why not? The ultrasparc is nothing special. Their point to point bandwidth is way below that of an Opteron. The Opteron should scale even better than UltraSPARC III. Someone just needs to build the interconnect:)
It seems like the economies of scale will catch up with Sun and they will eventually lose their high end hardware sales to cheaper x86 based solutions. Sun can make the high end dollars today, but can they make them in the future?
If Sun is making so much money on their high end servers, HP or IBM or Dell will want part of that action and will want to build a scalable system. They'll be able to build one cheaper than Sun by leveraging open source Linux development for the OS and leveraging Intel and AMD for the vastly expensive microprocessor development. Sun spends R&D on both the OS and the processor as well as the system. Their competitors can undercut Sun by setting a series of standards/reference-designs for the high-end datacenter machines (look at AMD's hypertransport for example).
What happens if Intel increases the IPC of their processor? Then their PR system doesn't scale with P4 frequency anymore. So when Intel introduces a faster bus (such as the 800MHz bus coming in the spring) and larger caches (like the 1MB cache on Prescott), the descrepancy between PR-rating and P4 frequency will grow.
Of course, by that time AMD will probably introduce another set of arbitrary PR ratings to scale with the higher IPC P4s.
When designing CPUs there is a constant re-evaluation of the costs and benefits of adding more cache. Adding more cache comes at the expense of spending that additional area on parts of the core processor which could have been used to improve performance (adding more execution units, more buffers, more registers, more decoders etc...)
The reason Sun has larger caches is because they want to optimize multiprocessor performance. In that scenario, the memory bus is an extremely valuable resource and reducing the amount of traffic to the bus by having a larger cache is more benficial than would be the case in a uniprocessor desktop. Their uniprocessor performance sucks. The UltraSPARC III is still an in-order processor (as opposed to the vastly superior out-of-order Pentiums and Athlons). Just check out their SPEC2000 scores.
As a shareholder in MSFT, I strongly disagree. They have paid lots of taxes over the years, well over the claimed 1.8% tax rate. I wish they didn't pay taxes, my stock would be worth more:)
If you want to use sales as a metric to measure size, that's fine.
I'm just pointing out that most wall street types use net income (aka - earnings) to measure the worth of a company. Net income is extremely well correlated to market cap. Therefore, most people on wall street use the earnings/market-cap metric to value the business.
Gross profit does not include operating expenses and is not what wall street uses to assign value to a business. Net income is the metric of choice.
Actual IBM earnings last quarter were: $1,313,000,000 (source: yahoo finance)
Compare to Microsoft: $2,726,000,000 (source: yahoo finance)
I guess that's why the market capitalization of Microsoft stock is more than 2x that of IBM. This ratio happens to be very well correlated to the earnings ratio. At the end of the day, earnings is all that matters when assigning value to a company and that is why most people claim that Microsoft is the largest software company.
Perhaps some of these bioinformation engineers should spend a little time on security. I tried to go to the website of one of the companies referenced in the Economist article and got a defaced website:
Well then compare the EV4/5 to it's conteporary RISC counterparts at the time and you'll find that indeed, they were speed demons compared to the SPARCs and MIPs of the time.
The speed demons won.
Going wider (more IPC) is hard. Exponentially hard in fact. The complexity, power and engineering effort explodes. Going faster via frequency is hard too but not quite as hard as going wider - you can ride the manufacturing benefits of process improvements.
I think what the original poster was trying to point out was that even if the P4 does less work per clock...if the clock is fast enough, it can beat a lower-clocked processor that does more-per-clock.
At the end of the day, performance is measured in time, not IPC.
If you remember the old days when DEC was on top, there was a raging debate in computer architecture. Who would win? The 'speed demons' or the 'brainiacs'? The speed demons would do less per-clock, but would run at a higher frequency than the brainiacs which would do more per clock but at a lower relative frequency.
The writer of this story talked about a 500MHz Alpha vs. a 200MHz Pentium. Obviously Alpha chose the speed demon and won.
Intel's P4 is a speed demon, whereas the Athlon is a brainiac. If you look at current performance as measured by SPEC or whatever benchmarks Anandtech or Tom's hardware have these days, you'll see that the highest end P4s beat the highest end Athlons.
DEC gets the kudos for choosing the high frequency microarchitecture, but interestingly, Intel gets a black eye for it.
As an engineer working on ridiculously complex projects, I welcome sophisticated software tools that make my life easier.
Software should be developed to make engineering more efficient. If tools today are doing things that you would have done manually 5 years ago, and you can't take advantage of it to do better things, then you are probably a weak engineer.
Frankly, if the entirety of your job can be encapsulated in a software algorithm, I question your value as an engineer.
It's not like Intel or AMD are unaware of the fact that most people do not need more processing power. They are forced to develop faster and faster processors due to competitive concerns.
If either Intel or AMD stop making faster chips, their competitor will price them out of the market with faster chips.
Given a choice between a PR3000 processor and a PR26000 processor at the same price, the consumer will buy the PR3000 processor since he/she will get the extra perf. for free. If one company runs far ahead of another company in terms of performance, they can undercut the competition on price.
The entire CPU business model is predicated on the fact that you must obsolete your own products or risk getting priced out of the market by your competitor.
Here's another thing to consider. If you are a software developer, you want to write software that will run well on the majority of machines. The majority of machines at any given point in time is far below the performance of the state of the art. It does not make much business sense to make your software run well only on the highest end machines because then your market is limited. So, most developers will develop for the average machine. Unless of course you are Microsoft and you can do whatever the hell you want.
Id, is a lone example of a company that pushes the limits of computing, but I consider them to be an outlier.
So, Microsoft and outliers like Id will drive the need for more CPU performance.
Slashdot Mirror
You haven't played the nude hacked Dead or Alive Beach Volleyball for XBOX yet have you?
If you have a mod chip, you can get progressive scan dvd playback with the right software.
Just wait 2 weeks and you can buy Opteron systems.
The UltraSparc IIIi is pathetic compared to the top of the line x86. They are 3 years behind at least. Their processor architecture is an ancient IN-ORDER design. Their peak point to point bandwidth is 3.2 GB/sec and DRAM bandwidth is 4.2 GB/sec. The Opteron is an out-of-order design with almost 2x the bandwidth. The SPECint2000 and SPECfp2000 scores will be double that of Ultra III. The Opteron offers 2x more performance and better scalability.
Just add 64 bit addressing and some high bandwidth point to point links and now you have a scalable x86 server solution. This is what Opteron brings to the table.
I can't believe Sun still wants to compete in the microprocessor market.
I totally agree that Sun cannot compete. How hard is it for a system vendor to add RAS features like hot-swap? Given the big fat margins that servers offer, it's only a matter of time before Dell et al. develop some killer servers to eat Sun's lunch.
Sun has to pay for OS development and processor development. Dell can just use linux and commodity processors from Intel or AMD and just foot the bill for the system development. Sure Linux isn't ready for the enterprise today, but what about 2 or 3 years from now?
Sun's vertical business model is obsolete. When common server/big-iron reference designs are developed, there is no way Sun can compete on the multiple fronts of OS, processor, and system design. It costs billions to develop competitive processors and the massive money and power of Microsoft is just barely able compete with Linux on the server. How is Sun ever going to be competitive? They're too inefficient.
Unless you have an SMT processor like the latest P4 which has 2 instruction pointers to execute 2 threads.
Why not? The ultrasparc is nothing special. Their point to point bandwidth is way below that of an Opteron. The Opteron should scale even better than UltraSPARC III. Someone just needs to build the interconnect :)
Your quad processor ultrasparc is slower than a state of the art uniprocessor x86 box.
It seems like the economies of scale will catch up with Sun and they will eventually lose their high end hardware sales to cheaper x86 based solutions. Sun can make the high end dollars today, but can they make them in the future?
If Sun is making so much money on their high end servers, HP or IBM or Dell will want part of that action and will want to build a scalable system. They'll be able to build one cheaper than Sun by leveraging open source Linux development for the OS and leveraging Intel and AMD for the vastly expensive microprocessor development. Sun spends R&D on both the OS and the processor as well as the system. Their competitors can undercut Sun by setting a series of standards/reference-designs for the high-end datacenter machines (look at AMD's hypertransport for example).
I think Sun's business model is in trouble.
I wonder how many people are feeling the same way that you are. Is AMD "Osbourning" themselves?
The whole PR rating thing seems kind of bogus.
What happens if Intel increases the IPC of their processor? Then their PR system doesn't scale with P4 frequency anymore. So when Intel introduces a faster bus (such as the 800MHz bus coming in the spring) and larger caches (like the 1MB cache on Prescott), the descrepancy between PR-rating and P4 frequency will grow.
Of course, by that time AMD will probably introduce another set of arbitrary PR ratings to scale with the higher IPC P4s.
When designing CPUs there is a constant re-evaluation of the costs and benefits of adding more cache. Adding more cache comes at the expense of spending that additional area on parts of the core processor which could have been used to improve performance (adding more execution units, more buffers, more registers, more decoders etc...)
The reason Sun has larger caches is because they want to optimize multiprocessor performance. In that scenario, the memory bus is an extremely valuable resource and reducing the amount of traffic to the bus by having a larger cache is more benficial than would be the case in a uniprocessor desktop. Their uniprocessor performance sucks. The UltraSPARC III is still an in-order processor (as opposed to the vastly superior out-of-order Pentiums and Athlons). Just check out their SPEC2000 scores.
As a shareholder in MSFT, I strongly disagree. They have paid lots of taxes over the years, well over the claimed 1.8% tax rate. I wish they didn't pay taxes, my stock would be worth more :)
Here are a couple of sources:
Yahoo annual data - this is taken from the government SEC filings
Microsoft data - this is just a check against the previous numbers - they are the same
If you want to use sales as a metric to measure size, that's fine.
I'm just pointing out that most wall street types use net income (aka - earnings) to measure the worth of a company. Net income is extremely well correlated to market cap. Therefore, most people on wall street use the earnings/market-cap metric to value the business.
Microsoft is an extremely powerful company, but not powerful enough to avoid paying taxes. Microsoft paid hefty taxes last year.
I refer you to the following evidence on yahoo finance.
Here are numbers from last quarter:
Income Before Tax $4,069,000,000
Income Tax Expense $1,343,000,000
Gross profit does not include operating expenses and is not what wall street uses to assign value to a business. Net income is the metric of choice.
Actual IBM earnings last quarter were:
$1,313,000,000 (source: yahoo finance)
Compare to Microsoft:
$2,726,000,000 (source: yahoo finance)
I guess that's why the market capitalization of Microsoft stock is more than 2x that of IBM. This ratio happens to be very well correlated to the earnings ratio. At the end of the day, earnings is all that matters when assigning value to a company and that is why most people claim that Microsoft is the largest software company.
Check out Shazam. They offer a unique service that does something similar to this.
Perhaps some of these bioinformation engineers should spend a little time on security. I tried to go to the website of one of the companies referenced in the Economist article and got a defaced website:
Well then compare the EV4/5 to it's conteporary RISC counterparts at the time and you'll find that indeed, they were speed demons compared to the SPARCs and MIPs of the time.
The speed demons won.
Going wider (more IPC) is hard. Exponentially hard in fact. The complexity, power and engineering effort explodes. Going faster via frequency is hard too but not quite as hard as going wider - you can ride the manufacturing benefits of process improvements.
I think what the original poster was trying to point out was that even if the P4 does less work per clock...if the clock is fast enough, it can beat a lower-clocked processor that does more-per-clock.
At the end of the day, performance is measured in time, not IPC.
If you remember the old days when DEC was on top, there was a raging debate in computer architecture. Who would win? The 'speed demons' or the 'brainiacs'? The speed demons would do less per-clock, but would run at a higher frequency than the brainiacs which would do more per clock but at a lower relative frequency.
The writer of this story talked about a 500MHz Alpha vs. a 200MHz Pentium. Obviously Alpha chose the speed demon and won.
Intel's P4 is a speed demon, whereas the Athlon is a brainiac. If you look at current performance as measured by SPEC or whatever benchmarks Anandtech or Tom's hardware have these days, you'll see that the highest end P4s beat the highest end Athlons.
DEC gets the kudos for choosing the high frequency microarchitecture, but interestingly, Intel gets a black eye for it.
What are you talking about?
The current Northwood P4's have 512K of L2 cache, that's double the size of the orignal Willamette P4's 256K cache.
Athlons have 256K of L2 currently. Barton, which will come out next year will have 512K.
Also, memory bandwidth is increasing.
Those score you quote for Clawhammer are actually for Sledgehammer (server parts with 1MB of cache).
The desktop Clawhammer scores will be lower with only 512k of cache.
It's not symmetric multithreading.
It's SIMULTANEOUS multithreading.
This means that both threads are in the processor pipeline simulatenously.
As an engineer working on ridiculously complex projects, I welcome sophisticated software tools that make my life easier.
Software should be developed to make engineering more efficient. If tools today are doing things that you would have done manually 5 years ago, and you can't take advantage of it to do better things, then you are probably a weak engineer.
Frankly, if the entirety of your job can be encapsulated in a software algorithm, I question your value as an engineer.
It's not like Intel or AMD are unaware of the fact that most people do not need more processing power. They are forced to develop faster and faster processors due to competitive concerns.
If either Intel or AMD stop making faster chips, their competitor will price them out of the market with faster chips.
Given a choice between a PR3000 processor and a PR26000 processor at the same price, the consumer will buy the PR3000 processor since he/she will get the extra perf. for free. If one company runs far ahead of another company in terms of performance, they can undercut the competition on price.
The entire CPU business model is predicated on the fact that you must obsolete your own products or risk getting priced out of the market by your competitor.
Here's another thing to consider. If you are a software developer, you want to write software that will run well on the majority of machines. The majority of machines at any given point in time is far below the performance of the state of the art. It does not make much business sense to make your software run well only on the highest end machines because then your market is limited. So, most developers will develop for the average machine. Unless of course you are Microsoft and you can do whatever the hell you want.
Id, is a lone example of a company that pushes the limits of computing, but I consider them to be an outlier.
So, Microsoft and outliers like Id will drive the need for more CPU performance.