Actually, it depends heavily on the type of workload. For media/scientific stuff, a large cache can be detrimental, as larger means slower and in this case under-utilized. Itanium actually does well on these because it's easy for the compiler to VLIW-ize the very regular code.
For OLTP-like server workloads, large caches are, however, crucial. Because they almost always wait for cache misses to resolve, these things barely manage to execute an instruction every couple of cycles, so actually having VLIW doesn't really mean much.
This doesn't mean that a P4 won't benefit from larger caches in a server configuration. There have been research papers looking at these workloads, and in general larger caches always help in such a configuration.
I'm a little bit confused. The problem is how to extract instruction level parallelism (ILP). The stacked register file doesn't really address this. The purpose of a stacked register file is to reduce the number of load/stores to the actual stack.
The main problem with it, however, is that it makes register renaming very difficult to implement. Register renaming is quintesential for out-of-order execution. BTW, the other big player that has an ISA with stacked register file - Sun - still does in-order processing, and this is actually one of the reasons their processors were dubbed "UltraSlows".
1. transistor count. You do need more transistors for decoding x86 into micro(mu)-ops, but in the end your L2(3) cache is gonna be >50% of your chip area. Interestingly enough, Itanium chips are overloaded with L3, and in fact, the first chip to break the 1billion transistors is an Itanium II chip. The good performance of Itanium comes a lot from its shitload of caches; nothing's preventing Intel from loading the P4 with caches though.
2. x86 is bad/ugly/dirty/whatever, however Itanium is not exactly clean either. The stacked register file is a good example of that. I personally prefer x86-64, which takes the evolutionary approach: fixes quite a few of the problems of x86, while still retaining the core features.
3. x86 chips do out-of-order execution; Itanium, OTOH relies on the compiler to schedule instructions and bundle them together. The main problem here is that doing instruction scheduling statically is much, much harder than doing it dynamically.
An average program has a basic block size that is less than 10 instructions. It's very hard to find parallelism within such small basic blocks, so to be efficient at all, you need to do profiling to build traces/hyperblocks. In fact, profiling on the Itanium can give you a performance boost of 30%. However, profiling is hardly desirable from a software developer's perspective
Is that what you're suggesting them to do? They make most of their money from Windows, Office and SQL Server licenses. You're effectively suggesting them to sabotage about
37% of their revenue.
The H-1 part I simply don't see. It's simply bringing people that are (much) more skilled than the average American. Furthermore, there's a cap on them - about 200k IIRC. This means that there's basically one H-1 worker per 1500 Americans. Finally, let's not forget that other countries are sometimes better in training specialists in some fields. It's very weird that these countries are very upset that they're losing their top specialists (after subsidizing their education), while in the States people feel threatened by them.
BTW, let me just point out that almost any kind of commoditization helps the economy, including the one of the workforce. For instance, post-WW2 Germany did benefit from cheap Turkish workforce, which contributed to their rebuilding effort. Well, they tend to forget that now:)
Jobs going to foreign countries or foreigners coming here to work ?
And btw, "free market" is just an idealized abstraction - with the risk of burning some karma I'll say it's pretty crude one, actually.
The notion of "free market" was developed during the Enlightenment or "Age or Reason", when Europe started to attribute everything to science and reason (after the Middle Ages attributed everything to the will of God). As a consequence, it is based on the fact
that players in the market make their decisions rationally, when, in fact, they very clearly do not. A free market also has failure points like monopolies, cartels, or the fact that it totally ignores environmental issues such as pollution.
The US is probably the closest country to a free market, however, it still does things like subsidizing agriculture products, and even steel a couple of years ago. As for Europe - they're actually much "worse" in not following the free market dogma.
As for your suggestions. 1. Saying "let's prevent people from getting H1s and work within the States" is clearly promoting isolationism and limitting a person's rights instead of more freedom. 2. yeah, let's just close China off... . You're probably ignorant of the fact that the US has been 0wn3d by them. They're basically subsidizing America's unbelievably-high debt. We can't really upset them too much.
Finally, I'll say that I'm not against economic freedoms. What I'm against is using the "free market" dogma like this.
A company that is publicly-traded is going to grow for as long as it can, as the stock market rewards growth prospects much more than profit margins (btw, M$'s profit margins are around ~30%).
That said, VoIP is a pretty reasonable direction for M$, it's something that they can easily embed in their OS and rip the benefits afterwards.
The
Warp drive
, or Alcubierre drive, isn't actually a phantasy. You need exotic matter (i.e. negative g force) to make it work, but otherwise it doesn't violate relativity.
There are many others: Pulsar timing, Astrometry, Gravitational microlensing, Transit method, Circumstellar disks, and... Direct observation (courtesy of wikipedia
It's actually not that difficult to make a standard-compliant browser. The real problem is that IE is extremely permissive, and most of this world's sites are built specifically for IE. This was a deliberate move by Microsoft.
The real problem with other browsers is that it's extremely difficult to emulate IE's behavior, and not w3c conformance.
is good until someone breaks it. In the best case scenario for Intel & media partners, it'll take a modchip (something on the memory bus, for instance) to bypass this. In the worst case scenario, software.
"Secure hardware" is an amazingly difficult thing to achieve (by secure I mean secure from its user, of course). For instance, in the late 90s, smartcards were hacked by figuring out bits from their keys with differential power analysis.
The security problem can be easily fixed in a new core revision. The main problem with Intel's hyperthreading is their somewhat simplistic implementation, which greatly limits performance (for instance, when you enable HT, most of your out-of-order structures are physically halved).
From a technical perspective, there's still a great deal of value from doing (proper) HT on top of a 2-core design - namely, you get 4 execution contexts. But that negatively impacts sales (very few would still need SMP, for instance)
Did you try the new high-end ones, with 10+MB/s read rate?
I'm asking because I want to use a CF drive myself, and CF->IDE adapters are dirt cheap (after all, the interface is almost identical from an electrical point of view)
Slashdot Mirror
For OLTP-like server workloads, large caches are, however, crucial. Because they almost always wait for cache misses to resolve, these things barely manage to execute an instruction every couple of cycles, so actually having VLIW doesn't really mean much.
This doesn't mean that a P4 won't benefit from larger caches in a server configuration. There have been research papers looking at these workloads, and in general larger caches always help in such a configuration.
The main problem with it, however, is that it makes register renaming very difficult to implement. Register renaming is quintesential for out-of-order execution. BTW, the other big player that has an ISA with stacked register file - Sun - still does in-order processing, and this is actually one of the reasons their processors were dubbed "UltraSlows".
2. x86 is bad/ugly/dirty/whatever, however Itanium is not exactly clean either. The stacked register file is a good example of that. I personally prefer x86-64, which takes the evolutionary approach: fixes quite a few of the problems of x86, while still retaining the core features.
3. x86 chips do out-of-order execution; Itanium, OTOH relies on the compiler to schedule instructions and bundle them together. The main problem here is that doing instruction scheduling statically is much, much harder than doing it dynamically. An average program has a basic block size that is less than 10 instructions. It's very hard to find parallelism within such small basic blocks, so to be efficient at all, you need to do profiling to build traces/hyperblocks. In fact, profiling on the Itanium can give you a performance boost of 30%. However, profiling is hardly desirable from a software developer's perspective
Is that what you're suggesting them to do? They make most of their money from Windows, Office and SQL Server licenses. You're effectively suggesting them to sabotage about 37% of their revenue.
BTW, let me just point out that almost any kind of commoditization helps the economy, including the one of the workforce. For instance, post-WW2 Germany did benefit from cheap Turkish workforce, which contributed to their rebuilding effort. Well, they tend to forget that now :)
And btw, "free market" is just an idealized abstraction - with the risk of burning some karma I'll say it's pretty crude one, actually.
The notion of "free market" was developed during the Enlightenment or "Age or Reason", when Europe started to attribute everything to science and reason (after the Middle Ages attributed everything to the will of God). As a consequence, it is based on the fact that players in the market make their decisions rationally, when, in fact, they very clearly do not. A free market also has failure points like monopolies, cartels, or the fact that it totally ignores environmental issues such as pollution.
The US is probably the closest country to a free market, however, it still does things like subsidizing agriculture products, and even steel a couple of years ago. As for Europe - they're actually much "worse" in not following the free market dogma.
As for your suggestions. 1. Saying "let's prevent people from getting H1s and work within the States" is clearly promoting isolationism and limitting a person's rights instead of more freedom. 2. yeah, let's just close China off ... . You're probably ignorant of the fact that the US has been 0wn3d by them. They're basically subsidizing America's unbelievably-high debt. We can't really upset them too much.
Finally, I'll say that I'm not against economic freedoms. What I'm against is using the "free market" dogma like this.
... as long as "they" can steer a small asteroid to Earth (1km in diameter), it's total loss for us.
That said, VoIP is a pretty reasonable direction for M$, it's something that they can easily embed in their OS and rip the benefits afterwards.
I just realized - that guy has a national presence!!!
1. No proper database connection pooling (and no, pconnect is not connection pooling).
2. Can't really run Apache2 in its threaded mode because a lot of php libraries are not thread safe (although php itself is).
Don't get me wrong, I *love* both php and mysql, but for highly-trafficked sites, jsp is definitely a better choice (from my own experience).
The major advantage of CFs is that they don't have seek time, so in practice a 10M/s harddrive is going to be slower than a 10M/s CF.
There are definitely ways of making a farm of x86 servers very reliable - see google, for instance.
My favorite one - Neutronium
The Warp drive , or Alcubierre drive, isn't actually a phantasy. You need exotic matter (i.e. negative g force) to make it work, but otherwise it doesn't violate relativity.
There are many others: Pulsar timing, Astrometry, Gravitational microlensing, Transit method, Circumstellar disks, and ... Direct observation (courtesy of wikipedia
The real problem with other browsers is that it's extremely difficult to emulate IE's behavior, and not w3c conformance.
It really feels like reading Freud.
... is to post an "argument" that's not falsifiable.
"Secure hardware" is an amazingly difficult thing to achieve (by secure I mean secure from its user, of course). For instance, in the late 90s, smartcards were hacked by figuring out bits from their keys with differential power analysis.
From a technical perspective, there's still a great deal of value from doing (proper) HT on top of a 2-core design - namely, you get 4 execution contexts. But that negatively impacts sales (very few would still need SMP, for instance)
I'm asking because I want to use a CF drive myself, and CF->IDE adapters are dirt cheap (after all, the interface is almost identical from an electrical point of view)
By comparison Google had a gross profit of 1.73B last quarter, with a couple thousand employees.
While IBM is doing fine, it's NOT very efficient these days.
That's why they do the graduate engineering/nursing mixers!
Although some of them (such as the Sikh religion) consider that people of different religions can also achieve "transcendence"
Dead Sea scrolls