Intel's Northwood 2.0A: the Robin Hood processor

Introduction
YEAR 2001 remains one of the worst years in recent history for many IT companies. Intel can be counted in this category since there were more thorns than roses in Santa Clara. Its last quarter was a bit better but still Intel has a hard job in competing with its one really serious competitor.

P4s were slower than Athlons all the time and the company had serious problems with providing enough high end processors to market.

With the beginning of this year, things may change to Intel's advantage since Intel has some fresh horses to run.

The first good thing is the verification of the chipset for this platform with SiS 645 as the current leader and ALi and ATi launching their new chipsets soon. The rebellious Via chipset will soon be upgraded with the introduction of the P4X333 and this DDR fight will start in the next quarter.

Intel decided to upgrade its Processors line with a new processor code-named Northwood.

What is Northwood?
We can describe this processor as the next generation P4 processor that is now shrunk to 0.13µ (micron) and has more cache than the previous, Willamette, .18µ micron Pentium 4.

This new baby has 512 KB cache which is enough for even small servers not to mention regular PCs. The doubling of cache comes from the fact that the die of processor is much smaller than it used to be on 0.18 µm and now Intel has some more space to use it. The result of this extra space is that it can double the processor cache. Some applications need more cache than others and every little helps when you're using several applications at once - in the jargon laden language of the computer industry, multitasking.

Northwood 2.0A in Asus P4B-M The new processor uses a regular 478-pin CPU packages with the cache being the only real difference. The shrink to 0.13µ technology also uses 1.5 volts rather than the 1.8 volts the old Willamette lamp used.

This gives a lower CPU temperature and so potentially better overclocking features.

Via chipset and Northwood chips
We wanted to try this chip on a DDR platform since we don't have a Rambust motherboard to plug it into and so we tried plugging it into a rebellious motherboard that dared to use the rebellious Via Pentium 4 chipset. After hours of trying to make things work as they should and multiple disk errors and blue screens, we decided to give it a shout with different board. With the Via chipset it was impossible to run or install any operating systems because of multiple errors and blue screams of defiance (BSOD).

The only board we could find here in Bosnia and Herzegovina on short notice was the Asus P4B-M based on an SDRAM 845 chipset. (No, we don't know why Intel didn't send us an i845D board either - seeing as they sent us the chip.)

Asus P4B-M
This is a micro ATX board with 3 PCI and 1 CNR slot and 1AGP expansion sot with onboard audio. The Asus board has three memory slots that uses relatively slow SDRAM PC-133 memory that can take up to 3GB of memory.

One thing that I like about Asus boards is PC Probe - a small application that measures temperature of the processor and can prevent heath damage. Average temperature on this board was about 35-40 Celsius or 98 till 110 F.

This board was not an overclockers dream since it has just FSB adjustment trough jumpers with the predefined values 100, 105, 133 and 141 MHz. Only 105 can be used as non uniform and as an overclocking frequency but on this processor you will get just 100 additional megahurts. Since the board was borrowed from Bosnian computer retailer www.genelec.ba I just didn't want to take any chances and burn it. Here I would like to express my gratitude to this Bosnian computer store for understanding us. Do them a favour and visit their site.

With the P4B-M board you get additional USB ports and the board uses pretty common Award Bios.

Asus P4B-M micro-ATX The only bottleneck with this quality board is that it is an SDRAM based chipset.

How did we test?
We decided to test this board against one Athlon platform. I used Athlon 1400 MHz with old Thunderbird core and great Epox 8KHA+ that uses KT266A chipset. I did some Athlon Xp 1800 + benchmarks but since we where as always on short notice with this test I deiced not to run this result and to use it in future test.

On Pentium 4 platform we used:

Asus P4B-M P4 478 board (i845 chipset)
256 MB SDR PC 133 memory
Western digital WD 400AB hard drive 40 GB 5400
Liteon 16x10x40 CD- RW
Nvidia reference Geforce 3 TI 500
Onboard audio

Athlon PC used

Epox 8KHA+ with KT266A chipset
256 MB Hyundai DDR 266 CL 2.5
Western digital WD 400AB hard drive 40 GB 5400
Liteon 16x10x40 CD- RW
Nvidia reference Geforce 3 TI 500
Onboard audio

All the test ran on Windows XP.

We ran the same benchmarks that we used in our Nforce review that can be found here.

Benchmarketing
We used the same benchmarks like we used in Nforce review. I have decided to make this my default set of tests that I will use in my future board reviews.

Sisoft Sandra 2001 TE.... even though we prefer non-synthetic tests it seems that Sandra is becoming some kind of de facto standard. We used the CPU, multimedia CPU test as well as the memory test. This test showed us some surprises.

We used 3Dmark 2001 pro to test the CPU. We used just final score that is based on 4 games from the tests.

Quake 3 Timedemo 1 - de facto OpenGl standard test

I used one more synthetic test called Streamd that shows you how fast memory runs. It runs from a DOS box.

We used Adobe Photoshop and I took a picture with a digital camera and it was picture of ATI Radeon 8500 that was 254 KB JPEG on my carpet and did the following thing.

I chose one performance killing procedure: I the took Filter called Radial blur, set amount to 100 and chose best quality settings. I measured how fast Photoshop finished on each chipset and got some interesting numbers.

If you ask why I chose to do this, I remembered that to hell with numbers, people want to know how fast these things are in the real world with real applications such as Photoshop.

Some time ago I used this in one of my processor review for some other magazines. As a warning this test took more than 300 seconds on Athlon 1400. More than 5 minutes to apply one effect on picture make us wonder how fast processors need to be to do this in real time. Maybe 300 times who knows?

I used SPECviewperf6_1_2 professional graphic test to see which "platform" is fastest for this kind of job. Choo choo.

The rest of my job was based on playing games or playing demos that are called benchmarks that do the job for you and show how many FPS they can achieve with your Board and graphic card.

We used the Dronez game demo that includes a benchmark and which measures frames per seconds. Dronez was set to 512 pixel textures and 24 bit Z buffer.

Serious Sam is a great 3D game which supports ATI Truform. We set the game to quality mode during testing.

Return To Castle Wolfenstein is a great Quake 3 engine based game and I used the demo with benchmark options. I used two demos where one is CPU and the other one is graphic card intensive.

Last but not least is Aquamark, a hard to get benchmark here in Bosnia.

All graphic game tests were run in two modes 640x480 in 16 bit color and 1024x768 in 16 bit color where the first one tests CPUs more and the second graphics power.

16 bit color was used since it is not so graphic card intensive especially on this 640x480 mode.

Benchmarketing
Here is one interesting result. While here I haven't use competition Athlon processors still we have nice charts. The arithmetical logical unit (ALU) of this P4 processor is not much faster than the Athlon 1.33 that this Sandra uses. But when we talk about FPU intensive calculation powered with SSE2 can be faster on this test. This is the only test where we saw this processor in its real light.

This was first from the purposeful loss of Intel's processor against Athlon 1.4 GHz. In 3Dmark you can see that SDRAM is killing this processor. Athlon is king of this test with about 900 point's advantage. Just a reminder we are talking about the same graphic card. You can blame the SDRAM bottleneck for these results.

Surprise, surprises at Quake as well. I could not believe with my bare eyes that P4 processor which is supposed to be faster in Quake 3 performed so badly on this specific test. 32 FPS is really large advantage on fastest settings. 18 FPS in 10x7 is unexpected. The SDRAM I845 was a dumb idea and here is the proof of this. SDR was responsible for these results.

We ran this test from windows XP and have some theoretical numbers about throughput of chipsets. You can see that in numbers KT266A is about 40 per cent faster than i845. This is a big problem of this chipset and as I said earlier it makes Northwood look bad. We have roughly 300 MB less than KT266A on this test and that's what we had in other benchmarks.

Radial Blur in Photoshop. Don't be surprised -- Northwood is not two times better on this test. Lower is better and you can see that this chipset makes Northwood look really bad. AMD wins this one by about 50 per cent.

This Nvidia nice good looking pixel and shader test again needs more throughput than SDRAM can offer and result of 10 FPS less on this 512 cache P4. Both platform had tied result on both resolutions with really minor differences.

Even though we are just a few days away from Serious Sam Second Encounter this game is a nice test especially when we are talking about tons of enemies with main goal to kill you. Athlon is still faster but not as much as it was in previous tests.

This is a CPU intensive test and it does not depend much on resolution of the game. This is simulation of multiplayer game with lot of blood and killing. County of Cebit do not like this "controversial game". 5 frames on this test is huge thing.

This second demo from the same game is not so CPU intensive and it uses more GPU power. We see a big difference and 20 FPS difference between two different processors.

Miracles can happen. On 10x7 Northwood shows its potential and outperforms Athlon by just one FPS. If you have in mind that the P4 buss needs 3.2 GB/s and SDRAM can offer just over 1 GB/s this is a great victory.

Conclusion
Surprise, surprise is what I experienced during these tests. I expected the pure victory of the P4 2 GHz of this processor but I learned one interesting thing that you need a bit more than just a fast CPU to win all the tests. Remember that the Mac G3 was two times faster than the equivalent Intel P2 at the time but the difference was not so great when the computer architecture was measured. You need a lot of memory bandwidth to send data to your processor. For me, Intel P4 SDRAM sucks and I do not recommend it to my worst enemies if I had them. You need at least a DDR chipset to run this processor so i845 B is the slowest thing that you should plug your Northwood P4 into.

I didn't use Athlon XP 1800+ since the gap would be even bigger and there was no need to test it at this short time how I had to test this CPU.

These results require second round of testing and I can guarantee you that I will test some DDR platforms with this processor against the Athlon XP. A fair fight would be the SiS 645 DDR 333, i850 Rambus or one of the Via boards, but the first two were unavailable at test time.

The Pentium 4 is likely faster than you can see on my charts but in the i845 SDRAM environment that's the best you could expect from it. Northwood should not be used with this chipset - it's like subjecting your overpriced CPU to grievous bodily harm. µ