Itanium 2 Performance

I prefer the AMD 64 bit chip, backwards compatibility is a requirement as far as I''m concerned. I have way too many games and apps that won''t get ported to a new arc and the 32 bit emulation is supposed to really, really suck on the IA-64.

If you ask me, backwards compatibility isn''t all it''s cracked up to be. Sure, it''s cool for the first few years, but after that, it''s just legacy crap that should be removed. Just look at Windows ME - how much of it''s crap-ness can be attributed to backwards compatibility?

If I had my way, I''d have all of you shot!

---

codeka.com - Just click it.

I prefer X86-64, for the simple reason that it does both compatibility and improvement. There are some very definate steps away from X86 in the 64-bit mode, but it doesn''t leave users out in the cold. Once the compatibility part is no longer necessary, they can move to the "64" part of the architecture exclusively.

quote:

---

Original post by Dean Harding
If you ask me, backwards compatibility isn''t all it''s cracked up to be. Sure, it''s cool for the first few years, but after that, it''s just legacy crap that should be removed.

---

You''ll find that the life cycle of large systems (e.g. Operating Systems, large engineering solvers, 3D modelling packages) isn''t that short. I have a friend who had to learn a programming language older than he was, Fortran77, when he signed up to work at an engineering firm. Their software is so big, and so complex, that it cannot possibly just be rewritten just like that . Backwards compatibility will add 5-10 years of life to big programs like this, and in this case will save the company millions (which is a lot of money because they''re only about a dozen people). The longer their old code makes money the longer they have to plan a 64-bit port, and in theory the longer they plan the port the smoother it will go.

quote:

---

I have way too many games and apps that won''t get ported to a new arc and the 32 bit emulation is supposed to really, really suck on the IA-64.

---

Itanium''s IA-32 execution (which is done in hardware) is very slow. However, there are 2 possible paths Intel may take:

1. Continue improving it (making a more complex chip.)
2. Remove it altogether and use dynamic binary translation.

I think the latter option is quite feasible. Much research has been done in this area and programs like FX!32 and Dynamo demonstrate that reasonable performance can be attained with this technique.

quote:

---

I prefer X86-64, for the simple reason that it does both compatibility and improvement.

---

Compatibility, yes. Improvement? I haven''t seen any data yet that indicates X86-64 will offer significant improvement over X86 in terms of performance.

The extended address space will be great for servers that need more than 64GB of RAM, but I''ve seen nothing so far to indicate that the Opteron''s performance will be a significant improvement over the Athlon''s, either now or later on down the road.

quote:

---

Once the compatibility part is no longer necessary, they can move to the "64" part of the architecture exclusively.

---

X86-64 will probably never be able to drop IA-32 compatibility. It''s still just an ugly hack to an already ugly ISA. It adds 8 more registers, which is nice, but pales in comparison to IA-64''s 128 GPRs and 128 FPRs. If I''m not mistaken, X86-64 still keeps 8 stacked floating point registers that X86 had.

It still uses variable length instructions (with a maximum length of 15 bytes), prefixes, and all that nasty stuff. Code density is nice, but it''s really not much of an issue any more on the sort of systems this processor is targeted at.

Even if the architecture were moved to 64 bits exclusively, it would still be ugly compared to its competitors -- namely IA-64 and POWER4.

How much more can be squeezed from out-of-order execution engines? The POWER4 is 8-way superscalar, whereas the Itanium 2 can only execute a maximum of 6 instructions in parallel. However, the Itanium can apparently sustain throughput closer to its maximum than POWER4 can, as might be evident by the benchmarks.

VLIW is certainly not some outdated 80''s idea without merit. IA-64 was designed with input from compiler people, contrary to what many say ("Compiler writers hate IA-64.") Maybe MS''s compiler team doesn''t like IA-64, but Intel has arguably one of the best (if not the best) compiler teams in the world and as a result, their compilers are actually doing pretty well.


Compatibility and support are major issues for the IA-64 right now, and I''ll agree. There simply isn''t any software support right now. The OS selection isn''t bad (Linux, HP-UX, Windows XP, and soon FreeBSD and maybe OpenVMS), but where is the rest of the software?

Hardware availability isn''t good either. I think only 500 Itanium systems have been shipped, and I don''t know if any Itanium 2 systems have been deployed (they''ve been available to order from HP for a while now, but they won''t have any in stock until later this year.) I don''t think any vendors besides HP are selling Itanium 2 systems (SGI, Dell, and others say they will have systems available.)

Pricing is steep, but the price-performance ratio is one of the best around. For $5K-$12K you can get an HP workstation that can blow away any workstations IBM has to offer in terms of shear processing power. But I guess there is no incentive if there''s nothing to run on these machines. Supposedly CERN has been very impressed with Itaniums, but I bet those guys write their own software anyway

Intel is definitely in a very tough situation right now: IA-64 is arguably now the king of the hill in terms of power, but it''s just not making any inroads against established proprietary systems such as those from Sun, IBM, and SGI. Nowadays, there are less architectures to choose from, so introducing something radically different is really difficult.

On the plus side, _if_ (that''s a big "if") IA-64 is adopted by a good number of vendors, it has the potential of offering state-of-the-art power without locking you into a specific company''s systems. After all, UltraSPARC is only available on Sun systems and POWER is currently only available in machines from IBM. Maybe IA-64 could actually make high end performance accessible to normal users?

Alpha was a similar situation, but look what happened to it I really hope the same thing doesn''t happen to IA-64, but it looks like it might.

Anyway, I just wanted to comment on the hardware itself. The concerns about compatibility, availability, and IA-64''s long-term success are definitely valid. I''m just a little tired of people claiming it''s an under-performing architecture

---
Bart

quote:

---

Original post by trzy
I'm just a little tired of people claiming it's an under-performing architecture

---

I'm tired of what anyone on the Internet says about all hardware . Hardware enthusiast sites are everywhere, and now everyone things they know oh so much about such-and-such video card architecture and this company's financial woes.

I've taken a wait and see approach. When the 64-bit world is upon us, I will take time to evaluate everything I can observe. Until then, I will ignore what everyone says. This might leave me a little behind the curve when the time comes, but at least I won't get frustrated by legions of anonymous Internet users talking out of their asses.

[edited by - Graylien on August 14, 2002 2:11:37 PM]

I''d be interested in seeing .NET performance on an Itanium. Since .NET is a JIT compiled language it has binary compatibility with any architecture/OS that implements the framework (in this case Windows.NET 64).

Since all our products at work are written in VB.NET and C# we should see immediate advantage to switching our servers to the Unobtanium (since the framework will take care of compiling to the new instruction set).

Epolevne

quote:

---

I'd be interested in seeing .NET performance on an Itanium. Since .NET is a JIT compiled language it has binary compatibility with any architecture/OS that implements the framework (in this case Windows.NET 64).

---

Good question. Intel has been researching Java under IA-64, but I wonder what the situation with .NET is. Will MS be able to take advantage of the Itanium? It will all depend on how capable MS's .NET JIT team is. They could very well deliver a less-than-spectacular product, and then investing in an Itanium just wouldn't be worthwhile.

I don't even know if Windows performs well under IA-64 (compared to Linux and HP-UX.)

---
Bart

[edited by - trzy on August 14, 2002 2:37:19 PM]

quote:

---

Original post by Graylien
You''ll find that the life cycle of large systems (e.g. Operating Systems, large engineering solvers, 3D modelling packages) isn''t that short.

---

Sure, the product we write at work is over 20 years old. I know the benefits of backwards compatibility. But Intel machines aren''t mainframes (whatever Intel is trying to tell people) - they''re comsumer PCs. The lifetime of a consumer piece of hardware is damn short, a couple of years tops.

If I had my way, I''d have all of you shot!

---

codeka.com - Just click it.

quote:

---

Original post by Dean Harding
But Intel machines aren''t mainframes (whatever Intel is trying to tell people) - they''re comsumer PCs. The lifetime of a consumer piece of hardware is damn short, a couple of years tops.

---

Don''t confuse the lifetime of PC with the architecture of the chip inside it. Code that works on a 286 still works on a Pentium IV. Itanium will be around for a long time, in various incarnations.

And of course, the Itanium and similar Intel machines are not mainframes. When I provided examples of large software systems (operating systems, large engineering solvers, 3D modelling packages, etc.), I specifically chose them so people wouldn''t think of old school mainframe stuff. The point is that large software systems can''t just be rewritten because they contain legacy code. This is just simply not practical, and your original post made it sound like you thought it was.