The Palominos are comming!

[Wombat]

Really? And why not?

Especially if it happens to be a reliable performer. Companies will go with whatever performs and runs their software. That's how Unisys and Samsung seem to do fine: fitting certain people's needs.

[isochar]

Companies go with *proven* performance and reliability. Intel dominates the corporate desktop for this reason.

IA-64 is their attempt at taking over the corporate server. From the performance that we've seen from the Itanium, it looks like Intel will be successful at grabbing that 40% server market share by 2005.

[superfly]

I think that it's unavoidable that AMD's hammer series of chips will use a longer pipeline so that it can acheive much higher clock speeds(5 ghz +).

Even the athlon with it's 12 stage pipeline is slower clock for clock that their previous K6-2/k6-3 series which had only a five stage pipeline,unfortunately that was one of the reasons why the k6 series of chips just coudn't be clocked very high.

Each and every new generation of chips that both intel and AMD released over their history featured longer pipelines,sometimes twice as long,so that they could be clocked higher,it's as simple as that.

[dZeus]

fall-out rate with the very fragile cores of the Thunderbird 1333 and 1400 models, as well as huge heat problems are just a few examples why many big OEMs don't go the AMD route.

You don't want to know how large percentage of the 1333 and 1400 CPUs get crushed cores (i.e. small flinters breaking off on the sides) and therefor have to be replaced.

[GT98]

dZeus,

Quick question, what makes the 1.33 and 1.4GHz processors more likely to get damaged vs slower models when they are physically the same thing?

Scott

[dZeus]

hehe... good question... officially there's no difference other than higher ratings for the core afaik....

but then again, the computer shop I work at part-time has had several cases of 1333 and 1400 thunderbirds going kaput (cracked core), especially when placing the cooler (mostly orbs with badly designed mounting mechanisms that put a lot of pressure on a particular spot of the thunderbird core), while almost close to nihil of the 1200 and slower models have had this problem. And this hasn't just happened with a _few_ 1333 and 1400 CPUs... with a whole lot actually (in the range of 10 - 30). And it's not that we sell the 1333 and 1400 a lot more than the slower models.

So I guess they use a different material for the 1333 and 1400 cores that has better heat-conductivity, but is also more easier to break

Management of my company has had contact with wholesalers who import AMD CPUs, and they told us that AMD was aware of this problem but nevertheless don't accept any RMA for physically broken cores....

so we get a lot of angry customers because of their policy.... fortunately the stuff has calmed down a lot after we immediately changed the notices and our RMA policy after we got this news from one of our wholesalers.

btw. the amount of broken cores also drastically lowered when we stopped selling Thermaltake Orbs... not that I mind, because they're piece of crap coolers imho

[Pneumatic]

So in your opinion what is the best cooler for the AMD's?

(short of peltiers and water )

[Wombat]

This still seems like a packaging issue to me. I don't know enough about AMD's fab-screen-package methods to know for sure though.

[GT98]

Originally posted by Kindness!
So in your opinion what is the best cooler for the AMD's?

(short of peltiers and water )

I would suggest getting something that mounts the motherboard its self like a Swiftech MC462-A. I got one and its expensive, but just about impossible to destroy a AMD chip this way.

I've built 2 other AMD systems (both 1GHZ machines) and I didn't break them, just have to take your time and dont force anything! One of them has an Orb type heatsink on it, but it was designed for a Socket 462 chips so I didnt have much problems putting it on....


Scott

[DentyCracker]

I've built several machines with all types of Durons and T'birds and haven't cracked a single core. If peple would just read , they would haverealised that the original orbs are not to be used on socket 462 chips d-uh. The hammer chips will almost certainly have longer pipelines to allow them to scale up in speed. I believe the performance rating is stupid. I want to know the clockspeed of the damn chip not some "indicator" of its speed. I refused to buy them cyrix PR crap chips. The performance rating is almost an indication that they have given up the MHz battle. I believe that with SOI and isotopically pure silicon they can get the Athlon core up to 2.5 GHz but we'll see. BTW clockspeed for clockspeed my Duron at 500 MHz kicked my K6-2 at 500's ass all over the floor. You seem to forget the little matter of a vastly improved FPU on the K7

[Topha]

i read somewhere, that AMD will show there first running hammer-chip this months on some conference.
also their hammer chip will use the performance rating, the 2gig version having a 3500+ rating, which, if it is true, means that they use 0.13 mikron (which they probably will) and not! use longer pipelines, because if they would, they wouldnt start at 2 ghz
i just hope its true

[superfly]

Remember,i wasn't talking about absolute performance between two different cpu's of different generations,because it depends on the type of app you're using and whether it needs a chip with a stronger fpu in the first place.


What i was saying is that the IPC performance(instructions per clock)are always better with chips that have a shorter pipeline,but having a short pipeline hinders the clock speed potential that any given cpu may be capable of,all things being equal of course.


That's why we see solutions like bigger,lower latency caches,better branch prediction,etc,to try and minimize pipeline stalls by having the most often used data already stored in the cache of the cpu where it's 10~15 times faster than accessing that data from system memory,no matter what memory type is used or how fast the fsb is.

[Wombat]

Things are changing though. Cache will still be really important, but IA64 is taking a bit of a different approach to things. VLIW over super-scalar architectures. It requires more from the compilers, but it can really help things along.

[superfly]

The performance of any given cpu is dependent on the quality(in terms of optimisation)of the compiler used and it's always been that way.


Look at how cpu makers have introduced new extensions like MMX,SSE,3D NOW,SSE2,to basically bypass inherent limitations in X-86 cpu's,but again,this requires updated compilers to take advantage of those new features since no developer will go to the trouble of writing code in assembly directly,no matter how good those extensions may be.


On the itanium however,the compiler is the critical element(way more than on conventional cpu's) that will make the IA-64 class of cpu's sink or swim,but the main problem is that not all software benefits from paralelism that's inherent in IA-64 architecture,no matter how good the compilers are.

[Wombat]

Parallelism is desired by almost everybody, EPIC just makes it easier. If it weren't running VLIW, then you'd be shooting for OOO execution. I've worked on OOO too. It's a tres cool feature, but damn it's nasty to debug.