I am a little skeptical about Conroe

[Viditor]

Obviously it's nowhere near 91% by volume, but it probably is close to 70% by revenue (remember that includes servers)...if you take away Enterprise servers (since they aren't a Conroe or 9xx system), it's probably closer to 40% by revenue.
Government is one of largest segments (not quite at the business level, but close), followed by consumers.

But all of that is irrelevant for these purposes...
Desktop systems (for business or otherwise) don't go through a lengthy qualifying process. Servers certainly do, but that isn't what we're talking about here.
The only reasons I can see that Intel is upgrading 9xx are:
1. Replacement of the Celeron line with 9xx
2. The full ramp of Conroe hasn't even started yet, and it is expected to take at least a year. There is a very long crossover period ahead where the 9xx will still be Intel's "sweet spot" due to lack of availability
3. The majority of computer buyers go for the low end

 

[IntelUser2000]

Obviously it's nowhere near 91% by volume, but it probably is close to 70% by revenue (remember that includes servers)...if you take away Enterprise servers (since they aren't a Conroe or 9xx system), it's probably closer to 40% by revenue.
Government is one of largest segments (not quite at the business level, but close), followed by consumers.

But all of that is irrelevant for these purposes...
Desktop systems (for business or otherwise) don't go through a lengthy qualifying process. Servers certainly do, but that isn't what we're talking about here.
The only reasons I can see that Intel is upgrading 9xx are:
1. Replacement of the Celeron line with 9xx
2. The full ramp of Conroe hasn't even started yet, and it is expected to take at least a year. There is a very long crossover period ahead where the 9xx will still be Intel's "sweet spot" due to lack of availability
3. The majority of computer buyers go for the low end

Add to that, Intel wouldn't want to have excess inventory of CPUs. I am sure 9xx CPUs would still be demanded considering how many people are still skeptical about Conroe. Also there are people who can get 9xx CPUs but probably not Conroe, which adds to another set of buyers. Plus, not all buyers are knowledgeable enough to wait for Conroe.

 

[Munky]

Originally posted by: IntelUser2000

. A possible reason for this may be that Intel needed that clock to actually beat AMD across the board (ie, maybe AMD's FPU was still superior at lower than 2.67GHz) because it wanted to show that its technology was superior in every way possible.

According to released information, the FPU is much better than the one AMD Athlon 64 has so that's no problem. Also no need to have a CPU that performs 5% better than previous fastest CPU when it can be clocked 20-30% higher(if at 2.167GHz as you suggested).

Link to that released information? AFAIK, the A64 has 3 FPU's, and the Conroe core has 2 FPU's.

 

[Accord99]

Originally posted by: munky
Link to that released information? AFAIK, the A64 has 3 FPU's, and the Conroe core has 2 FPU's.

http://realworldtech.com/page.cfm?ArticleID=RWT030906143144&p=6

Conroe is 2-3X more powerful than K8 in FP per cycle.

 

[imported_Questar]

Obviously it's nowhere near 91% by volume, but it probably is close to 70% by revenue (remember that includes servers)...if you take away Enterprise servers (since they aren't a Conroe or 9xx system), it's probably closer to 40% by revenue.

Wow, dude, you a really off on that one.

Retail sales are 9% of the market. While some of the direct market is consumer, the vast majority is business. To think the business market is 40% is ignorant of the volumes of equipment enterprises buy.

Desktop systems (for business or otherwise) don't go through a lengthy qualifying process.

Wrong again. I have already begun qualiying systems that are not annouced yet. The earliest I would be buying these for production refresh operations would be Q1 '07.

2. The full ramp of Conroe hasn't even started yet, and it is expected to take at least a year. There is a very long crossover period ahead where the 9xx will still be Intel's "sweet spot" due to lack of availability

There's no such thing as a "full ramp". Conroe is being produced on production lines today. They are shipping production chips today.

I assume you mean that Conroe hasn't been ramped to full production volumes yet. Yes, that is true, there are not three fabs shipping Conroe today.

 

[swtethan]

they have am2 benches if you wanna see now

 

[imported_Questar]

Source for that stat please?

http://news.com.com/Strike%20three%20fo...2100-1014_3-6045843.html?tag=nefd.top?

 

[Munky]

Originally posted by: Accord99

Originally posted by: munky
Link to that released information? AFAIK, the A64 has 3 FPU's, and the Conroe core has 2 FPU's.

http://realworldtech.com/page.cfm?ArticleID=RWT030906143144&p=6

Conroe is 2-3X more powerful than K8 in FP per cycle.

That link doent mention the K8 anywhere. From looking at the diagrams, I see 3 ports and only 2 of them have an FPU. A port can only issue a command to one of its units at a time, so when doing raw FPU math a Core cpu still has to use 2 FPU's compared to 3 of a K8. But since the FPU's are 128 bit, that may even out the field, or even bring a slight advantage to the Core, but I still dont see any technical basis of the claim that it has 2-3x the FPU performance of a K8.

 

[Hard Ball]

Originally posted by: munky

Originally posted by: Accord99

Originally posted by: munky
Link to that released information? AFAIK, the A64 has 3 FPU's, and the Conroe core has 2 FPU's.

http://realworldtech.com/page.cfm?ArticleID=RWT030906143144&p=6

Conroe is 2-3X more powerful than K8 in FP per cycle.

That link doent mention the K8 anywhere. From looking at the diagrams, I see 3 ports and only 2 of them have an FPU. A port can only issue a command to one of its units at a time, so when doing raw FPU math a Core cpu still has to use 2 FPU's compared to 3 of a K8. But since the FPU's are 128 bit, that may even out the field, or even bring a slight advantage to the Core, but I still dont see any technical basis of the claim that it has 2-3x the FPU performance of a K8.

I would like to run some MunkyMark SpecFP to see what that sucker is really capable of.

 

[Accord99]

Originally posted by: munky
That link doent mention the K8 anywhere. From looking at the diagrams, I see 3 ports and only 2 of them have an FPU. A port can only issue a command to one of its units at a time, so when doing raw FPU math a Core cpu still has to use 2 FPU's compared to 3 of a K8. But since the FPU's are 128 bit, that may even out the field, or even bring a slight advantage to the Core, but I still dont see any technical basis of the claim that it has 2-3x the FPU performance of a K8.

Conroe can do 1 128-bit packed SSE add and 1 128-bit SSE multiply per cycle. That gives 4 double precision flops/cycle. The K8 can only do 2 flops/cycle. Conroe can also do a 128-bit store and 128-bit load in the same cycle.

 

[Munky]

Originally posted by: Accord99

Originally posted by: munky
That link doent mention the K8 anywhere. From looking at the diagrams, I see 3 ports and only 2 of them have an FPU. A port can only issue a command to one of its units at a time, so when doing raw FPU math a Core cpu still has to use 2 FPU's compared to 3 of a K8. But since the FPU's are 128 bit, that may even out the field, or even bring a slight advantage to the Core, but I still dont see any technical basis of the claim that it has 2-3x the FPU performance of a K8.

Conroe can do 1 128-bit packed SSE add and 1 128-bit SSE multiply per cycle. That gives 4 double precision flops/cycle. The K8 can only do 2 flops/cycle. Conroe can also do a 128-bit store and 128-bit load in the same cycle.

The K8 has 3 FPU's - FMUL, FADD, FSTORE, which already gives it 3 FLOPS per cycle. I'll have to look at the documentation more closely to examine the details, but based from what I've seen, the Core would have a 2x theoretical advantage at most because it works with 128-bit data in its FPU's.

 

[Accord99]

Originally posted by: munky
The K8 has 3 FPU's - FMUL, FADD, FSTORE, which already gives it 3 FLOPS per cycle. I'll have to look at the documentation more closely to examine the details, but based from what I've seen, the Core would have a 2x theoretical advantage at most because it works with 128-bit data in its FPU's.

A store is not considered a floating point calculation. Conroe can do a 128-bit store and load in the same cycle, and has considerably higher bandwidth cache as well.

 

[sandorski]

As time goes on I've become increasingly skeptical about any Hardware Previews. It's always best to wait for actual released parts before declaring anything.

 

[Hard Ball]

Originally posted by: Accord99

Originally posted by: munky
That link doent mention the K8 anywhere. From looking at the diagrams, I see 3 ports and only 2 of them have an FPU. A port can only issue a command to one of its units at a time, so when doing raw FPU math a Core cpu still has to use 2 FPU's compared to 3 of a K8. But since the FPU's are 128 bit, that may even out the field, or even bring a slight advantage to the Core, but I still dont see any technical basis of the claim that it has 2-3x the FPU performance of a K8.

Conroe can do 1 128-bit packed SSE add and 1 128-bit SSE multiply per cycle. That gives 4 double precision flops/cycle. The K8 can only do 2 flops/cycle. Conroe can also do a 128-bit store and 128-bit load in the same cycle.

err, NO,

FADD and VFADD share the same logic in CMA, and FMUL and VFMUL share the same; these units also do FMOV; and LD and ST unit is shared with arith-logic and vector ALUs.

 

[Accord99]

Originally posted by: Hard Ball
err, NO,

FADD and VFADD share the same logic in CMA, and FMUL and VFMUL share the same; these units also do FMOV; and LD and ST unit is shared with arith-logic and vector ALUs.

See page 29 on the IDF presentation:

https://www35.cplan.com/cbi_export/PS_MATS001_278703_125-1_FIN_v1.pdf

login: idf
password: spring2006

 

[Hard Ball]

Originally posted by: Accord99

Originally posted by: Hard Ball
err, NO,

FADD and VFADD share the same logic in CMA, and FMUL and VFMUL share the same; these units also do FMOV; and LD and ST unit is shared with arith-logic and vector ALUs.

See page 29 on the IDF presentation:

<a target=_blank class=ftalternatingbarlinklarge href="https://www35.cplan.com/cbi_export/PS_MATS001_278703_125-1_FIN_v1.pdf">https://www35.cplan.com/cbi_export/PS_MATS001_278703_125-1_FIN_v1.pdf</a>

login: idf
password: spring2006

That slide shows little, if anything of value about the datapath or the composition of the number of execuation units. And how exactly does that contradict any of the things that I just said?

 

[Accord99]

Originally posted by: Hard Ball
That slide shows little, if anything of value about the datapath or the composition of the number of execuation units. And how exactly does that contradict any of the things that I just said?

What are you trying to say exactly and how does it disprove that Conroe can do 4 DP flops/cycle plus 128-bit packed store and 128-bit packed load?

 

[Hard Ball]

Originally posted by: Accord99

Originally posted by: Hard Ball
That slide shows little, if anything of value about the datapath or the composition of the number of execuation units. And how exactly does that contradict any of the things that I just said?

What are you trying to say exactly and how does it disprove that Conroe can do 4 DP flops/cycle plus 128-bit packed store and 128-bit packed load?

That's not what I'm trying to say at all.

What I was saying is that :
-FADD and VFADD sharesthe same logic, which is capable of scalar (32|64) OR 4X32 OR 2X64; which is also capable of FMOV by the right FS code in the reg file.
-FMUL and VFMUL share the same logic, which is capable of scalar (32|64) OR 4X32 OR 2X64; and is also capable of FMOV.
-LD and ST are handled by a separate datapath and is shared by scalar AL/FP and Vector.

 

[Viditor]

Originally posted by: Questar

Obviously it's nowhere near 91% by volume, but it probably is close to 70% by revenue (remember that includes servers)...if you take away Enterprise servers (since they aren't a Conroe or 9xx system), it's probably closer to 40% by revenue.

Wow, dude, you a really off on that one.

Retail sales are 9% of the market. While some of the direct market is consumer, the vast majority is business. To think the business market is 40% is ignorant of the volumes of equipment enterprises buy.

If as you say business is 91% and retail is 9% (BTW, 9% is the US retail percentage of the total global market, not the total percentage of retail globally)...that leaves 0% for consumer-direct (which is larger than retail alone), government, education, scientific institutions, etc...

Desktop systems (for business or otherwise) don't go through a lengthy qualifying process.

Wrong again. I have already begun qualiying systems that are not annouced yet. The earliest I would be buying these for production refresh operations would be Q1 '07.

I see...then might I ask:
1. who or what you work for?
2. why don't you use SIPP or CSIP programs like most businesses?
3. What exactly do you "qualify" on a desktop business platform that takes over a year??

2. The full ramp of Conroe hasn't even started yet, and it is expected to take at least a year. There is a very long crossover period ahead where the 9xx will still be Intel's "sweet spot" due to lack of availability

There's no such thing as a "full ramp". Conroe is being produced on production lines today. They are shipping production chips today.

I assume you mean that Conroe hasn't been ramped to full production volumes yet. Yes, that is true, there are not three fabs shipping Conroe today.

Yes, a "full ramp" is ramping to 100% production volumes (at least it's been that since I began studying it in the 90's). However, that doesn't just mean that 3 Fabs are shipping...it means that all Conroe capable Fabs are producing at full capacity.

 

[Duvie]

Dont forget business reports a few weeks back as Intel was announcing some less then rosy news...It seems they are having an issue with a large supply of chips in the warehouse...not being sold...They stated this amount was more then waht was needed for general warranty replacement...

I also think we wont see the speed grades like anything over 3ghz for awhile (2.66 may be a push as well and not liklely at the prices being speculated)...i think they will want to move some of these stockpiled smithfields and preslers before bringing in chips that will beat their next line of chips by 30-40% and cost significantly less then the XE line they just launched (955-965s)....

Not smart business sense...

 

[Childs]

Intel Core Duo's at 2Ghz are faster than an 3800+ X2, and sit in between a X2 4200 and X2 4400. Hopefully this afternoon I can see how it does against an Opteron this afternoon. If the Core Duo is faster clock for clock than the X2, then I have no doubt that Conroe will be even better. Core Duo's arent necessarily cheap tho, so its hard to imagine the $316 price tag for something that bests anything AMD has by 20%. Doesnt seem to make much business sense, unless its old school gangsta vendetta sh1t.

 

[designit]

Says: Bob Valentin: Architect
This is Illegal and he can be criminally charged for calling himself "Architect"
Architect legal definition is Architectural (designing buildings, structures, homes, etc)
If he means "electrical circuitry designer" then he must use the word " Electrical microchip architecure engineer"
Another one of those stuck up guy who wished to be called "Architect" and who envies the real "Architects".
Do you call a circuit board assemblers (the persons who solders things) "Surgeon"?
NO.

 

[ZOXXO]

Originally posted by: Duvie
Dont forget business reports a few weeks back as Intel was announcing some less then rosy news...It seems they are having an issue with a large supply of chips in the warehouse...not being sold...They stated this amount was more then waht was needed for general warranty replacement...

...snip...

Not smart business sense...

Link?

 

[Viditor]

Originally posted by: ZOXXO

Originally posted by: Duvie
Dont forget business reports a few weeks back as Intel was announcing some less then rosy news...It seems they are having an issue with a large supply of chips in the warehouse...not being sold...They stated this amount was more then waht was needed for general warranty replacement...

...snip...

Not smart business sense...

Link?

Here's one, though you could do a Google search for "Intel warns" and find many more...

 

[Viditor]

Originally posted by: Childs
Intel Core Duo's at 2Ghz are faster than an 3800+ X2, and sit in between a X2 4200 and X2 4400. Hopefully this afternoon I can see how it does against an Opteron this afternoon. If the Core Duo is faster clock for clock than the X2, then I have no doubt that Conroe will be even better. Core Duo's arent necessarily cheap tho, so its hard to imagine the $316 price tag for something that bests anything AMD has by 20%. Doesnt seem to make much business sense, unless its old school gangsta vendetta sh1t.

Huh? Core Duo performs the same in most, worse in some, and better in very few benchmarks vs the X2 3800.
AT article