- Status
razor512
Distinguished
- Jun 16, 2007
- 2,134
- 71
- 19,890
[citation][nom]r0x0r[/nom]Please don't tell us you actually think they're $4 each... >_[/citation]
they need to lower the price to $4 each,
I am tired of maya taking like 16 hours to render something on a quad core.
intel needs to sell these CPU's for $4 each so I can build a few systems that can use those CPU's and act as render slaves.
they need to lower the price to $4 each,
I am tired of maya taking like 16 hours to render something on a quad core.
intel needs to sell these CPU's for $4 each so I can build a few systems that can use those CPU's and act as render slaves.
[citation][nom]nottheking[/nom]That doesn't tell the whole story; that's like saying that the PS3's Cell Broadband Engine has 1 core, and the Xbox 360's Xenon has 3, hence the latter is superior.Like that, Itanium and POWER7 can't really be compared... Even less so, since they're entirely different architectures. (Itanium being IA-64, vs. IBM-POWER for the CBE, Xenon, and POWER7)Itanium tends to have fewer cores, but a lot more complexity per core, allowing each core to process WAY more per clock cycle than any other architecture; hence, it's considered a unique design; while the POWER architecture is entirely 'RISC,' (Reduced Instruction Set) Itanium goes well beyond the x86-standard 'CISC' (Complex Instruction Set) and is in fact perhaps the world's only 'EPIC' (Explicitly Parallel Instruction Computing) architecture. Rather than making the CPU figure out how to order the instructions, EPIC/Itanium includes structuring at the compiler level to control branching, allowing for little confusion or delay for the CPU handling complex, branching instructions. All this works out to EACH thread being capable of handling 6 instructions per clock cycle. That means that while the upcoming POWER7 may have more threads at 32 to 8, Tukwila handles more instructions per clock cycle, at 48 to 32... Couple that with the fact that POWER7 has to cut its instruction complexity to achieve so many threads, and Tukwila almost certainly will provide far more capability at handling complicated instruction trees. So it's all a matter of design; POWER7 will likely win in the FLoPS ring, due to outright having more cores, (and a whopping 4 floating-point units per core) while Tukwila will certainly be better at the Instructions/second race.[/citation]
There are a lot of people who don't really understand the Itanium at all. First of all, it's not in the same league as anything with the Nehalem. Comparisons between the two will tend to show the superiority of the Nehalem without the context of why.
For one, this is two nodes behind the newest Nehalems, and is made on 65nm. More to the point, the transistors are more reliable and less susceptible to soft errors. This requires more size and is generally slower, but, when you need things to be right, you can't run them on a toy that runs fast.
However, the assertion that the Itanium shows better IPC is just not consistent with benchmarks. The idea of an in order processor that instead relies on the compiler to do accurate scheduling sounded good, except such a compiler is impossible to create. We don't have benchmarks for the Tukwila yet, but previous versions did not show better IPC on integer workloads, than even the dreadful x86 processors like Core 2, and particularly Nehalem. This and the fact they can't clock nearly as high (which should give them an advantage on IPC, since caches can be clocked much faster vis-a-vis the core) should make it clear these things are not the IPC beasts they were envisioned to be.
POWER 7 looks to be a superior product by almost all measures. But, that's not an indictment against the IA-64 instruction set. They're on a 65nm node, on a very old architecture that hasn't been upgraded nearly as extensively as IBM has with POWER.
The next version of Itanium, Poulson, looks to be more of a "ground up" effort, and could change everything we think we know about the Itanium. The Tukwila is a nice improvement, but the core isn't changed a whole lot, although the much better system interface is going to help quite a bit.
Also, the Itanium core is NOT more complex than x86. It's actually quite small and simple. That's kind of the point. Move the complexity to the compiler, and get parallelism from it. The problem is, so far, it hasn't worked.
There are a lot of people who don't really understand the Itanium at all. First of all, it's not in the same league as anything with the Nehalem. Comparisons between the two will tend to show the superiority of the Nehalem without the context of why.
For one, this is two nodes behind the newest Nehalems, and is made on 65nm. More to the point, the transistors are more reliable and less susceptible to soft errors. This requires more size and is generally slower, but, when you need things to be right, you can't run them on a toy that runs fast.
However, the assertion that the Itanium shows better IPC is just not consistent with benchmarks. The idea of an in order processor that instead relies on the compiler to do accurate scheduling sounded good, except such a compiler is impossible to create. We don't have benchmarks for the Tukwila yet, but previous versions did not show better IPC on integer workloads, than even the dreadful x86 processors like Core 2, and particularly Nehalem. This and the fact they can't clock nearly as high (which should give them an advantage on IPC, since caches can be clocked much faster vis-a-vis the core) should make it clear these things are not the IPC beasts they were envisioned to be.
POWER 7 looks to be a superior product by almost all measures. But, that's not an indictment against the IA-64 instruction set. They're on a 65nm node, on a very old architecture that hasn't been upgraded nearly as extensively as IBM has with POWER.
The next version of Itanium, Poulson, looks to be more of a "ground up" effort, and could change everything we think we know about the Itanium. The Tukwila is a nice improvement, but the core isn't changed a whole lot, although the much better system interface is going to help quite a bit.
Also, the Itanium core is NOT more complex than x86. It's actually quite small and simple. That's kind of the point. Move the complexity to the compiler, and get parallelism from it. The problem is, so far, it hasn't worked.
liquidsnake718
Distinguished
- Jul 8, 2009
- 1,379
- 0
- 19,310
the question people should really be answering is........ CAN IT RUN A SERVER OF CRYSIS ONLINE WHERE EVERY SINGLE LAYER IS PLAYING WITH AT LEAST HIGH TO VERY HIGH SETTINGS AND AT LEAST 2X AA...... ON A SERVER WITH 30FPS AVG.....
beat that MAXED OUT
beat that MAXED OUT
micky_lund
Distinguished
- Mar 17, 2009
- 672
- 0
- 18,990
tenaciousleydead
Distinguished
- Dec 6, 2004
- 812
- 0
- 18,990
- Jul 16, 2009
- 4,821
- 0
- 22,780
[citation][nom]ben850[/nom]"Don't expect this to be something that you'd have at home to run Crysis, "there you go guys.. don't even ask![/citation]
Doesn't answer the question though does it...
Interesting that this comes out after IBM's Power7 chips. Methinks Intel have stuff they deliberately keep in the drawer till someone try to release a product that competes.
Doesn't answer the question though does it...
Interesting that this comes out after IBM's Power7 chips. Methinks Intel have stuff they deliberately keep in the drawer till someone try to release a product that competes.
ethanolson
Distinguished
- Jun 25, 2009
- 318
- 0
- 18,780
These run Unix, OpenVMS (the best OS ever), Linux and NSK. The coolest thing about these Itaniums, though, is that it runs a version of Windows that doesn't crash.
That's right, crash-proof Windows. Who'da thunk?
That's right, crash-proof Windows. Who'da thunk?
- Jul 16, 2009
- 4,821
- 0
- 22,780
[citation][nom]ethanolson[/nom]These run Unix, OpenVMS (the best OS ever), Linux and NSK. The coolest thing about these Itaniums, though, is that it runs a version of Windows that doesn't crash.That's right, crash-proof Windows. Who'da thunk?[/citation]
Some might say that Windows never was the crashing problem to begin with and that previous crashes attributed to the OS were in fact problems associated with the type of chip normally pumped out to the consumer market.
Shame history and fanboys will argue black is white that Windows was the cause all along... hey ho...
Some might say that Windows never was the crashing problem to begin with and that previous crashes attributed to the OS were in fact problems associated with the type of chip normally pumped out to the consumer market.
Shame history and fanboys will argue black is white that Windows was the cause all along... hey ho...
ethanolson: You should check out what Linus Torvalds(founder of Linux) had to say about itanium, I shall paraphrase for you:
"it sucks, it's slow as hell, and needs to die"
back_by_demand: XP Pro Corporate on "corporate stable" motherboards and (x86) cpus doesn't crash either.
"it sucks, it's slow as hell, and needs to die"
back_by_demand: XP Pro Corporate on "corporate stable" motherboards and (x86) cpus doesn't crash either.
- Jul 16, 2009
- 4,821
- 0
- 22,780
[citation][nom]c0rp0rate_stabl3[/nom]"back_by_demand: XP Pro Corporate on "corporate stable" motherboards and (x86) cpus doesn't crash either.[/citation]
Interesting, that only strengthens the case that it is "user" error or cheap OEMs with incompatible hardware...
Interesting, that only strengthens the case that it is "user" error or cheap OEMs with incompatible hardware...
- Status
TRENDING THREADS
-
Question User RX 580 8gb red devil powercolor, asking original BIOS- Started by PizzaInGame
- Replies: 5
-
-
-
News Microsoft updates Windows 11 24H2 requirements, CPU must support SSE4.2 or the OS will not boot- Started by Admin
- Replies: 19
Facebook
Twitter
Instagram