Im not a regular to this forum because I know I cant post with the level of knowledge that most of the folks in this forum has, but I do come here to read and to get away from the childish crap in the other forums. My question for the super eggheads is what is going to make the P4 northwood the super processor that Im reading and thinking it will be? I would like to hear some of your personal views, because things here often read better than on sites that Ive visited. Does anyone have any GOOD links with early test for this chip. Im a performance junkie who loves to have the lastest and greatest and I plan to get the northwood when its released. Thanks in advance for your input and thanks for keeping at least one forum on sharkey a joy to come to and learn from.

Edited because I alway go back and read my post on this forum LOL!!!!

The new Intel Pentium4 "Northwood" utilizes the new Intel 0.13m fabrication. The L1 cache is still the same, but the L2 cache has been upgraded to 512k, which should make a nice improv. Instead of having alluminum interconnects, the Northwood will have new copper interconnects, my guess is that the new core will scale from 2Ghz upto 3.6Ghz, although it would'nt be out of the question for it to hit 4Ghz - of course Intel isn't AMD. It'll use a new socket called mPGA478, which uses 55 more pin's than the original P4. Due to the die shrink, the power consumption is said to be at 1.3v, compared to 1.7, the change in die - size is also quite remarkable (meaning bigger server range), from 217mm down to 116mm.

It was rumoured that JacksonTech would make it into the P4 and its said to be featured in the Prestonia core (the successor to the Foster).

First of all, nobody should feel reluctant to post here. Questions are just as valuable as answers, perhaps even more so. If one person has a question you can be that others have the question as well.

Northwood isn't going to be a fundamentally different processor then the current P4. The primary difference (as Kunal stated above) is that it is made using a .13 Cu process. Intel has estimated (Sorry, I can't find the link) that this process will eventually allow the P4 to run 65% faster. This would put the P4 in the 3.5 GHz range.

The L2 cache has been bumped up from 256 KB to 512 KB. Some people disagree with me, but I feel this is a significant improvement for the P4. The Designers of the P4 gave it a large 128 Byte cache line. This helps make more efficient use of the FSB bandwidth, but the downside is it can decrease the efficiency of the cache. This means the P4 can only store 2048 lines in it 256 KB cache. The PIII by comparison can have 8192 cache lines in a 256 KB cache. If the memory is organized properly, this is not a problem since the total amount of cache is the same. This is not always the case however.

The final source of improvement in Northwood will come in the form of minor errata fixes. Every processor has errata. Things that do not work as the designers planned. Many of these "bugs" are "fixed" using micro-code to replace the defective circuit. Others cause the processor to have unnecessary bubbles in its pipeline, etc. None of these are visible to the end user, but they all take away from performance a little. Northwood will fix some of these and should perform a little better as a result.

Overall, I would expect a 5%-10% per clock performance increase over the current P4. Not a big boost, but when added to the frequency increase it is still nice. One thing to keep in mind is that this chip may be a good over clocker. It's FSB will (eventually) support 133 MHz (X4) operation sometime next year. I wouldn't be surprised if most 2.2 GHz Northwood's reached 2.6GHz - 2.8GHz, but this is pure speculation at this point.

Any news on who will be making an mpga478 mobo? (Aside from Intel of course)

edit: sorry double post

Abit already has a socket478 Mobo.
Its the KH7II-Raid.
See it @ www.abit.com.tw.

As far as the nothwood goes,
i dont think i could state it any better
than the guys above did.

quote:Originally posted by jester22c:
Any news on who will be making an mpga478 mobo? (Aside from Intel of course)

AFAIK the new socket is just a packaging change with some power and ground thrown in. Existing MB should need very little in the way of changes to use the new socket, so I expect everybody will have them before much longer. I also think there is a very strong possibility that we will see adaptors to allow the new chips to be used on older boards, so the socket 423 boards may not have to be thrown out yet.

quote:The Designers of the P4 gave it a large 128 Byte cache line. This helps make more efficient use of the FSB bandwidth, but the downside is it can decrease the efficiency of the cache

Another factor to be considered when looking at cache lines and therefore why Intel chose such a long L2 cache line is run length.

Run-length refers to the average number of consecutive (wrt memory locations) instructions that are executed before the PC (Program Counter) jumps elsewhere in memory.

On most RISC processors the run legth tends towards 4 words (16-bytes on a 32-bit processor). x86 tends towards a run length of 8 words (but not 32 bytes - more on this next).

A complication to this model is that RISC processors generally have fixed length instruction sets - almost always 32-bits. x86, on the other hand, has a variable length instruction set and instructions tend to be longer than 32-bits.

So therefore, if you have an average run-length of 8-words and your average word length (I'm guessing on this variable, btw!) is say 8-bytes (64-bits), then you would require a cache memory footprint of 64-bytes for storage.

As the run-length is an average it makes sense to store a cache line considerably longer than your average so that you reduce memory fetches from the same vicinity in memory (which, as you know has a latency and therefore performance penalty).

IMO the performance gain from having such a long cache line, based on x86's run length characteristics, more than makes up for the latency in the longer fetch.

Furthermore, I believe that Intel originally intended a large L3 cache which would have provided enough space so that the small number of available cache lines was not an issue.

I also understand the current P4 had its FPU castrated due to size and cost restraints. With the move to .13um they may put back the FPU unit the way it was originally designed. If they do, this should substantially help performance, even when the program doesn't support SSE/SSE2 instruction optimizations.

I just wish we could see that new Motorola technology come to CPUs soon. As it stands right now CPU manufacturers use manufacturing tweaks to extract as many MHz as possible out of a current design. When they reach the limit of the design they basically have two choices to continue ramping clock speed. They can shrink the die, which is time consuming, expensive, and has limitations as we approach the end of what we can do with current lithography technology. The other option is to increase the depth of the pipeline. The problem here is that IPC is reduced as the pipeline depth increases, thus requiring more speed just to make up for this. This is one of the main reasons that even AMD's procs can meet and beat P4s on benchmarks even though it is at a 1.1GB/sec FSB/memory bandwidth disadvantage, and 200-300MHz clock difference.

quote:Originally posted by Un4given:
I also understand the current P4 had its FPU castrated due to size and cost restraints. With the move to .13um they may put back the FPU unit the way it was originally designed. If they do, this should substantially help performance, even when the program doesn't support SSE/SSE2 instruction optimizations.

I very much doubt that this will happen. Over time, in the P4's target market, FP intensive application are going to be more and more constrained by memory performance and less constrained by FPU throughput. This makes the second FPU expendable.

While were on the Northwood topic:

So exactly how much smaller is the core of the Northwood compared to the Willy, and how many transistors does it have now?

How much heat will it generate?

Will the Northwood still have a price advantage over the Willy, considering its L2 is doubled, and assuming the Yields are comparable?

Final question. Are there any other less important performance tweaks Intel performed to the Northwood that we know about?

Thanks alot!

Livecoma: While were on the Northwood topic:

So exactly how much smaller is the core of the Northwood compared to the Willy, and how many transistors does it have now?

Northwood is estimated at 140mm^2. Willamette is 217mm^2. Northwood will have more transistors, due to the larger L2 cache.

How much heat will it generate?

Substantially less. I am guessing at 2.0GHz, it will generate ~40-45W, compared to 75.3W for Willamette.

Will the Northwood still have a price advantage over the Willy, considering its L2 is doubled, and assuming the Yields are comparable?

Northwood will cost less to make, due to a smaller die. However, expect a larger premium, since Intel can claim performance enhancements.

Final question. Are there any other less important performance tweaks Intel performed to the Northwood that we know about?

None that have been announced. However, I am confident that the extra L2 cache will have a >5% peformance improvement alone, and depending on the application, there may be other performance enhancements as well.

Arcadian

I agree

I agree

halflife3d-

Please stop spamming these boards, if you don't have something important to say, please refrain from posting (twice) "I agree" I have seen you so this in other parts of the board...that is called flooding you will get banned.

Which is the version of the P4 that will have "HyperThreading" or the use of two dies on one chip?

HyperThreading, or its former name, JacksonTech, will be present in the 0.13Cu P4 server line. The current server line of P4 chis, called P4-Xeon, which have the codename Foster, is virtually identical to the P4 line. The successor to the Foster(0.18u) is the Prestonia line of P4-Xeon chips,(0.13Cu). This line will have many more differences. It will include the HyperThreading feature. Due to the die size shrink, we will finally start to see larger cache on the P4-Xeon's. Prestonia's with higher cache are expected to start at 1.6Ghz.

quote:Originally posted by Kunal:
HyperThreading, or its former name, JacksonTech, will be present in the 0.13Cu P4 server line. The current server line of P4 chips, called P4-Xeon, which have the codename Foster, is virtually identical to the P4 line. The successor to the Foster(0.18u) is the Prestonia line of P4-Xeon chips,(0.13Cu). This line will have many more differences. It will include the HyperThreading feature. Due to the die size shrink, we will finally start to see larger cache on the P4-Xeon's. Prestonia's with higher cache are expected to start at 1.6Ghz.

Actually, you're 95% correct. All except for the part about large cache Prestonia cores. Before that, I believe there will be a large cache Foster core, with 512KB or 1MB of on-die L3 cache. Hyperthreading will also be enabled, but these chips will be on the .18u Aluminum process, like the current (P4)-Xeon. These are set to launch in late Q4 or Q1 at 1.6GHz or better (also the first chips with Hyperthreading).

Prestonia will be the first Xeon chip at .13u, and it will have 512KB of L2 and Hyperthreading, but no L3 cache. Their launch is set for Q1, too.

Finally, the large cache Prestonia is called Gallatin, and that will come in the second half of 2002, and feature 1MB or 2MB of on-die L3 cache, as well as Hyperthreading, and 512KB of L2 cache. These "Large Cache Prestonia" chips will clock much, much higher than 1.6GHz. I would probably guess 2.6GHz on the conservative side. Gallatin will likely compete with Sledgehammer, if AMD is on time with that chip.

Arcadian

quote:Originally posted by Abit400:
Which is the version of the P4 that will have "HyperThreading" or the use of two dies on one chip?

Hyper threading does not mean the chip has two cores. What happens is hyper threading is the core itself can execute instructions from two different threads at the same time. The various execution resources like ALU's, FPU's, SIMD units, etc are issued instructions from either thread depending on what execution resources are free and what instructions can be executed.

The reason it looks like two cores is that the OS (Win2K, Linux, etc) wasn't designed to run a chip that could handle more then one thread at once. Therefore some extra hardware was included to make it "look" like something the OS could recognize.

the 478 pin p4 has 256k of l2 cache not 512 like someone posted here.the 478 pin is using the sdram which doesn't have the bandwidth that rdram has and will cause some bottlenecking because of this.when they come out with a motherboard chipset that will use ddr memory things will improve quite a bit.i would go with a amd 1.4 t-bird and wait till intel gets it's act together.tom's hardware has a review of the 845 chipset that these new p4 -478 pin cpu's have.the 0.13 micron is going to be a great chip once they get the memory where it will run neck and neck with the processor but that will be next year.by then they should have a 64 bit cpu that will be blazing fast compared to now.we have serial ata coming out and ata 133-7200 rpm harddrive,usb 2.0 ,agp 8x and more firewire components and also wireless highspeed internet service and then there is internet 2 which will have a lot more bandwidth than internet 1 which we use now.when internet 2 is released we will be able to download video a lot faster and the quality of the video will be hdtv resolution .the future holds a lot of very exciting computer technologies that is going to be so much better and faster than what we have now that it is going to be unbelieveable.hang on to your hat.
tommy

quote:Originally posted by butcherbeard:
Im not a regular to this forum because I know I cant post with the level of knowledge that most of the folks in this forum has, but I do come here to read and to get away from the childish crap in the other forums. My question for the super eggheads is what is going to make the P4 northwood the super processor that Im reading and thinking it will be? I would like to hear some of your personal views, because things here often read better than on sites that Ive visited. Does anyone have any GOOD links with early test for this chip. Im a performance junkie who loves to have the lastest and greatest and I plan to get the northwood when its released. Thanks in advance for your input and thanks for keeping at least one forum on sharkey a joy to come to and learn from.

Edited because I alway go back and read my post on this forum LOL!!!!

OK, let's try this again. There are two released versions of the Pentium4, with a third around the corner

Pentium4 423pin Willamette (0.18)
Pentium4 478pin Willamette (0.18)
Pentium4 478pin Northwood (0.13)

Now, the 423pin P4 Willamette will work with existing i850 RDRAM chipsets and VIA's P4X266 DDr chipset. This chip as 256 L2 cache.

The 478pin P4 Willamette will work with newer revisions of the existing i850 chipset, which have the 478socket. These chip's will also work in i845 boards. I do not see any reasons why these chips won't work in future northwood boards such as Tulloch. This chip also has 256 L2, just more pins.

Finally theres the 478pin P4 Northwood. In terms of compatability, look at the boards the 478 willamette works on. This completely new core for the P4 DOES have 512 L2 cache.

Once again, 478 pin P4's are not restricted to SD-RAM, they work on DDr and RD.

quote:Originally posted by tparker:
the 478 pin p4 has 256k of l2 cache not 512 like someone posted here.the 478 pin is using the sdram which doesn't have the bandwidth that rdram has and will cause some bottlenecking because of this.when they come out with a motherboard chipset that will use ddr memory things will improve quite a bit.i would go with a amd 1.4 t-bird and wait till intel gets it's act together.tom's hardware has a review of the 845 chipset that these new p4 -478 pin cpu's have.the 0.13 micron is going to be a great chip once they get the memory where it will run neck and neck with the processor but that will be next year.by then they should have a 64 bit cpu that will be blazing fast compared to now.we have serial ata coming out and ata 133-7200 rpm harddrive,usb 2.0 ,agp 8x and more firewire components and also wireless highspeed internet service and then there is internet 2 which will have a lot more bandwidth than internet 1 which we use now.when internet 2 is released we will be able to download video a lot faster and the quality of the video will be hdtv resolution .the future holds a lot of very exciting computer technologies that is going to be so much better and faster than what we have now that it is going to be unbelieveable.hang on to your hat.
tommy

Just a few corections.
Socket 478 boards are out with both SDram and RBDram configurations.

Socket 478 chips that are out right now are the same as socket 423 chips.

Northwood will be on socket 478, but will also offer other changes. These changes include .13 fab with cu interconects. It will have 512kb l2 cache. Other than that it will only offer a few minor bug fixes.

As far memory bandwidth the current RBDram configurations are the best out, and future RBDram sytems will maintain that position. DDR will be offered with the P4, but RBDram will still be the top performer.

About 64 bit processors. Intel has no current plans for a 64bit chip in the desktop market that i know of. AMD has bet the farm on there 64/32 bit hammer, and may buy the farm if it fails. Currently 64 bit chips offer no performance gains unless the entire architecture is changed. AMD has not done this, so i dont see the point in there chip.

As far as other things comming, i would say that has little to do with this thread, so i wont comment.

To answer the original question.
The Northwood will be a good chip, because it will offer slight ipc performance gains and up to 65% greater clock speeds. In other words it will be a great overclockable chip and keep the stability and durability intel is known for.

quote:Originally posted by tparker:
the 478 pin p4 has 256k of l2 cache not 512 like someone posted here.the 478 pin is using the sdram which doesn't have the bandwidth that rdram has and will cause some bottlenecking because of this.when they come out with a motherboard chipset that will use ddr memory things will improve quite a bit.i would go with a amd 1.4 t-bird and wait till intel gets it's act together.tom's hardware has a review of the 845 chipset that these new p4 -478 pin cpu's have.the 0.13 micron is going to be a great chip once they get the memory where it will run neck and neck with the processor but that will be next year.by then they should have a 64 bit cpu that will be blazing fast compared to now.we have serial ata coming out and ata 133-7200 rpm harddrive,usb 2.0 ,agp 8x and more firewire components and also wireless highspeed internet service and then there is internet 2 which will have a lot more bandwidth than internet 1 which we use now.when internet 2 is released we will be able to download video a lot faster and the quality of the video will be hdtv resolution .the future holds a lot of very exciting computer technologies that is going to be so much better and faster than what we have now that it is going to be unbelieveable.hang on to your hat.
tommy

LostGod covered most of the points, but I'd like to add that the P4 Northwood will also have some non-bug fix related tweaks.

As far as AGP 8x USB 2.0 and ATA133, the biggest issue in the entire graphics subunit right now is the on board memory bus speed (that is the clock of the video RAM). AGP 8x would offer extremely marginal, if any gains at all in realworld performance.

Also, the same applies for ATA133. Devices now can barely offer the bandwidth offered by ATA66, much less ATA100. SCSI can more effectively use the bandwidth of the newer (80,160 and now even 320) interfaces due to its design.

And a final note, the USB 2.0 vs. Firewire (both 400 and onwards at 800 and perhaps 1600Mbps speeds) war seems interesting. It might end up where we are now again with the cheaper USB 2.0 solutions becoming the standard for mainstream applications (which with increased bandwidth could accomidate better webcams, A/V devices etc.) while Firewire at 400 and 800Mbps flavors will again occupy the higher-end niche. Intel is backing USB 2.0 and with XP supporting both technolgoies, I can't see either failing, but rather becoming successful mutually exclusive of each other much like USB 1.0/1.1 and IEEE1394 are now.