Not only Tejas, but Tulsa and Jayhawk canned aswel

[pauldh]

I know Intel is loaded, but canning all these expensive projects sure sounds bad for them. What is wrong with Intel? Hope they surprise us with something good in the near future.

ABIT IS7, P4 2.6C, 512MB Corsair TwinX PC3200LL, Radeon 9800 Pro, Santa Cruz, TruePower 430watt

 

[P4Man]

> I think that if they managed to tweak the core for clocks
>of 2.5-3Ghz and go for a full 533/667/800Mhz FSB, they
>might just manage to get a product out the door that's more
>than respectable.

All depends on *when*. Its just not likely intel will be able to produce anything near those specs before halfway 2006 on 65nm, by which time 3+ GHz dual cored 90nm K8's will have been on the market for quite a while.

Also, don't forget dual core=2xTDP. Now with a 2 GHz Dothan that would be quite manageable (2 GHz Dothan is 34W I think ?). Now up the FSB to 800 MHz, increase vCore to achieve higher clocks, and skew the process towards higher clocks, and you're looking at 50+W easily. Now increase the clock by 50% (your claim, I don't believe it will clock anywhere near that high), and your back at 70-80W. Dual core of that ? No way.

>If Dothan is indeed superior to Banias

What I've heard, Dothan==Banias+2MB cache. Its a straight shrink, minor relayout, added cache. So yes it will be "superior" but not by much at all.

>and they can crank two dothan cores on a processor by 2005,
>clock it above 2.5Ghz (which they'll probably try to do),

No way, not on 90nm, and not at >2,5 GHz. It would be hotter than precott, and not even faster on single threaded apps.

> increase FSB

Thats a given.

>and deploy the 64-bit extensions in Dothan,

Ah.. thats the $1M question. Not this year IMHO, and not before the end of next year either.

>then I don't think they're at all out of the game.

Depends what game. High performance desktop chips, I don't think a dual core dothan will be competitive with 90nm K8's.. At least it will be mixed bag, maybe good on multithreaded apps, but it will suck balls on anything singlethreaded or FP intensive. Not too mention I expect it to be 32 bit only for quite some time.

= The views stated herein are my personal views, and not necessarily the views of my wife. =

 

[Mephistopheles]

Well, it's just that Dothan is not just a relayout, as far as I know; there are a few architectural changes (don't ask me which ones, I don't remember... I'll try researching a little later)

I read it somewhere that dothan was supposed to deliver 10-15% faster performance per clock than Banias. Plus, Banias top speed is 1.7Ghz, and Dothan will be launched at 2Ghz, which is already an 18% clock speed gain; If indeed Dothan is 10-15% faster per clock too, then a 2Ghz Dothan would be around 29-35% faster than a 1.7Ghz Banias. This is <i>extremely optimistic</i>, mind you, but we'll be learning more on monday. This would be putting the 2Ghz Dothan amongst the high-end products...

<i><font color=red>You never change the existing reality by fighting it. Instead, create a new model that makes the old one obsolete</font color=red> - Buckminster Fuller </i><P ID="edit"><FONT SIZE=-1><EM>Edited by Mephistopheles on 05/08/04 03:46 PM.</EM></FONT></P>

 

[P4Man]

>I read it somewhere that dothan was supposed to deliver
>10-15% faster performance per clock than Banias

~10% sounds about right for the cache increase

>This would be putting the 2Ghz Dothan amongst the high-end
>products...

I wouldnt count on it. It will be a killer mobile chip, little doubt there, but against high end desktop chips, it won't be competitive. And that is against "old" 130 nm, single channel Athlon 64's running 32 bit software. Fast forward 6-9 months, and it will get ugly I think.

= The views stated herein are my personal views, and not necessarily the views of my wife. =

 

[Mephistopheles]

I wouldnt count on it. It will be a killer mobile chip, little doubt there, but against high end desktop chips, it won't be competitive.

Yes, well, not amongst the highest-end like 3.2+ A64 and P4s, but it might just manage to compete against the 2.8C... And that would be, agreed, obviously a killer mobile chip performance.

<i><font color=red>You never change the existing reality by fighting it. Instead, create a new model that makes the old one obsolete</font color=red> - Buckminster Fuller </i>

 

[upec]

If you like to use cheap PSU and Case then it will not cost you a lot. Most good ATX case have very good air flow. Why do we need a new stardard for PSU too? The BTX PSU offer better air flow too??

 

[Mephistopheles]

And if Dothan were to be made a desktop part, you still have to consider that i855 usually runs single-channel DDR266, DDR333 tops, with a 400Mhz FSB. So the total FSB bandwidth is 3.2GB/s, with 2.7GB/s memory bandwidth available, tops. Getting this all the way up to 6.4GB/s with a fully-fledged 800Mhz FSB and dual channel DDR400 (keep in mind that Intel pushed for DDR2-533! This will put them in the position of wanting to deploy the best possible memory system in the least amount of time... yikes).

If they can do that to the memory subsystem, I think a Dothan-like architecture would benefit a lot... But this depends heavily on how the future desktop architecture would react to extra bandwidth. Netburst needed it.

<i><font color=red>You never change the existing reality by fighting it. Instead, create a new model that makes the old one obsolete</font color=red> - Buckminster Fuller </i>

 

[juin]

No one have think that intel can just go back to NW core.Intel just cannot go dual core banias they will lose too much on desktop.All ISV have compile there software for Netburst.Just think about 3D studio max lightwave or Doom 3 engine that use SSE2 and is multitread for HT.

Technologie wise SSE 3 and TNI wont have much sense if use on banias.Ht is also not support.SSE2 SSE are not super scalar so long pipeligne cpu will performe better.Prescott new branch prediction didicate shift rotate Imul larger write buffer can be use on NW core.All DDR-2 tech got some sense if use with netburst that is not much affect by latency unlike K8 P6.Larger cache have allwayse help more P7 that K7-8 or P6.Intel will lose if they flush P7.

i need to change useur name.

 

[P4Man]

>ntel just cannot go dual core banias they will lose too
>much on desktop.All ISV have compile there software for
>Netburst.Just think about 3D studio max lightwave or Doom 3
>engine that use SSE2 and is multitread for HT.

Banias (and Dothan) support SSE2, and if an application benefits from HT, it will benefit from dualcore. It makes perfect sense. SSE3 should also be easy to incorporate, if AMD can do it with just a microcode update (no redesign), I'm sure intel will find it just as easy. SSE3 might well be in Dothan already, and if not, it shouldnt take long.

>Intel will lose if they flush P7.

Who said they would ? Tejas got canceled, but no one said Prescott will not live on, and might well be shrunk to 65nm. I'm also not sure if intel would lose more with a dual core Dothan or with a uber hot Prescott however. Time will tell.

= The views stated herein are my personal views, and not necessarily the views of my wife. =

 

[Snorkius]

The fifth pentuim.

They should call it Pentium squared.

P²...

A long long time ago, but I can still remember, how that music used to make me smile... <A HREF="http://www.nexus.hu/zonix/DIGGER.MID" target="_new"><b><font color=blue>Digger rulz</font color=blue></b></A>

 

[Mephistopheles]

Well, it should be possible to modify Dothan architecture in order to better make use of existing Netburst-optimized software. After all, they've got all blueprints imaginable for everything... They've also got resources. What they don't have a lot of is time. 😱

<i><font color=red>You never change the existing reality by fighting it. Instead, create a new model that makes the old one obsolete</font color=red> - Buckminster Fuller </i>

 

[endyen]

That, my friend is single core thinking. They have 3 X the time, as they have three teams working on it (Tejas, Tulsa and Jayhawk). The result may not be spectacular at release, but it will be "good enough", and will grow well.

 

[Mephistopheles]

Well, I suppose that what we're dealing with here is a company with a gargantuan amount of resources that are being reallocated and redirected for dual-core, 64-bit enabled processors by 2005 and the best possible performance. Question: what could possibly go wrong? Everything depends on it and they know that. Interesting things to come, heh. 😎

<i><font color=red>You never change the existing reality by fighting it. Instead, create a new model that makes the old one obsolete</font color=red> - Buckminster Fuller </i>

 

[juin]

SSE scale linear to clock speed Dothan core will never be close to P7 clock speed even lower to K8.Leaving intel with a lose situation there.In the middle of 2005 how many game and apps wont use SSE 1.


SSE 3 2 instruction were made to help HT they 9 others dont do much compare to SSE2.Tejas instruction was also made for a P7.

On standart X86 instruction and branch.2 alu vs 2 twin pump alu that reach 0.5 clock latency and reach very high clock speed.P6 have a bit faster FPU that P7 except that P7 reach much higher clock speed.

You can compare SPEC result for P4EE and Banias 1.7GHZ.There more that 50% difference with ICC.

HT vs dual core.Alpha team got the same question 7 year ago better to go with 2 small core or 1 big core.On a 8 alu 6 FPU risc 4 way SMT vs a 4*2 ALU 3*2 FPU.The big core have come to be faster in multitreads and single tread.Others limitation of dual core cache corenhency a level of cache must be share or interface before the FSB.AMD will go with a write back to others L2 from systemes request interface.(manage the HT link MCH and connect to the L2)Rumor say intel will go with a share L2 leaving to have to rework all the L2 cache that will be to long.So in the real world .......
2 core that push about 32 bit of data per cycle to write back
Each 3 cycle a L1 miss will occur
Writing new data from memory


ALL this work for what 2 year before nephalem show it face that feature a multi core SMT high single tread performance.
Just be realistic a bit USA R&D have a lot of probleme after merced Prescott.What intel will with them.They should do like ATI just go back to the basic Gefore 2 Radeon 9700 Radeon X800 Itanium 2 K8 NW.Just work on Basic thing.Go back to a 90NM 1 MB NW that can scale up to 4.4 GHZ will do the job that can be ready in 6 month and in a 1 and half add what go thing you have do with prescott and tejas and 64 Bit mode.That for desktop part on a DP and others P7 will not hold the candle vs K8.Either intel push itanium as fast as they can (impossible anyway) or call for a miracle.

i need to change useur name.

 

[P4Man]

>I read it somewhere that dothan was supposed to deliver
>10-15% faster performance per clock than Banias.

I guess THG's review (though it leaves much to be desired), proves your estimate wrong. Told you it was basically a straight shrink... +5% performance/clock, slightly lower TDP, significantly higher standby and deep sleep power consumption.. I'm not overwhelmed.

= The views stated herein are my personal views, and not necessarily the views of my wife. =

 

[Mephistopheles]

It would seem so. However, I still have to stress that that review is very badly representative of Dothan's performance. They should have used more benchmarks that stress the actual CPU. I mean, it is completely pointless to say "oh my, they have the same memory performance", since they use the same type of memory and same chipset. Or even to say "A 1.7Ghz Banias is as good as a 2Ghz Dothan in games!" - these benchmarks were obviously video-card stressing. The CPU has little to do with that.

Granted, I was too optimistic. But THG's review is as bad as it gets.

<i><font color=red>You never change the existing reality by fighting it. Instead, create a new model that makes the old one obsolete</font color=red> - Buckminster Fuller </i>