Pentium 4 1.6 Northwood overclocking
Introduction
As many of you are no doubt aware, Intel recently launched their revamped range of Pentium 4 desktop processors, codenamed Northwood. The differ from the original Pentium 4's, codenamed Willamette, in a number of important respects. Firstly, they will only be available in Socket 478 form. This appears to be Intel's preferred pin layout, one that should remain for the foreseeable future. Secondly, and perhaps of greater significance to us, is the change from a 0.18 micron to a 0.13 micron manufacturing process. The reduction is core size has a number of knock-on benefits for Intel. A smaller core allows a greater number of CPU's to be manufactured from a given wafer. The inherently smaller transistors also allow Intel to drop the core operating voltage from 1.75v to a scant 1.5v.
The Northwood is also outfitted with copper interconnects, something which the competing Athlon XP has been doing for some time. Not content with just a die shrink, Intel also decided to increase the L2 cache from 256k to 512k. The extra cache is the real clincher for the Northwood, one that should provide a discernable benefit in real applications.
The 0.18 micron Willamette Pentium 4 managed to hit 2Ghz without too much difficulty, the deep pipeline of the P4 made it relatively easy for Intel to scale to 2 Ghz. This bodes well for the potential overclockability of the Northwood CPU's. Today we'll see just how well. The Northwood was officially released in 2.0 Ghz and 2.2Ghz variants respectively. One by-product of a smaller manufacturing process is the increased number of CPU's per wafer. This in turn means that it's more cost-efficient for Intel to manufacture the Northwood. A decision was made to manufacture lower speed Northwoods so that system integrators could simply substitute the newer, cheaper to produce Northwoods, in place of the existing Willamettes. Thus saw the birth of the 1.6A and 1.8A P4, the A denoting their Northwood status.
Today we'll have a look at the baby of the bunch, the 1.6A. The fact that it's a 0.13 micron part should mean that it overclocks well, especially considering its relatively low rated clock-speed. Firstly we'll give you a quick run-down of our test system.
System Setup
- Abit BD7 I845D Motherboard
- Intel Pentium 4 1.6A 'Northwood' CPU
- 256Mb Samsung PC2700 DDR Memory Module (CAS2)
- Leadtek GeForce3 Ti500 64Mb
- 2 x 60Gb Maxtor D740X 7200rpm hard drives
- Liteon 32x12x40 CDRW
- Liteon 16x DVD
- D-Link 530TX NIC
- 21" Sony G500 FD monitor
- Windows XP Professional Build 2600.xpclient.010817-1148
- DetonatorXP 23.12 NVIDIA drivers
- Pifast v41.exe
- LAME MP3 Encoding, LAME 3.91 with RazorLame 1.15 front-end.
- XMpeg 4.2A DVD Encoding
- 3DMark 2001 Professional Second Edition
- SiSoftware Sandra v2002.2.8.64
- Serious Sam: The Second Encounter Demo
- Quake 3 v1.30
Benchmarks
Let's see how SiSoft Sandra sees our little Northwood running at stock speeds from both the CPU and memory standpoints.
As you can see, the CPU speed is being reported as 1.61Ghz. A little investigation shows that our motherboard, the Abit BD7, overclocks the FSB ever so slightly. At stock speeds we get an actual clock speed of 1608Mhz. The memory, being DDR, is run asynchronously to the FSB at stock speeds. Our numbers are in-line with the reference I845D board.
Now let's see just how well these Northwoods overclock. We had high expectations of this CPU and we were not disappointed. This little CPU raced across the 2Ghz mark with ease, passed the 2.2Ghz mark with default voltage and finally settled at a rock-solid 2.4Ghz with only a minor bump in voltage, All we can say is WOW. Your mileage may vary, but if this is any kind of indication of how well the lower Northwoods clock, you'll not be disappointed. The new Celeron 300A, most definitely. Our CPU, cooled only by the retail heatsink and fan, at 2400Mhz with 1.625v under load, hovered around 54c at full load. Let's have a look at how Sandra sees it.
It sure looks impressive. A 50% overclock with only a minimal rise in voltage (1.625v). The last time we saw overclocks like this was with the Athlon AXIA stepping from AMD. As you can see, the memory is now being run synchronously to the FSB.
Now let's see just how much impact the substantially increased clock-speed plays in our favourite benchmarks.
Firstly. a new addition to our benchmarking suite is Pifast. Put simply, it's a calculation of Pi to X decimal places. This benchmark is heavily FPU dependant, something the Willamette was not overly strong at. Let's see how we get on. We've set it to calculate Pi to 10 million places using the fastest method available.
Even at 1600Mhz, 130 seconds isn't too shabby. As a point of reference, an XP1500 / KT266A combination would take 122 seconds for the same task. The XP is noted for having an impressive number-crunching ability. The situation changes dramatically once the CPU is overclocked. 91 seconds is the result of another 600Mhz of raw CPU power allied to slighter greater memory throughput, impressive stuff.
We'll next turn our attention to another CPU-intensive application, MP3 encoding. We're using Lame 3.91 with both MMX / SSE2 optimisations to encode a 481MB custom WAV file into 128kb/s MP3.
As clock speeds are raised, we see an almost linear increase in performance. It appears that MP3 encoding is entirely FPU dependant. A 50% increase in a real-world application is nothing to be scoffed at, especially considering that only a minor bump in voltage was required. Numbers are almost meaningless when considered in isolation. Again, as a point of reference, our XP1500 / Iwill combination managed to complete the task in 161 seconds. The Athlon's impressive FPU at work again.
Another new entry to our benchmarking suite is DVD encoding, We know that a lot of you are interested in encoding, indeed, it's one of things that Intel tout about the P4. We're sure that you've seen the numerous adverts that allude to the proficiency of the Pentium 4 to all things media related.
We're using Xmpeg 2.0, a derivative of the popular Flask encoder, coupled with the Div X 3.20 codec. We've found this combination to be the most stable in our stress tests. Three Kings is the DVD of choice, it's mixture of action and dialogue make it an excellent benchmarking test. The DVD is encoded in full-screen format into YUV2 format. The black borders are cropped to save unnecessary encoding time.
We've almost reach the magical real-time framerate of 25fps at 1600Mhz, we positively crash through that barrier at 2400Mhz. It seems the Pentium 4's tag of media king is not totally without justification. 35fps at 2400Mhz is simply staggering, a figure that, until recently, was exclusively reserved for dual-processor equipped machines.
We see an almost linear increase in encoding performance, something that is always pleasing. We can surmise that it's not quite linear due to the lack of bandwidth afforded at 150 FSB.
We've demonstrated that an increased clock-speed brings about tangible gains in real-world performance. Let's focus our attention to another facet of computing that relies on huge CPU power, namely gaming.
3DMark 2001SE from the folks at MadOnion, although being an entirely synthetic benchmark, is an excellent tool for quantifying changes that occur when CPU speeds / memory throughput is increased. 3DMark 2001 was benchmarked at the default settings of 1024x768x32, vertical sync was disabled. Remember, we're using a Geforce 3 Ti500 clocked at 240/500, so our results may be differ from yours.
7615 marks from default clock speeds is pretty impressive in itself, it puts it around 400 marks ahead of a 1600Mhz Willamette P4. The extra L2 cache certainly seems to play a part in this test. As a comparison, we got 7683 marks with the Nforce / XP1500 combination.
It's pretty evident that 3DMark scales well with increased clock speed. At 2400Mhz, we achieve 8949 marks. This is the highest Ti500 default score we've seen at Hexus. Notice how the car and lobby tests particularly benefit from greater CPU speed. Nature on the other hand is almost completely video card limited.
Serious Sam 2 is the gorgeous new first-person-shooter from Croteam. It's is also an extremely useful benchmark, stressing the CPU and memory sub-system to to the limit. Here we're using the inbuilt Valley Of Jaguar timedemo.
We used the Normal graphics preferences here at 1024x768x32 and set the dem_bprofile variable to 1 (/dem_bprofile=1 at the console) so the game would generate the demo statistics for it.
Again, we can see a tangible gain in performance, one that is achieved through simply upping the speed on our processor. a 34% rise in performance is extremely impressive, it just goes to show how responsive Serious Sam 2 is to a change in basic speed. As a point of reference, our XP 1500 at 1660Mhz / 166fsb on the Iwill XP333R motherboard scored 67.2fps in the same test. The Pentium 4 had consistently under-performed in this benchmark until now. We surmise that the extra L2 cache is certainly helping the Northwood's cause.
We couldn't finish off a review without including the old-timer that is Quake 3. We've run the test at both 512 fastest and at 1024x768x32 Quality setting. 512 fastest is useful for illustrating the gains from a faster sub-system as it's not so card limited, 1024 Quality shows the gains that we can expect at a decent games-playing resolution. Onto the numbers.
The Pentium 4 has traditionally been the king-of-the-hill at Q3, the Northwood reinforces this view. We gain an extra 76 fps at 512 fastest when going from 1600Mhz to 2400Mhz, 344 fps is the fastest we've ever seen, . Perhaps more pleasing is the 44 fps gain at 1024 Quality, a resolution that many of us play at. It goes to show just how bandwidth hungry a game like Quake 3 is.
Conclusion
We set out to investigate the overclockablility of the 1.6Ghz Northwood, we came away surprised and pleased with the results. Although all of the ingredients for decent overclocking were present, a 0.13 micron fabrication, low voltage and a low rated clock-speed, we still didn't know just how well this CPU would do. 2.4Ghz is went well beyond our expectations.
From scouring various overclocking forums, we have gleaned that the average overclock for a 1.6Ghz Northwood is around 2.3Ghz, one that is usually achieved with a minor increase in voltage. It's probably safe to say that 2Ghz is almost a certainty, not bad for a processor that retails for around £140.
We've demonstrated that tangible gains are accrued when running at 2400Mhz, gains that make a difference in real-life applications and games. We've also seen just how well successful the 0.13 micron fabrication has been. Until recently, I'd have scoffed at the thought of a 1600Mhz processor running at 2400Mhz with just the retail air-cooling.
We couldn't let this processor go without seeing what it would do with more voltage. We upped the voltage to 1.85v (something that is not recommended unless you have excellent cooling) and began raising the clocks. Just have a look at what we got to.
