Just hours before the kick off of Super Bowl XXXVIII in Texas we have Intel kicking off the new Pentium 4 Prescott processors and the latest Pentium 4 Extreme Edition processor clocked at 3.4GHz. The long awaited Prescott comes with some major core changes: 1 MB L2 cache, 16 KB L1 data cache, SSE3, and built on the 90-nm process. With the launch comes four versions of the Prescott (2.8E / 3.0E / 3.2E / 3.4E GHz) with a low-cost Prescott version at 2.8GHz and 533MHz FSB speed that comes without HyperThreading. The 3.4GHz Extreme Edition also is out now and it will also be included in our review.
NetBurst Architecture Changes: Longer Pipelines
The first thing that should be understood with the Prescott is that the Netburst's main branch prediction/recovery pipeline has been lengthened from 20 stages in the Northwood to 31 stages in Prescott. If you've been an enthusiast or in the IT industry for some time,then you know that this has been done before and at an initial performance loss. Intel went from 10 stages in the Pentium III to 20 stages in the original Pentium IV's and the result was a performance decrease. As you will soon see, some applications do take a performance hit due to the change, but the increase in stages will allow Intel to ramp up to higher clock speeds thanks to the new design. Intel has confirmed that we will have "4GHz in 04", but it's rumored and has been seen on documents that the Prescott design will get up to ~5GHz before it hits the ceiling.
The Northwood's have 512 KB, but as the above CPU-Z image shows the Prescott has 1 MB of L2 cache. The die size dropped from 127 to 125 (mm2), but has 125 million transistors compared to Northwood's 55 million transistors. At 3.4GHz, Prescott has a maximum cache bandwidth of an impressive 108 GB per second. Additionally, Intel doubled the L1 data cache from 8 KB to 16 KB.
More Instructions: SSE3
Intel has added 13 new processor instructions to help make life easier for software programmers and has called them SSE3.
fisttp: fp to int conversion
addsubps, addsubpd, movsldup, movshdup, movddup: complex arithmetic
lddqu: video encoding
haddps, hsubps, haddpd, hsubpd: graphics
monitor, mwait: thread synchronization
At the current time not many applications take advantage of SSE3, but in due time the advantages may very well be seen. Think of this as an investment in the future and as applications begin to use the new instructions we will see what impact they have on overall system performance.
The Test Processors:
Individual Test Systems:
- Intel P4 3.2 "C", Intel P4 3.2 "E", Intel P4 3.2 "EE", Intel P4 3.4"EE" -- ABIT IC7-MAX3 (Intel 875P), 1GB (2x512MB) Kingston PC3500 HyperX @ DDR400 (2-3-2-5), nVidia 5900 Ultra 256mb(Forceware 53.03 drivers), 120GB Seagate SATA150 HDD, Windows XP w/SP1 and DX9B.
- AMD A64 FX-51 -- ASUS SK8N (nForce3-150 w/ 3.13 Forceware Drivers), 1GB (2x512MB) Corsair PC3200 Registered XMS @ DDR400 (2-3-2-5), nVidia 5900 Ultra 256mb(Forceware 53.03 drivers), 120GB Seagate SATA150 HDD, Windows XP w/SP1 and DX9B.
- AMD A64 3000+ -- Chaintech ZNF3-150 (nForce3-150 w/ 3.13 Forceware Drivers), 1GB (2x512MB) Kingston PC3500 HyperX @ DDR400 (2-3-2-5), nVidia 5900 Ultra 256mb(Forceware 53.03 drivers), 120GB Seagate SATA150 HDD, Windows XP w/SP1 and DX9B.
- AMD Athlon XP 3200+ Barton -- ABIT AN7 (nForce2 Ultra 400 w/ 3.13 Forceware Drivers), 1GB (2x512MB) Kingston PC3500 HyperX @ DDR400 (2-3-2-5), nVidia 5900 Ultra 256mb(Forceware 53.03 drivers), 120GB Seagate SATA150 HDD, Windows XP w/SP1 and DX9B.
Testing Procedure :
All testing was done on a fresh install of Windows XP Professional build 2600 with Service Pack 1A and DirectX 9.0b. All benchmarks were completed on the desktop with no other software programs running. No overclocking was done on the video card durning any of this review. We did disable the audio, USB, Firewire, and lan features if found in the BIOS menu for all the testing completed during this review.
Now for the results!
Memory Bandwidth Testing:
AIDA32 Version 3.90.5:
AIDA32 is a professional system information, diagnostics and benchmarking program for Win32 platforms. It extracts details of all components of the PC. It also tests the actual read and write speeds of your memory to give a fairly accurate look of true memory performance.
ScienceMark 2.0 Beta:
ScienceMark 2.0 is different from other benchmarks, in a sense the benchmark tests a series of different memory bandwidth algorithms. To top it all off the assembly source for these copy routines is available online to help assure the benchmark is not biased towards any one platform in particular.
Results: It's fairly obvious that the Prescott has an improved memory throughput. The recent P4 processors have not been "starved" when it comes to memory bandwidth, but more can't hurt!
FutureMark; 3dmark2001 SE, Build 330:
Massive Development; AquaMark3:
The AquaMark3 executes a complete state-of-the-art game engine and generates 3D scenes designed to make the same demands on hardware as a modern game. The utilized game engine, the krass? Engine, has been used in Aquanox and AquaNox 2: Revelation as well as in the upcoming RTS Spellforce by Phenomic Game Development. AquaMark3 utilizes recent hardware features of the new DirectX 9 API, such as PixelShader 2.0, while staying fully backward compatible to DirectX 8 and 7 graphics hardware.
SuperPi 1.1e :
SuperPi calculates the number Pi in this raw number crunching benchmark. The benchmark is fairly diverse and allows the user to change the number of digits of Pi that can be calculated. In this benchmark we ran SuperPi to 4 million places.
Results: Thus far the Prescott is able to hang close with the Northwood and even passes up all the 3.2GHz processors during Super Pi testing. Let's move on to some more benchmarks.
MAXON; CINEBENCH 2003:
CINEBENCH 2003 is the free benchmarking tool for Windows and Mac OS based on the powerful 3D software CINEMA 4D R8. The tool is set to deliver accurate benchmarks by testing not only a computer's raw processing speed but also all other areas that affect system performance such as OpenGL, multithreading, multiprocessors and Intel's new HT Technology. Again, higher Frames/Second and lower rendering time in seconds equal better performance.
Results: Here we see that the performance of the Prescott is hurt by the additional 11 stages, but the Prescott is not far behind the Northwood.
Case Lab CFD Solver:
The benchmark testcase is the AGARD 445.6 aeroelastic test wing. The wing uses a NACA 65A004 airfoil section and has a panel aspect ratio of 1.65, taper ratio of 0.66, and a quarter-chord sweep angle of 45. This AGARD wing was tested at the NASA Langley Research Center in the 16-foot Transonic Dynamics Tunnel and is a standard aeroelastic test case used for validation of unsteady, compressible CFD codes. The CFD grid used to model this problem consists of 67,435 nodes and 366,407 tetrahedral elements.
Results: Here we see that the performance of the Prescott is better than the Northwood thanks to the additional 512kb. The Intel 3.4 EE takes the crown with almost a 40 second lead over the Prescott.
Standard Performance Evaluation Corporation; SPECviewperf 7.1.1:
SPECviewperf 7.1 is a professional OpenGL rendering benchmark that can evaluate the performance of OpenGL rendering in a professional workstation environment.
Results: We have found this benchmark to be heavily bottlenecked by the video card being used, but some performance differences can still be noted.
FutureMark; Bapco SYSmark2004:
SYSmark2004 provides an application-based benchmark that accurately reflects usage patterns for business users in the areas of Internet Content Creation and Office Productivity.
Results: Intel dominates across the chart on SYSmark2004! The Extreme Edition processors led the pack closely followed by the Prescott and then the Northwood.
ScienceMark 2.0 Beta:
Science Mark 2.0 is an attempt to put the truth behind benchmarking. In an attempt to model real world demands and performance, ScienceMark 2.0 is a suite of high-performance benchmarks that realistically stress system performance without architectural bias. For the Memory Testing, higher numbers represent better performance. On the remaining tests, lower seconds represent better performance.
Primordia "calculates the Quantum Mechanical Hartree-Fock Orbitals for each electron in any element of the periodic table." We ran the benchmark on default using Argon as our element.
Results: We found that the Prescott had a hard time during this benchmark and actually ended up being the slowest of the seven processors we benchmarked.
NovaLogic; Comanche 4:
The Comanche 4 benchmark demo is a unique benchmark as it represents a real-world gaming experience. It contains the single player Eagle's Talon mission from the game as well as a detailed cinematic. This DirectX 8.1 benchmark demo measures your system's performance in the standard frames per second format.
Results: The Intel 3.4GHz Extreme Edition brought the victory to Intel, but we found the Intel 3.2E almost 10 FPS behind the 3.2C.
Epic Games; Unreal Tournament 2003:
Using the full installation of Unreal Tournament 2003 along with the newest patch gives us a very nice real world benchmark! We also used [H]ardOCP's UT2003 Benchmarking utility version 2.1. A resolution of 1024x768 was tested in Direct3D on the built-in CPU test.
Results: It looks like the Intel 3.2E and 3.2C are on the same level when it comes to our in-game testings of UT2003.
Activision; Call Of Duty:
Using the full installation of Call of Duty along with the newest patch gives us a very nice real world benchmark! We also used a custom demo for our testing at 640x480.
Results: Just as in UT2003 we found the 3.2GHz Prescott and Northwood neck to neck. This time around the Intel processors take a significant lead over AMD.
After looking at the benchmarks it is clear the processor has it's advantages and disadvantages. Sure it's slower than Pentium 4 Northwood chips at the same clock speed, but that was just the case with the original Pentium 4 (Willamette) hit the market and look now! The addition of extra L1 and L2 cache was a smart move by Intel because without it performance would be lower in many of the benchmarks we looked at. It's also too soon to judge the impact of the new SSE3 instructions. Since there are only 13 new instructions (SSE2 had 145) I wouldn't expect a huge increase, but time will tell.
Right now the Prescott is a tad slower than the Northwood at equal clock speeds, runs hotter, and draws more power, but thats not the whole story. Intel has told us that the Prescott will get better as Intel comes out with higher clock speeds. The sweet spot for the Prescott is rumored to be around 3.8-4.0GHz, which are speeds that are not too far away.
One thing you might have noticed in this review is that I didn't touch on temperatures or overclocking of the Prescott. I will be putting out individual reviews covering both of those issues and both are important areas to cover. The Prescott sample we have overclocks very well, which could be different from the retail versions. Just to give you a heads up we are seeing 4GHz+ with little effort on our Engineering Sample.
The new 3.4GHz Extreme Edition that we tested along with the Prescott was very impressive, as well as the $999 price tag that comes along with it. Overall the Intel 3.4 EE and AMD A64 FX-51 led the majority of benchmarks and are both still the CPU's to have if the budget allows.
Overall the first new processor of 2004 seems to be a peak ahead for the near future. In the upcoming months we have Balanced Technology Extended (BTX) interfaces, new CPU sockets, PCI Express, and hopefully the new software to use SSE3 and other new technologies.
Legit Bottom Line:
Intel was able to launch the Prescott processor with major core architecture changes without having a negative impact on performance. Intel now has a core that can be ramped up to 4GHz and beyond with ease!