Deneb Samples are almost out

[JumpingJack]

Other than the 64 bit thing, where else has AMD got ahead of the industry?

Usually it is Intel with their new SSE extensions driving things.

I think in the multithreaded scaling department.... well, hard to argue it is all relative.

Intel utilized the only (only emphasized) advantage of the FSB since it provides a quick solution for increasing core count -- giving us quad cores early. It sacrifices potential performance (i.e. not quite as good scaling), but provides a good boost ... AMD, by nature of their design, took much more time and went monolithic... better scaling, but not quite as good single threaded (where in general is where the performance boost was still needed) ...

These are really the only two examples I can think of that form the basis of my opinion.... I did not claim it was a good opinion, just an opinion ...

EDIT: New instructions though are harder to state in this context... it is a chick or the egg type thing with things like SSE, etc etc. But if memory serves ... AMD launched 3DNow! before SSE -- sorta the counter to MMX and it was indeed much better.


Jack

[gosh]

AMD seems to be way ahead of the needs of the industry in general. They provide 64-bit extensions but ahead of when the industry was really ready for it (in desktop, server was begging for it hence they did really well in server), with the monolithic quad they deliver better multithreaded performance over their single threaded capability, but most software was still single threaded so it did not shine as bright.

yes! that was probably their biggest error with phenom or the error might have been there because they underestimated that programmers need to adapt to the lowest common dominator. I also think that they underestimated how much they was hurt by the slower L3 cache because they needed to handle situations that hardly exists in apps. If Intel had been faster out with the Nehalem then applications would have been more threaded now.
I think that Intel will do what they can in order to change how programmers build their applications. Because the Nehalem is too expensive and Core 2 will soon be old. Deneb is probably on the spot when it is out.

This is only the first step in scaling applications. Next step then there might be new languages where the compiler itself will thread functionality for the programmer. if you have 8 or more cores then the programmer don't need to think any more how many threads that is used. The programmer is just going to create as many as needed.

[Helmore]

*snip*
This is only the first step in scaling applications. Next step then there might be new languages where the compiler itself will thread functionality for the programmer. if you have 8 or more cores then the programmer don't need to think any more how many threads that is used. The programmer is just going to create as many as possible.

Slight correction there. They should already start preparing for 16 and higher core machines, even though that is very tough to do.

[freeloader]

Other than the 64 bit thing, where else has AMD got ahead of the industry?

Usually it is Intel with their new SSE extensions driving things.

First X86 CPU to 1ghz, first X86 dual core processor, hypertransport and the onboard memory controller?

[Chad Boga]

First X86 CPU to 1ghz, first X86 dual core processor, hypertransport and the onboard memory controller?

When Jumping Jack said ahead of the industry, he meant in a way that didn't help AMD out for a good while and at an opportunity cost to taking a slightly different approach.

The things you listed are not what Jumping Jack or myself talking about as they aren't of the same nature.

[freeloader]

When Jumping Jack said ahead of the industry, he meant in a way that didn't help AMD out for a good while and at an opportunity cost to taking a slightly different approach.

The things you listed are not what Jumping Jack or myself talking about as they aren't of the same nature.

Ok, point taken. Although Hypertransport certainly has done them some good in the last few years. So good that Intel copied it (QPI).

[Hornet331]

First X86 CPU to 1ghz, first X86 dual core processor, hypertransport and the onboard memory controller?

mhz is pretty much pointless.. intel had first 2ghz etc etc.

Pentium D beat the X2 for beeing the first dualcore cpu for consumer.

Point for amd (thought HT is an evolution of Alphas EV7 Bus )

i386 SL

[Hornet331]

It's like he decided, "Go sell crazy somewhere else... we're all stocked up here".

http://enthusiast.hardocp.com/articl...50aHVzaWFzdA==

*pewh*

and i thought i was in some kind of alternate univers.

kyle is still kyle.

[freeloader]

mhz is pretty much pointless.. intel had first 2ghz etc etc.

Pentium D beat the X2 for beeing the first dualcore cpu for consumer.

Point for amd (thought HT is an evolution of Alphas EV7 Bus )

i386 SL

We can argue the symantecs for the i386SL or the first "real" dualcore processor, but it's been done ad nauseum already. Peace....

[JumpingJack]

First X86 CPU to 1ghz, first X86 dual core processor, hypertransport and the onboard memory controller?

Yeah, like Chad said above ... the 1 GHz race was exciting, AMD did it by a few weeks as I recall, hypertransport/IMC were implementations for competitive performance, this was not ahead of what the market was ready for.

64-bit, though, was a good example. Server accepted it readily, they needed the extra memory address space, but for the bulk of the market, the volume of the market... it was way ahead of it's time -- AMD pushed 64-bit into desktop when it really wasn't needed or wanted. Even today, the adoption is still slow but improving (i.e cheap memory to 8 gigs, Vista 64 readily available with drivers). But it took 3-4 years before 64-bit really penetrated the mainstream.

Aside from that, let's not get into who did what first. That is a different ball of wax.

[Drwho?]

First X86 CPU to 1ghz, first X86 dual core processor, hypertransport and the onboard memory controller?

The first x86 CPU with memory controler on dice was the i80386sx/sl, and the "i" is for Intel! The rest of the claim is as bogus.

[Drwho?]

just kidding ... ok?
tired of hearing this memory controler stuff that is not true!

[duploxxx]

as usual in these amd threads pages full of crap posts

fyi it's about deneb, not some stupid race who was first in what....

one thing is for sure, thx to AMD we finally got rid of that crappy pentium iv architecture. (that should have stopped @ northwood)

[justthefax]

as usual in these amd threads pages full of crap posts

fyi it's about deneb, not some stupid race who was first in what....

one thing is for sure, thx to AMD we finally got rid of that crappy pentium iv architecture. (that should have stopped @ northwood)

Nehalem is basically a great validation of AMD's architectural choices -- on-die memory controller, Hypertransport, small L2, big L3, monolithic quadcore etc. People can argue on the details, whether inclusive or exclusive cache hierarchies are more effective etc, but the gist is the same.

Too bad AMD wasn't pragmatic enough about when to introduce features, though. Kinda like the R600 was so forward-looking that it was beaten by Nvidia. Until the R7xx generation they didn't focus on optimizing for NOW rather than 2 years in the future. Hopefully AMD has applied that lesson on the CPU front in Deneb and Bulldozer. I still can't believe AMD decided to aim for quad-core or bust in 2007 when the market was -- and is -- dominated by dual-core. They don't seem to have a new dual core design even now, apart from broken quads.

[duploxxx]

Nehalem is basically a great validation of AMD's architectural choices -- on-die memory controller, Hypertransport, small L2, big L3, monolithic quadcore etc. People can argue on the details, whether inclusive or exclusive cache hierarchies are more effective etc, but the gist is the same.

Too bad AMD wasn't pragmatic enough about when to introduce features, though. Kinda like the R600 was so forward-looking that it was beaten by Nvidia. Until the R7xx generation they didn't focus on optimizing for NOW rather than 2 years in the future. Hopefully AMD has applied that lesson on the CPU front in Deneb and Bulldozer. I still can't believe AMD decided to aim for quad-core or bust in 2007 when the market was -- and is -- dominated by dual-core. They don't seem to have a new dual core design even now, apart from broken quads.

your compare is on personal desktop happenings, and you are right they did not bring what was required for the market at that time.

BUT amd's focus is on server and believe me the future there is for sure on quad core and higher, both intel and amd cpu prices are way more focused on the quad core offerings for months now and all dualcore have an EOL mark for a long time .

[Hans de Vries]

just kidding ... ok?
tired of hearing this memory controler stuff that is not true!

mmmh,

François, the first Intel processor with on-die DRAM memory controller was this one...

http://www.cpu-galaxy.at/CPU/Intel%2...heet_Nov71.pdf



Regards, Hans

[Shintai]

mmmh,

François, the first Intel processor with on-die DRAM memory controller was this one...

http://www.cpu-galaxy.at/CPU/Intel%2...heet_Nov71.pdf


Regards, Hans

I think the reference is high volume mainstream parts.

[mAJORD]

Wow now that's a nice surprise, Hans on our forums..

Love your website



/ot

[Drwho?]

mmmh,

François, the first Intel processor with on-die DRAM memory controller was this one...

http://www.cpu-galaxy.at/CPU/Intel%2...heet_Nov71.pdf



Regards, Hans

you are perfectly right Hans, I was talking about x86 processor, able to "boot dos" ... Hehehe ...
Mr Ruiz had the bad habit to attribute himself a lot of things that are not his, and your coming here is nice, and remind everybody that Hector was bragging about something Intel did in 1971 ... ;-)

by the way, I really love your web site! Keep the History straight!

Francois

[Helmore]

Nehalem is basically a great validation of AMD's architectural choices -- on-die memory controller, Hypertransport, small L2, big L3, monolithic quadcore etc. People can argue on the details, whether inclusive or exclusive cache hierarchies are more effective etc, but the gist is the same.

I'd say Intel would still go that route with Nehalem even if AMD would not have done all that in Barcelona. Although I don't think Intel would really try to be as competitive as they are doing now without AMD, even though AMD is not crushing Intel at the moment. In such low competition scenario though, a Nehalem like chip could show up a lot later and that is not good either.

BTW in 2 years from now, what kind of role will Itanium have in the market place?

On Deneb, is there any word on pricing yet? What about RD890? I haven't heard anything of that chip set for a while and it even seems to be delayed to halfway through 2009.

[Shintai]

BTW in 2 years from now, what kind of role will Itanium have in the market place?

Still a billion$ business. Itanium tho will get alot cheaper platform. Since with QPI. You could mix an Itanium with say a X58 chipset.

And I think the goal is to make Xeons and Itaniums dropin compatible.

But gonna be fun to see the first 32nm Itanium (3?), Since they skip 45nm.

On Deneb, is there any word on pricing yet? What about RD890? I haven't heard anything of that chip set for a while and it even seems to be delayed to halfway through 2009

Same as current K10s I say. RD890 have never been 2008 as far as I know.

[Helmore]

Still a billion$ business. Itanium tho will get alot cheaper platform. Since with QPI. You could mix an Itanium with say a X58 chipset.

And I think the goal is to make Xeons and Itaniums dropin compatible.

But gonna be fun to see the first 32nm Itanium (3?), Since they skip 45nm.



Same as current K10s I say. RD890 have never been 2008 as far as I know.

Well you would almost forget that Intel also has Itanium next to all the other things it does, that's only because Itanium is not that much in the spotlight at the moment.

I didn't say RD890 was supposed to be launched this year, I heard it is supposed to launch in late Q1, probably Q2, but since they are going to use RD790 on their upcoming socket AM3 platform, it seemed like they are delaying it's launch. Could just be me though, I don't know .

[Drwho?]

I'd say Intel would still go that route with Nehalem even if AMD would not have done all that in Barcelona. Although I don't think Intel would really try to be as competitive as they are doing now without AMD, even though AMD is not crushing Intel at the moment. In such low competition scenario though, a Nehalem like chip could show up a lot later and that is not good either.

BTW in 2 years from now, what kind of role will Itanium have in the market place?

On Deneb, is there any word on pricing yet? What about RD890? I haven't heard anything of that chip set for a while and it even seems to be delayed to halfway through 2009.

Trust me, right now, Intel is running fast, even without AMD around, what you are talking about is just in the imagination of some people.

This is a very personal view, I lived through this as one of the P4 performance guy, I was an insider on this, trust me, if you look at the memory controler of Nehalem, you ll figure out that on many main stream programs, it does not help much yet, proving that AMD claims about memory controler were base less.

Just remember, Competition is good, but it is not always what drives people. In my case, I try to make the CPU faster, for same reason other people will tune up their Nissan GTR ... For the love of it, and I can tell you that Ronak and his team are the same, they love CPU architecture, and they work every day for making it faster and better.
Forget the fan boys discussions, it is all about "pimp my ride" thinking, get it fast, get it good, and please, give us the right to be proud of our archivements.

Let's keep the olympic spirit, that the best we can do for all of us.

"Olympism seeks to create a way of life based on the joy found in effort, the educational value of a good example and respect for universal fundamental ethical principles. "
Pierre de Coubertin

[duploxxx]

I didn't say RD890 was supposed to be launched this year, I heard it is supposed to launch in late Q1, probably Q2, but since they are going to use RD790 on their upcoming socket AM3 platform, it seemed like they are delaying it's launch. Could just be me though, I don't know .

finally some remarks back on topic

they are not going to use the rd790 for am3, it has always been called 890 and 880, but in the end the difference between 7 and 8 series are small features and off course ddr3 layout.

they just brought few am2+ package deneb into 2008, while am3 deneb was always slated for q12009 and so is the chipset. But you can put any am3 deneb in a am2+ mobo.

[Hornet331]

3 days till we see if shanghai delivers or not.