AMD Instanbul 6 core demoed

[thilanliyan]

http://www.techreport.com/articles.x/16448

 

[DrMrLordX]

Why did Constantinople get the works? That's nobody's business but the Turks'.

Very interesting nevertheless. It's dangerous for the article's author to assume that clockspeeds were reduced on the Istanbul system versus Shanghai, but the Stream results certainly are interesting. It would be nice to see results of a single-threaded benchmark to see if the test Istanbul system has a higher overall IPC than Shanghai, but again, without knowing the clockspeeds involved, that might not help so much.

 

[Cookie Monster]

48 core goodness with 8P configurations?

 

[DrMrLordX]

Since it's a drop-in for Socket F, sure, why not?

 

[Viditor]

To sum up...

1. Istanbul is 2P, 4P, and 8P like other Opterons
2. Drop-in for any socket F with split power planes
3. The 24-core Istanbul box hit about 42,000 MB/s
4. Has "HT Assist"...a snoop filter that functions to reduce traffic on socket-to-socket HyperTransport links by storing an index of all caches and preventing unnecessary coherency synchronization requests. The filter will be programable in the BIOS.
5. AMD estimates that 5% of their customers will use the chip as a drop-in.
6. Expected in H2 of this year

 

[exar333]

Originally posted by: DrMrLordX
Why did Constantinople get the works? That's nobody's business but the Turks'.

Very interesting nevertheless. It's dangerous for the article's author to assume that clockspeeds were reduced on the Istanbul system versus Shanghai, but the Stream results certainly are interesting. It would be nice to see results of a single-threaded benchmark to see if the test Istanbul system has a higher overall IPC than Shanghai, but again, without knowing the clockspeeds involved, that might not help so much.

I got your quote there DrMrLord! I was thinking the same thing myself...LOL

 

[Idontcare]

Originally posted by: DrMrLordX
It's dangerous for the article's author to assume that clockspeeds were reduced on the Istanbul system versus Shanghai

If AMD is targeting these things for drop-in replacements then that means the installations are already operating within a designed-in thermal budget.

To put 50% processing logic in a socket but keep the GHz the same would require the thermal output to rise considerably.

In other words to keep an Istanbul within say the 65W TDP class of a shanghai that it would be replacing would most likely require a decrease in core frequency relative to the 65W TDP shanghai being replaced since these are both 45nm chips.

 

[SunnyD]

I forget - is Istanbul supposed to be an MCM solution?

 

[Gikaseixas]

Originally posted by: Idontcare

Originally posted by: DrMrLordX
It's dangerous for the article's author to assume that clockspeeds were reduced on the Istanbul system versus Shanghai

If AMD is targeting these things for drop-in replacements then that means the installations are already operating within a designed-in thermal budget.

To put 50% processing logic in a socket but keep the GHz the same would require the thermal output to rise considerably.

In other words to keep an Istanbul within say the 65W TDP class of a shanghai that it would be replacing would most likely require a decrease in core frequency relative to the 65W TDP shanghai being replaced since these are both 45nm chips.

Good logic there but they could also have made new improvements here and there to achieve lower power envelops within a certain core speed.

Are they planning to release a desktop version and when? Too lazy to look around hehe

 

[exar333]

Will the six cores be sharing the same size cache as the previous 4 cores? If so, will this cause the cores to be more cache-starved?

 

[Idontcare]

Originally posted by: Gikaseixas
Good logic there but they could also have made new improvements here and there to achieve lower power envelops within a certain core speed.

That is why I included the caveat "since these are both 45nm chips".

Sure they can make some improvements, but your not going to see a near 50% reduction in the core-logic power consumption which would be necessary in order to enable a 50% increase in core-logic operating at the same frequency consuming the same amount of power.

That's called a node-shrink, not a new stepping.

I'm not trying to make stuff up here, just stating the obvious.

Do dual-core 65nm opterons operate at the same clockspeed and TDP as quad-core 65nm opterons?

If you was a betting man, what would you be inclined to assume will be the case when it comes to quad-core 45nm opterons versus hex-core 45nm opterons in terms of clockspeed and TDP?

 

[Markfw]

Originally posted by: Gikaseixas

Originally posted by: Idontcare

Originally posted by: DrMrLordX
It's dangerous for the article's author to assume that clockspeeds were reduced on the Istanbul system versus Shanghai

If AMD is targeting these things for drop-in replacements then that means the installations are already operating within a designed-in thermal budget.

To put 50% processing logic in a socket but keep the GHz the same would require the thermal output to rise considerably.

In other words to keep an Istanbul within say the 65W TDP class of a shanghai that it would be replacing would most likely require a decrease in core frequency relative to the 65W TDP shanghai being replaced since these are both 45nm chips.

Good logic there but they could also have made new improvements here and there to achieve lower power envelops within a certain core speed.

Are they planning to release a desktop version and when? Too lazy to look around hehe

This also assumes that socket F can go no higher than 65 watt. Is that true ? What if all socket F platforms are certified to say even as high as 125 watt per socket ?

I hate assumptions.

 

[Phynaz]

Originally posted by: Viditor

2. Drop-in for any socket F with split power planes

And BIOS support.

I smell another Phenom upgrade debacle.

 

[Flipped Gazelle]

Originally posted by: ExarKun333

Originally posted by: DrMrLordX
Why did Constantinople get the works? That's nobody's business but the Turks'.

Very interesting nevertheless. It's dangerous for the article's author to assume that clockspeeds were reduced on the Istanbul system versus Shanghai, but the Stream results certainly are interesting. It would be nice to see results of a single-threaded benchmark to see if the test Istanbul system has a higher overall IPC than Shanghai, but again, without knowing the clockspeeds involved, that might not help so much.

I got your quote there DrMrLord! I was thinking the same thing myself...LOL

Would have been a great time to introduce Socket 1453!

:beer:

 

[AstroManLuca]

Originally posted by: Flipped Gazelle

Originally posted by: ExarKun333

Originally posted by: DrMrLordX
Why did Constantinople get the works? That's nobody's business but the Turks'.

Very interesting nevertheless. It's dangerous for the article's author to assume that clockspeeds were reduced on the Istanbul system versus Shanghai, but the Stream results certainly are interesting. It would be nice to see results of a single-threaded benchmark to see if the test Istanbul system has a higher overall IPC than Shanghai, but again, without knowing the clockspeeds involved, that might not help so much.

I got your quote there DrMrLord! I was thinking the same thing myself...LOL

Would have been a great time to introduce Socket 1453!

:beer:

I had to Google that one.

 

[atari030]

I smell another Phenom upgrade debacle.

The only 'upgrade debacle' that might have happened would have been if some uninformed oaf tried to upgrade their system with a new processor without first investigating their BIOS support for it.

I've got an older ASUS MN2 SLI Deluxe board. AM2 socket. Nvidia nForce570 chipset. The original 5000+ BE I put in the system of course worked. With a BIOS upgrade I put in my current Phenom 9550 (AM2+) simple as pie. In fact, with this old *ss board I can put in a brand new PhII 940 and that will also work.

What I smell is FUD, caused more by the ignorant user than any issues with the vendor.

 

[Idontcare]

Originally posted by: Markfw900
This also assumes that socket F can go no higher than 65 watt. Is that true ? What if all socket F platforms are certified to say even as high as 125 watt per socket ?

I hate assumptions.

No my example was merely using 65W to make a point. The point would be if 125W quad-core exist and 125 hex-core exist then it is practically guaranteed those two 125W TDP cpu's will not be operating at the same clockspeed. That's all I am attempting to communicate.

 

[Cookie Monster]

Originally posted by: SunnyD
I forget - is Istanbul supposed to be an MCM solution?

Since your post got ignored

Here is the answer to your question.

Istanbul die shot

 

[dmens]

^

artist impression, not a real die shot

 

[Phynaz]

Originally posted by: atari030

I smell another Phenom upgrade debacle.

The only 'upgrade debacle' that might have happened would have been if some uninformed oaf tried to upgrade their system with a new processor without first investigating their BIOS support for it.

I've got an older ASUS MN2 SLI Deluxe board. AM2 socket. Nvidia nForce570 chipset. The original 5000+ BE I put in the system of course worked. With a BIOS upgrade I put in my current Phenom 9550 (AM2+) simple as pie. In fact, with this old *ss board I can put in a brand new PhII 940 and that will also work.

What I smell is FUD, caused more by the ignorant user than any issues with the vendor.

I suggest you do a little research before you start calling people ignorant.
It will help prevent you from looking foolish.

 

[Flipped Gazelle]

What Phenom upgrade debacle? I thought, for the most part, Phenom upgrades went fairly well as long as users waited for the proper BIOS to be made available for the mobo in question.

The B2 Phenoms were real dogs, though.

 

[atari030]

Two things:

1. Yes, my research is based on my own personal experience with the platform as I actually own the hardware in question. I'd hazard a guess that's more than you can say as you probably don't actually own any of this type of hardware, or at least, didn't experience this particular issue personally. My apologies in advance if that's not the case. If you did, I'd like to hear the details of the issues you had. If it's the silly 125 vs 140W thing on the 780 boards, that's a whole other side issue. However, just as with the BIOS update, user research of board compatibility for new CPUs is the issue at hand. IMO, one has no business upgrading PC components, especially CPUs, if you don't know that you need to look into these types of things first.

2. If you read again, the term 'ignorant user' was not directed at yourself. It was directed at those people that just assumed a particular CPU would work in their board without doing the proper research.

 

[Phynaz]

Originally posted by: Flipped Gazelle
What Phenom upgrade debacle? I thought, for the most part, Phenom upgrades went fairly well as long as users waited for the proper BIOS to be made available for the mobo in question.

The B2 Phenoms were real dogs, though.

After AMD spent a year or more shouting "drop in replacemnt", it turned out that BIOS updates were made for less than 50% of existing boards.

Combine that with the 140w power draw, and you have AMD with a lot of the proverbial egg on it face.

Then add in the Quadfather upgrades that never occured and you start to see a pattern.

I think there was a third case, but it's late Friday and the brain is shutting down

 

[Flipped Gazelle]

Originally posted by: Phynaz

Originally posted by: Flipped Gazelle
What Phenom upgrade debacle? I thought, for the most part, Phenom upgrades went fairly well as long as users waited for the proper BIOS to be made available for the mobo in question.

The B2 Phenoms were real dogs, though.

After AMD spent a year or more shouting "drop in replacemnt", it turned out that BIOS updates were made for less than 50% of existing boards.

Combine that with the 140w power draw, and you have AMD with a lot of the proverbial egg on it face.

Then add in the Quadfather upgrades that never occured and you start to see a pattern.

I think there was a third case, but it's late Friday and the brain is shutting down

Not AMD's fault re: the lack of BIOS updates. Where'd you get the "less than 50%" number? I remember there being problems early on, then the problems seemed to disappear into the ether.

AFAIK, the only Phenom @ 140w is the 9950. The 9600 and lower were 95w parts.

Quadfather was a real fringe thing, a petty attempt to draw attention away from Intel. It never should have, um, not happened.

 

[Viditor]

Originally posted by: Flipped Gazelle

Originally posted by: Phynaz

Originally posted by: Flipped Gazelle
What Phenom upgrade debacle? I thought, for the most part, Phenom upgrades went fairly well as long as users waited for the proper BIOS to be made available for the mobo in question.

The B2 Phenoms were real dogs, though.

After AMD spent a year or more shouting "drop in replacemnt", it turned out that BIOS updates were made for less than 50% of existing boards.

Combine that with the 140w power draw, and you have AMD with a lot of the proverbial egg on it face.

Then add in the Quadfather upgrades that never occured and you start to see a pattern.

I think there was a third case, but it's late Friday and the brain is shutting down

Not AMD's fault re: the lack of BIOS updates. Where'd you get the "less than 50%" number? I remember there being problems early on, then the problems seemed to disappear into the ether.

AFAIK, the only Phenom @ 140w is the 9950. The 9600 and lower were 95w parts.

Quadfather was a real fringe thing, a petty attempt to draw attention away from Intel. It never should have, um, not happened.

That's the info I have as well...The number of boards affected was fairly minimal and nowhere NEAR 50% (though I must admit that I don't know what the real number is).

Also, I don't see any reason to expect any problems at all on Istanbul as far as drop-ins are concerned. AMD has a really good track record on backwards compatability (certainly better than Intel's), and I don't see them reiterating the drop-in capability this close to launch if there was any possible problem. JMHO...