Competition For AMD’s “Hammer”: P4 With 3.6 GHz And 3.33 GHz
Initially, the news was that AMD’s Hammer would surely be clawing its way to us at the end of this year. Then came the next announcement: not just yet. Even the chip manufacturer’s best contacts around the world sometimes fail to shed any light; then there’s that gut feeling that tells the insider that it won’t be long now, that there’s still potential in the Athlon XP types with the Thoroughbred core. So let’s first banish all uncertainty: the Athlon successor will be launched at the start of 2003 at the earliest.
When AMD recently introduced the fast Athlon XP 2600+ with a modified Thoroughbred B core At The Last Second: AMD’s Trump Card – The Athlon XP 2600+, marketing strategists at Intel immediately began to signal a change of course so that the 3.06 GHz variation on the Pentium 4, originally planned for January 2003, will already be available this year. Interestingly, Intel is confident enough to pit the Pentium 4 against AMD next year, too, where the P4 will run up against the long-anticipated AMD Claw Hammer, which, like the current processors, will be sold under the Athlon label. That’s because the new P4 with Prescott core was initially postponed until mid-2003, leaving only Intel able to field high clock speeds.
That said, in view of the use of the 0.13 µm manufacturing technology, clock speed has its limits. The highest level of development for the P4 with Northwood core will be the 3.6 GHz version. If the rumors are true, Intel won’t launch it until the first half of 2003.
The Intel Pentium 4 with 3.6 GHz: it won’t be on the market for another ten months. The FSB will remain at 133 MHz (corresponds to 533 MHz quad-pumped).
It’s become standard for THG to take forthcoming processors at an early stage and test them for performance and compatibility at the Munich laboratory. Already in February of this year we were able to present the P4/2666, P4/2533 and P4/2400 in Behind The Silicon Curtain: Exclusive Test Of The P4/2666 With 533 MHz Rambus – at the time, the most that was available was the P4/2200.
With the current top version there’s almost 1000 MHz more against the top score back then (2666 MHz) – exactly 3600 MHz. Compared to the current top model with 2.8 GHz, there are virtually no differences; only the core speed has been raised by 29%. The Front Side Bus continues to operate at 133 MHz, the multiplier is 27 (the two weaker versions run on 25 and 23). So, there’s virtually no trace of any increase to 166 MHz FSB speed – in contrast to the enemy camp.
In order to provide an extensive overview of the performance of all the P4 types for the coming ten months, we’ve also included the P4/3333 and P4/3066.
The design of future P4 CPUs will not change. Starting with the 3.06 GHz version (to be launched in November 2002), virtual multi-processing (Hyperthreading) will be supported.
Intoxicated With Speed: AMD Athlon v. Intel P4
The table below gives details on product launches from AMD and Intel over a period of two and a half years. For comparison purposes, we concentrated on the Athlon and the Pentium 4 because these are the image-bearers for each company.
The neck-and-neck race between AMD Athlon and Intel Pentium 4 began more than two years ago, with Athlon becoming the first manufacturer to break the 1 GHz barrier. The details in the table listing the time between launches is also interesting, as it makes clear that AMD took its time at the beginning to successively raise the clock speed on the Athlon.
All the same, the time between the launch of the Athlon 1333 and the 1200 MHz variant was 155 days. Intel took an even longer respite when it took 190 days between the launch of the P4 with 1.5 GHz and the 1.7 GHz variant. Competition has stiffened, particularly in the last 12 months. That’s clear from the shorter times between individual product launches.
Nevertheless, after the P4/2533 was introduced at the beginning of May, Intel let another 112 days pass before launching its current top model, the P4/2800. The reason for this is not hard to find: arch-rival AMD had a tough time raising the clock speed in order to give Intel some stiff competition in terms of performance.
Date | Intel | AMD |
XX.XX.2003 | P4 3.33 GHz | |
XX.XX.2003 | Athlon XP 3000+ | |
XX.11.2002 | P4 3.06 GHz (Hyperthr.) | |
30/09/2002 | Athlon XP 2800+ (40 days) |
|
26/08/2002 | P4 2.8 GHz (112 days) |
|
21/08/2002 | Athlon XP 2600+ (72 days) |
|
10/06/2002 | Athlon XP 2200+ (0,13 µm) (89 days) |
|
06/05/2002 | P4 2.53 GHz (34 days) |
|
02/04/2002 | P4 2.4 GHz (85 days) |
|
13/03/2002 | Athlon XP 2100+ (65 days) |
|
07/01/2002 | P4 2.2 GHz (0,13µm) (133 days) |
Athlon XP 2000+ (63 days) |
05/11/2001 | Athlon XP 1900+ (27 days) |
|
09/10/2001 | Athlon XP 1800+ (97 days) |
|
27/08/2001 | P4 2.0 GHz (89 days) |
|
04/07/2001 | Athlon 1400 (104 days) |
|
30/05/2001 | P4 1.7 GHz (190 days) |
|
22/03/2001 | Athlon 1333 (155 days) |
|
21/11/2000 | P4 1.5 GHz (0,18 µm) | |
18/10/2000 | Athlon 1200 (135 days) |
|
05/06/2000 | Athlon 1000 (0,18 µm) |
Pentium 4 with 3600 MHz: the FSB clock will remain at 133 MHz, the multiplier is 27.
Ride Like The Wind: P4/3600 With 533 MHz RDRAM
We can’t repeat it often enough: only in conjunction with high-speed RDRAM memory (PC1066) is the Pentium 4/3600 able to perform optimally. This does not mean, however, that DDR400 memory is not qualified at all, a fact which has been proven by previous tests. Even a platform with the Intel 845G chipset and fast DDR333 memory (CL2) can stand its ground in many disciplines against a system with normal RDRAM memory (PC800). Yet when RDRAM with 533 MHz (PC1066) or even 600 MHz (PC1200) is deployed, the ways part.
For the time being, DDR memory simply cannot hold its own – the RDRAM platform is always ahead. For this reason, we also tested all P4 processors with the fastest configuration – otherwise it would beg the question of why we would deliberately throttle a 3.6 GHz P4 by using DDR333 or DDR400 with single storage technology.
However, as soon as the dual DDR chipsets from Intel become available next year, the tables are likely to turn in favor of DDR technology.
Pentium 4/3333 during the test.
A view of the BIOS of the Asus P4T533 motherboard.
Pentium 4/3066 during the test.
A view of the BIOS of the Asus P4T533 motherboard – the P4 with 3.06 GHz will be launched this year.
Comparison Of All P4 CPUs
Intel P4 Cores | Willamette | Northwood | Northwood “A” | Prescott |
Clock Speed | 1.3 – 2.0 GHz | 1.4 – 2.6 GHz | 2.26 – 3.60 GHz | 3.60 GHz – 5.xx GHz |
FSB clock | 100 MHz | 100 MHz | 133 MHz | 166 MHz |
FSB speed | Quad-Pumped (400 MHz) | Quad-Pumped (400 MHz) | Quad-Pumped (533 MHz) | Quad-Pumped (666 MHz) |
L1-Cache (Trace) | 12 K µ-Ops | 12 K µ-Ops | 12 K µ-Ops | ? |
L1-Cache (Data) | 8 KB | 8 KB | 8 KB | ? |
L2-Cache (Data) | 256 KB | 512 KB | 512 KB | 512 KB |
Process | 0.18 µm | 0.13 µm | 0.13 µm | 0.09 µm |
CPU Die Size | 217 mm2 | 146 mm2 | 146 mm2 | ? |
Number of Gates | 42 Million | 55 Million | 55 Million | ? |
Gates per size | 193,548 gates/mm² | 376,712 gates/mm² | 376,712 gates/mm² | ? |
CPU Core voltage | 1.75 V | 1.50 V | 1.50 V – 1.525 V | 1.25 V |
Amperage max. | 43 A | 41.7 A | 44.9 A | ? |
Power consumption max. | 75.3 Watt | 62.6 Watt | 68.4 Watt | ? |
Hyperthreading | no | no | yes, from 3.06 GHz | yes |
Basis For Discussion: Memory Performance AMD vs. Intel
The two benchmark suites SiSoft Memory and PC Mark 2002 Memory give clues to the performance of the memory interface. For this, we compared the Pentium 4/2533 with the fastest Athlon XP 2600+ – each on different but comparable platforms equipped with identical peripherals.
The SiSoft test shows that DDR333 memory is marginally slower than 400 MHz RDRAM memory (PC800). The outcome is different with the PC Mark 2002: here, the P4’s DDR platform even outstrips the RDRAM system. As before, the RDRAM platform with its 533 MHz memory clock speed is still unmatched, this system winning by a length in every test. Even with the fastest DDR333 memory (CL2), the AMD Athlon XP 2600+ doesn’t quite reach the performance of an optimally configured DDR-based P4 system.
Test Setup and Details
Intel Hardware (Socket 478) | |
Processors 133 MHz FSB 533 MHz Memory Clock |
Pentium 4 3.6 GHz (3600 MHz) Pentium 4 3.3 GHz (3333 MHz) Pentium 4 3.0 GHz (3066 MHz) Pentium 4 2.8 GHz (2800 MHz) Pentium 4 2.6 GHz (2666 MHz) Pentium 4 2.5 GHz (2533 MHz) Pentium 4 2.4 GHz (2400 MHz) Pentium 4 2.3 GHz (2266 MHz) |
Processors 100 MHz FSB 400 MHz Memory Clock |
Pentium 4 2.4 GHz (2400 MHz) Pentium 4 2.2 GHz (2200 MHz) Pentium 4A 2.0 GHz (2000 MHz) Pentium 4A 1.8 GHz (1800 MHz) Pentium 4A 1.6 GHz (1600 MHz) Pentium 4 2.0 GHz (2000 MHz) Pentium 4 1.4 GHz (1400 MHz) |
Motherboard and Memory | Asus P4T533-C (Intel 850E chipset) Revision: 1.01 Bios: 1006 BETA 002 (07.08.2002) 2x 256 MB RDRAM, PC800, 533 MHz, 40ns, Infineon 2x 256 MB RDRAM, PC1066, 533 MHz, 32ns, Kingstone |
Driver | Intel 850E Driver V 4.00.1013 (7.06.2002) Intel IAA Driver V 2.2.0.2126 |
AMD Hardware (Socket 462) | |
Processors 133 MHz FSB 166 MHz Memory Clock |
Athlon XP 2600+ (2133 MHz) Athlon XP 2200+ (1800 MHz) Athlon XP 2100+ (1733 MHz) Athlon XP 2000+ (1666 MHz) Athlon XP 1900+ (1600 MHz) Athlon XP 1800+ (1533 MHz) Athlon XP 1700+ (1466 MHz) Athlon XP 1600+ (1400 MHz) Athlon XP 1500+ (1333 MHz) Athlon 1400 (1400 MHz) |
Processors 100 MHz FSB 133 MHz Memory Clock |
Athlon 850 (850/100/133 MHz) |
Motherboard and Memory | Epox EP-8K3A+ (VIA Apollo KT333 chipset) Revision: 1.0 Bios: 8k3a2328 (19.06.2002) 512 MB DDR-SDRAM, CL2.0, 166 MHz, PC2700, Winbond |
Drivers | VIA KT333 Driver 4 in 1 Version: 4.40(a)P3 INF File Version: 1.60a AGP Driver Version: 4.20a IDE Filter driver Version: 1.20 |
Common Hardware | |
Graphics Card | GeForce 4 Ti 4600 (MSI MS-8872) Version: 2.00 Memory: 128 MB DDR-SDRAM Memory Clock: 650 MHz Chip Clock: 300 MHz |
Hard Drive | 40 GB,6L040J2 , Maxtor UDMA100, 7200 rpm, 2 MB Cache |
Network | D-Link DFE-530TX (10/100 Mbit) |
CDROM | Asus 52x |
Drivers and Software | |
Graphics Driver | Detonator 4 Series Version: 29.42 |
DirectX | Version: 8.1 |
OS | Windows XP, Build 2600 |
Benchmarks and Settings | |
Bapco Sysmark 2002 | Version 1.0 |
Quake III Arena | Patch V1.16 640×480 – 16 bit / 1024 x 768 – 32 bit Timedemo1 / demo demo001 / nv15demo command line = +set cd_nocd 1 +set s_initsound 0 command line = +set cd_nocd 1 +set s_initsound 0 Graphics detail = Normal |
3DMark 2000 Pro | Version 1.1 – Build 340 1024 x 786 – 16 bit Default Benchmark |
3DMark 2001 SE | Version 1.1 – Build 340 – Patch Build 330 1024 x 786 – 32 bit Default Benchmark |
PCMark 2002 Pro Pack | Build 100 CPU and Memory Tests |
SiSoftware Sandra Standard 2002 SP1 | Version 2002.6.8.97 CPU MultiMedia / CPU Arithmetic / Memory Bandwidth Benchmark |
Newtek Lightwave | Version 7.5 – Build 572 Render First Frame = 1 Render Last Frame = 60 Render Frame Step = 1 Rendering Bench “SKULL_HEAD_NEWEST.LWS” Show Rendering in Progress = 320×240 Ray Trace Shadows, Reflection, Refraction, Transparency = on Multithreading = 8 Threads |
DivX Pro | Version 5.0.02 Xmpeg – Version 4.5 150 MB VOB Datei, 3780 Frames, 720 x 480 pixel, 25 fps Audio = off, Crop = on, Video Format = YV12 Performance/quality = Slowest, Encoding bitrate = 780 kbps |
Pinnacle Studio 7 | Version 7.13.5 Rendering – DVD Compatible no Audio |
Lame | Version 3.92 32 bit Dos Promt, 178 MB Wave File, 44100 Hz 32 – 320 Kbit sampling |
e-merge Winace | Version 2.2 178 MB Wave file, Compression = Best, Dictionary = 4096 KB |
Maxon Computer | Cinema 4D XL 7- Version 7.303 Rendering in 1024 x 786, “Radiosity-Stairs.c4d” |
magix mp3 maker platinum | Version 3.04 D 178 MB Wave file, 44100 Hz, VBR = on and Quality |
SPEC Viewperf | Version 7.0 1280 x 1024 / 32 bit / 85 Hz, Vsync = off |
Comanche 4 | 1024 x 768 / 32 bit / Audio = off |
discreet | 3ds Max 4.2 4.26 Service Pack 1 Characters “Dragon_Charater_rig” Rendering Single, 1024×768 |
Benchmarks Under Windows XP
OpenGL Performance | Quake 3 Arena “Demo 1” and “NV15 Demo” |
3D Rendering | SPEC Viewperf 7 (new) Lightwave 7.5 Build 572 Cinema 4D XL 7.303 3D Studio Max 4.26 SP1 |
DirectX7 Games | 3D Mark 2000 Pro (Version 1.1) |
DirectX8 Games | 3D Mark 2001 SE (Version 1.1) Comanche 4 (new) |
MP3 Audio Encoding | Lame MP3 Encoder 3.92 mp3 Maker Platinum 3.04 |
MPEG-2 Video Encoding | Pinnacle Studio 7.13.5 |
MPEG-4 Video Encoding | XFlask 4.5a and Divx 5.02 Pro |
Office Performance | Sysmark 2002 |
Archiving | WinACE 2.2 |
CPU and Multimedia Bench | PC Mark 2002 SiSoft Sandra 2002 SP1 |
We applied different benchmark tests to give an overall, balanced picture of the performance of the P4 processors that will be appearing over the next ten months. The benchmark results from a total of 35 different CPUs provide a more transparent overview. For the purposes of comparison, all AMD Athlon XP processors were included, as were the classic Athlon with Thunderbird core and the slowest contestant – the chronologically challenged Athlon 850.
We calculated OpenGL performance with four different Quake 3 tests – the Direct3D performance from the DirectX packet was gauged with the 3D Mark 2000 Pro (Version 1.1 and based on DirectX 7) and the 3D Mark 2001 SE (based on DirectX 8).
The various MPEG encoding benchmarks form a comprehensive test scenario: with the help of Lame MP3 Encoder and mp3 Maker Platinum, a 178 MB WAV file was encoded into MPEG-1 Layer 3 format. The MPEG-4 test is an established standard: it involves converting a file from a commercial DVD-ROM into MPEG-4 format using Xmpeg 4.5 and the Dvix codec 5.02 Pro. In addition, an MPEG-2 file is created using the Pinnacle Studio 7 video-editing software.
The standard repertoire also includes quantifying rendering performance with Lightwave from Newtek in its Version 7.5, 3D Studio Max Version 4.26 and Cinema 4D XL 7.303. Archiving is highly relevant in practice, and we performed this task with the new WinACE 2.2 packer.
We turned to the Sysmark 2002 benchmark to check office performance. A comprehensive 3D benchmark suite is supplied by the new Version 7 of the SPEC Viewperf benchmark. And, finally, we mustn’t forget SiSoft Sandra 2002. The new versions of 3D Studio Max 5.0 and Pinnacle Studio 8.1 got to us too late to be included for testing purposes.
OpenGL Performance “Quake 3 Arena”: Record with 416 frames!
For the four time-demo runs of Quake 3 Arena there’s no contest: the 3600 MHz Pentium 4 beats the entire test pack.
DirectX 7 Games: 3D Mark 2000
3D Mark 2000 demonstrates the Direct3D performance of DirectX 7 under Windows XP. The winner by a wide margin is the Intel P4/3600, with 17078 points. Freaks need special cooling and a patched graphics driver for this score!
DirectX 8 Games: 3D Mark 2001 SE
3D Mark 2001 demonstrates the Direct3D performance of DirectX 8 under Windows XP. The upshot of this test is that the P4 leads the field with 13,745 points. Off-the-shelf PC systems usually score 2000 points!
DirectX 8 Hardcore Game: Comanche 4
Comanche 4 is one of the first games on the market to support DirectX 8. The P4/3600 leads the pack with 68 fps. When older processors are used, the new game can hardly be played.
MP3 Audio Encoding: Lame MP3 Encoder 3.92
With the help of Lame MP3 encoder, a 178 MB sound file is converted from WAV format to MPEG-1 Layer 3 format in Windows XP. The chart shows that the P4 with 3.6 GHz is in the lead position.
MP3-Audio-Encoding: MP3 Maker Platinium
MPEG-4 Video Encoding: Xmpeg 4.5a and Divx 5.02 Pro
Memory performance is an important yardstick in MPEG-4 encoding. The P4/3600 manages encoding just shy of 51 frames – that’s faster than double real-time.
MPEG-2 Video Encoding: Pinnacle Studio 7
The P4/3600 can create an MPEG-2 film with Pinnacle Studio 7 in 51 seconds – almost 10 seconds faster than the P4 with 2.8 GHz. A fantastic performance!
SiSoft Sandra 2002 Benchmarks: CPU and Multimedia
SiSoft Sandra Benchmark 2002 showed that the benchmark doesn’t really represent any realistic performance scores.
Multimedia Performance: PC Mark 2002
In both benchmark tests, the P4/3600 comes out ahead of all other CPUs. This benchmark is based on old program code and lacks optimization for AMD and Intel.
Office/Internet Performance: Sysmark 2002
Living up to expectations, the P4/3600 ends up in first place in all three sub-disciplines.
Archiving: WinACE 2.2
Archiving data is a very practical application. With the help of the new WinACE 2.11 packer under Windows XP, a 178 MB WAV file was packed while the clock was running. The 3.6 GHz P4 did the job in just under 144 seconds – the 2.8 GHz P4 took a total of 162 seconds.
3D Rendering: Newtek Lightwave 7.5
Optimizations for Pentium 4 processors really make themselves felt with the Lightwave benchmark – the 3.6 GHz P4 takes another gold medal.
3D Rendering: Cinema 4D XL 7.303
A similar result is produced in the Cinema test: the 3.6 GHz P4 is in the lead.
3D-Rendering: 3D Studio Max 4.26
For this benchmark, ten pictures from the “Rabbit” scene are calculated at a resolution of 800 x 600 pixels. Since the software manufacturer optimized Service Pack 1 in cooperation with Intel, things generally look bad for the AMD Athlon XP.
3D Rendering Performance: SPEC Viewperf 7
The SPEC benchmark reveals varying results: due to the load caused by the code, there were breakdowns during the DRV8, DX07 and PROE01 tests.
Conclusion: 3.6 GHz P4 Outclasses The x86 World – Waiting For The AMD Hammer
The test comparisons show the performance values of the P4 processors that won’t officially be available until the middle of 2003. The top model (the 3.6 GHz Pentium 4) will probably make its dйbut in June of next year. At that time, the P4 will be competing against the Claw Hammer-based AMD Athlon 3xxx+. By then, the P4 models with 3.06 GHz and 3.33 GHz will have been introduced. In any case, it has already been decided that the 3.06 GHz version will be available in November of this year. There’ll soon be company for the speed monster when AMD introduces the Athlon XP 2800+. More details can be gleaned from the roadmap.
At this point, here is an important announcements on the future top model P4/3600: there is a 29% increase in clock speed over the current top model with 2.8 GHz. The primary concern was how the increase in speed was reflected in the performance values. The result we got was that in all benchmarks, the CPU landed in first place by a long shot. The increase over the P4/2800 is between 7% and 29%, depending on the application. The biggest performance leap is with rendering applications, such as Cinema 4D XL, Lightwave 7.5 and 3D Studio Max 4.2. This means that performance increases linearly with clock speed gain – now that’s perfect scaling! Smaller growth can be witnessed in MPEG-encoding tests, such as MPEG-1, MPEG-2 and MPEG-4. The lowest rise in terms of percent was with the 3D games (DirectX7 and DirectX8); here, the increase was between 7% (3D Mark 2001) and 15% (Comanche 4). It should be mentioned that these test results don’t depend on CPU performance alone. In any case, the 3.6 GHz P4 scored the absolute best marks — at Quake 3, for example, with 416 fps, or in the Sysmark test, with 328 points.
All the same, the benchmark results only tell half the truth. On the one hand, the P4/3600 benefits from its high clock speed in conjunction with the rapid RDRAM memory (533 MHz, PC1066). At the same time, though, it should be noted that applications are strongly adjusted to the code extensions SSE and SSE2, which make the Pentium 4 in particular look better. The current AMD Athlon XP is unable to take advantage of the SSE2 optimizations for applications. This feature will be offered only by the Athlon successor “Claw Hammer.”
Since we can only evaluate the performance scale of P4 CPUs of the next ten months, there is no direct comparison with the Athlon Hammer at present. THG’s view is that the introduction of a potent, if not overpowering, “hammer” is more than desirable. It will bring a breath of fresh air into the market, and the ambitious user will profit from the 64-bit option in the bargain. Intel needs more pressure once again! But first, the hurdles that lie ahead in the coming months will have to be cleared, yet again. For all that, the 3.6 GHz P4 has set the bar pretty high. But it’s early yet, and Intel’s 2.8 GHz P4 is on the market.
Update on Sep 12, 2002: The Picture Issue
We have a Pentium 4 in the lab that is completely unlocked. This makes it possible to choose a multiplier of any value as desired (higher and lower values), as long as you make the necessary manipulations to the core voltage on the motherboard. Our source had originally asked us not to publish a photo of this special CPU. Now the situation has changed – we will show a photo of the CPU, but we still refrain from naming our source. By the way, the results of a Pentium 4 3.6 GHz (at standard FSB clock) are still valid.
Regarding the photo that has been much talked about in discussion groups, Tom’s Hardware Guide has never claimed this to be a Pentium 4 3.6 GHz. Instead, the caption reads: “The design of future P4 CPUs will not change. Starting with the 3.06 GHz version (to be launched in November 2002), virtual multi-processing (Hyperthreading) will be supported.”
Judging by the discussions, it strikes us that only few have recognized the scalability of the Pentium 4. Without overclocking the Front Side Bus, it’s already possible to run an Intel processor at a CPU clock of 3.6 GHz. The article describes this technological potential that future Intel processors enable today. It wasn’t about a traditional product launch for a CPU that is available on the market.
In order to make things clear, here are two original photos:
A Pentium 4 CPU for OEMs and editorial tests. This CPU is locked starting from a specific clock speed. With any clock speed falling below this limit, the multiplier can be adjusted. For example: with a P4 that has been locked starting at 2.4 GHz, tests can also be made at clock speeds of 2.3 GHz and 2.2 GHz. 2.5 GHz, however, is not possible if the FSB remains at its standard clock frequency.
A special Pentium 4 CPU from an undisclosed source. This CPU is completely unlocked. Any multiplier value can be selected until the stability limit has been reached.