The 1 GHz Intel Pentium III seems to be the subject of much controversy, as many claims have been made about its inability to run in a dual CPU configuration. HardwareCentral has been following the discussion closely and decided to put an end to all the rumors and get a couple of GigaHertz Pentium IIIs and a dual CPU motherboard and find out what exactly is the truth of the matter.
Looking at the datasheets we could find no reference about what exactly prohibits it from running in a dual CPU configuration, other than the rather de-motivating: “NOTE: These processors cannot be used in Dual Processor (DP) applications.” But after consulting some of our industry contacts we found out that the 1 GHz Pentium III supposedly uses a different stepping. Fortunately, nothing was mentioned about pins not bonded to the CPU-die, or other hardware modifications on the chip itself that would render our efforts fruitless. In fact, upon closer examination of the datasheet it seems that the 1 GHz Pentium III uses the exact same stepping, substrate revision and CPUID as the 866 MHz, 850 MHz, etc. Pentium IIIs, the only difference we saw on the datasheet was the higher clockspeed.
All we could now think of that would limit it from actually running in a dual CPU configuration are cooling requirements. The junction temperature, Tjunc, or simply the temperature of the CPU-die, of a Pentium III operating at a GHz must not reach temperatures in excess of 60C°. That explains the rather large cooling solutions as we’ve seen on many commercially available 1 GHz systems, but doesn’t match our experience testing the 1 GHz Pentium III a couple of weeks ago. Running our benchmarks, we used a normal-sized cooling solution in a standard ATX-midi tower case, and saw no temperatures in excess of 50C°.
However most dual CPU motherboards have very little room between the CPU slots, making mounting a cooling solution large enough to meet the CPUs’ temperature requirements can be quite a challenge. Furthermore, most dual CPU motherboards are used in workstation or server configurations, which would make meeting these temperature requirements an absolute necessity. Instability or system crashes are not an option in these configurations. And that currently is the only valid explanation we could think of for not certifying this CPU for dual CPU configurations.
Modifying Slockets
There are two versions of the Intel Pentium III, the SECC2 Slot-1 and the FCPGA form factor. The Slot-1 comes factory assembled and leaves little room for any modifications, whereas the FCPGA has to be interfaced with a ‘slocket’, that can be easily modified and offers an excellent way to experiment with different configurations without actually modifying the CPU. As we mentioned in the introduction, the 866 MHz, 850 MHz, etc. FCPGA Pentium IIIs use the exact same stepping, substrate revision and CPUID as the 1 GHz Pentium III, so our most likely test candidates are two FCPGA 866 MHz Pentium IIIs.
But looking at the wide variety of slockets on the market today we quickly realized that the majority support SMP for dual Celerons, not for dual FCPGAs. The slocket we’ve used in earlier reviews, Iwill’s Slocket II, was unfortunately no exception. It did support a single FCPGA Pentium III, but had no SMP support.
After having closely examined the Pentium III and Celeron datasheets, we realized that the SMP pin on the Pentium III was in a different location and thus probably not connected. Because the Slocket II does support dual Celerons, pin ‘AN15’ is connected, but pin ‘N33” needed for dual FCPGA Pentium III support is not.
Pentium III Processor for the SC242 at 450 MHz to 866 MHz and 1.0 GHz Datasheet
Pentium III Processor for the PGA370 Socket 500MHz to 866 MHz Datasheet
We decided to modify these slockets to find out whether we could get a dual Pentium III FCPGA system up and running. The modification steps below are based on our findings and worked on our configuration. Since configurations vary, they may not work on every setup. We, however, have put considerable time and effort behind verifying them and found them to be compliant with the Pentium III datasheet and the i820/440BX chipset datasheet bus and voltage requirements. We can take no responsibility, however, for any damage, loss of data or other problems that may result from the use of these modifications on your configuration.
First remove the plastic outer casing of the slocket, to gain access to the slocket’s interior. Then set the slocket’s back facing upwards with the Slot-1 connector pointing downwards (see picture). All of the modifications used the same slocket orientation. Obviously the CPU needs to be removed prior to modifiying the slocket.
Iwill Slocket II
Iwill Slocket II Front
Iwill Slocket II Back
Next cut three (3) traces, in the upper left corner of the slocket, between jumper rows two (2) and three (3), connected to PIN3 of JP8 (used to select the CPU type, PPGA (Celeron) or FCPGA (Pentium III)) and PIN2 and PIN1 of JP8. Then solder a wire (red) to the left-most jumper pad. This wire (red) is to be connected to pin N33 (see Pentium III Processor for the PGA370 Socket 500 MHz to 866 MHz Datasheet) which is the pin in the seventh (7) column in the third (3) row from the upper right corner of the PGA-370 socket.
Iwill Slocket II Cutting Traces
Iwill Slocket II Attaching Wire
Iwill Slocket II Pin N33
Iwill Slocket II Soldering Wire/Pin N33
Now we need to make sure the CPU’s bus and voltage requirements are met; all signals need to adhere to Intel’s GTL+ bus protocol. We can easily meet these specifications by mounting a 56 Ohm ¼ watt resistor between GND and PIN2 of JP8. In order to connect the resistor to GND scrape some of the PCB’s protective coating off to create a solder pad large enough to accommodate the resistor. But we first have to make a connection (green) between PIN1 of JP8 and the trace previously leading up to PIN2 of JP8.
Iwill Slocket II Solder Pad/GND
Iwill Slocket II 56 Ohm Resistor
Iwill Slocket II Attaching Wire
Iwill Slocket II Attaching Wire
Now we only need to make a connection (yellow) between PIN2 of JP8 and the 56 Ohm resistor. Then we are done with the modifications to the slocket’s back side.
Iwill Slocket II Attaching Wire
Iwill Slocket II Attaching Wire
Iwill Slocket II Overview
Now we only need to make a connection (yellow) between PIN2 of JP8 and the 56 Ohm resistor. Then we are done with the modifications to the slocket’s back side.
Iwill Slocket II Attaching Wire
Iwill Slocket II Attaching Wire
Iwill Slocket II Overview
The last modification we need to make is removing SMD (Surface Mounted Devices) resistor R3, located on the frontside of the slocket, inside the PGA-370 socket. After having removed R3 make sure all wires are soldered properly, and to the correct PCB and CPU pins.
Iwill Slocket II Locating R3
Iwill Slocket II Removing R3
Dual FCPGA Pentium IIIs
Modifying the slockets is one thing, finding a suitable motherboard with 133MHz front side bus (FSB) support as well as enough room between the CPU slots to accommodate a cooling solution capable of cooling a dual 1GHz configuration is another.
Fortunately Iwill has such a motherboard; the DS133-R, based on Intel’s i820 chipset, has dual RDRAM sockets, built in sound and 10/100Mbps LAN, voltage and busspeed tweaking options and one of the nicest board layouts we’ve seen on a dual motherboard so far. One other big advantage is the sheer size and number of capacitors and other power supply components mounted on this board, which greatly contribute to its speed and stability. Iwill did not cut any corners here.
We were very fortunate to get our hands on a couple of Pentium III 866 MHz FCPGA Engineering Sample CPUs with the same stepping, substrate revision and CPUID as the 1 GHz Pentium III. Initial testing showed that we were indeed able to run both these CPUs in SMP, but we had to increase the VIO and Vcore voltages to 3.5 volts and 1.75 volts respectively to make them run at a GHz and make the dual 1GHz a reality.
Pentium III FCPGA 1GHz SMP
Dual FCPGA Benchmark Setup
The benchmarks we used to evaluate the dual FCPGA Pentium III’s performance are: Bapco SYSmark 2000, which actually consists of a whole slew of popular applications ranging from 3D modeling to word processing, SiSoft Sandra 2000, a synthetic benchmark which measures CPU, FPU, memory and multimedia performance. The configuration used with all benchmarks consisted of:
Hardware Configuration
CPUs : 2 x Intel Pentium III FCPGA ES 866 MHz
Motherboard : Iwill DS133-R i820
Slot-1/FCPGA Adapter : 2 x Iwill Slocket II
Memory : 2 x Samsung 128 MB ECC PC800 RDRAM
Videocard : Elsa Erazor X^2, GeForce DDR
Harddisk : Quantum Atlas V U160 SCSI 36,7GB
CDROM : Plextor UltraPlex 40x Max
SCSI-Adapter : Adaptec SCSICard 29160 U160
Floppy : Generic 1.44 MB
Case : Generic Midi-ATX
Powersupply : Enlight 250W AMD-Approved
The display resolution for all benchmarks was set to 1024x768x16 unless stated otherwise. The software used to test all combinations consisted of the following:
Software Configuration
OS : Microsoft Windows 2000 Pro, typical installation
Applications/business : Bapco SYSmark 2000 v1.0
CPU/FPU : SiSoft Sandra 2000
Multimedia/FPU : SiSoft Sandra 2000
Disk Performance : SiSoft Sandra 2000
Dual FCPGA Benchmark Results
Dual FCPGA 133×7.5
SYSmark 2000 : 197
Sandra 2000 CPU/FPU : 5394/2665
Sandra 2000 MM/FPU : 6285/8444
Sandra 2000 Memory : 477/447
Sandra 2000 Disk : 14089
Dual FCPGA Intel Pentium III 1 GHz SYSmark 2000 Benchmarks
Dual FCPGA Intel Pentium III 1 GHz Sandra 2000 Memory Benchmarks
Dual FCPGA Intel Pentium III 1 GHz Sandra 2000 CPU/FPU Benchmarks
Dual FCPGA Intel Pentium III 1 GHz Sandra 2000 MM/FPU Benchmarks
Dual FCPGA Intel Pentium III 1 GHz Sandra 2000 Disk Benchmarks
Benchmark Evaluation
SYSmark 2000
The SYSmark 2000 scores show a significant improvement over the recently tested 1 GHz Intel Pentium III running on the 440BX platform. This system scored 181 SYSmark points. We have to take into account, though, that the 440BX chipset was being clocked at 133 MHz, which was 33% over spec. If we were to compare a 440BX platform running at its official 100 MHz bus the difference would be more pronounced. Furthermore not all of the applications included in SYSmark 2000 benchmark have SMP support, and thus ran on a single CPU. Also, that system used Microsoft Windows 98SE, whereas we used Windows 2000 Professional here. In summary, the comparison made above can only be used for illustrative purposes.
SiSoft Sandra 200
SiSoft Sandra 2000 provided us with the synthetic Dhry- and Whetstone as well as memory and SIMD/SSE/MMX/FPU scores. These showed that the dual FCPGA Pentium III setup is no slouch in terms of raw CPU/FPU performance. The memory throughput showed some of the memory throughput the i820/RDRAM combination is actually capable of, clearly leaving all other setups in its dust. What we can also clearly see from this benchmark is the scaling in memory bandwidth as the busspeed increases.
Conclusion
Looking back on all the modifications we’ve made to get the dual 1 GHz FCPGA Pentium III up and running, it’s clear that this is not something everyone will be able to reproduce. But it proves the feasibility of a dual 1 GHz Pentium III configuration, although we have to note that a powerful cooling solution is a requirement to get it up and running stable.
Furthermore, the Iwill DS133-R i820 motherboard proved to be a well-performing motherboard that puts the benefits of RDRAM to good use, while combining all the features we’ve been waiting for: support for the 133 MHz system bus, dual RDRAM memory channels and AGP 4X, as well as dual CPU support. We had no troubles operating the same setup at 1066 MHz, although it was very unstable, nor did we experience any of the memory or AGP problems often attributed to the i820 chipset.
In summary, Intel once again proved that its CPUs and chipsets can be made to do things that are ‘not supported’ according to the datasheets. And Iwill has taken the i820 chipset to another level, as we’ve never had an i820 motherboard running so fast and stable. Furthermore, it worked wonders with the modified Slockets while running two FCPGA Pentium IIIs at a GHz. Also Iwill has notified us that a new revision of the Slocket II is currently in the works that will support FCPGA SMP out of the box, making the configuration of a dual CPU system a matter of plugging the CPUs in the slockets, no soldering required.