Intel Released New Celeron Processors
On the beginning of January 1999 Intel launched two new Celeron CPUs, each for the well known Slot1 as well as for the new Socket370. The new Celeron CPUs are running at 66 MHz front side bus clock and at the core clock speeds of 366 and 400 MHz. They include the 128 kB on-die L2-cache, known from the previously released Celeron 300A and Celeron 333. This L2-cache boosts the performance of Celeron into the same area as Pentium II CPUs at the same clock speed, so that there’s hardly any performance advantage of Pentium II over Celeron anymore. The new Socket370 looks pretty similar to Socket7 when you have a first peek at it, but it turns out that Socket370 has an additional row of pins. Socket370 is a conversion of Slot1 to a socket, running at the same bus protocol as the Pentium II, the GTL+ bus protocol. Thus it can work with the known Pentium II chipsets, as 440BX, 440LX, 440EX and the new 440ZX from Intel or the Apollo Pro chipset from VIA. The reason behind Intel’s move back to a socket rather than a slot-solution is pretty simple. Now since the L2-cache fits onto the CPU-die, there’s no additional space needed anymore, as it is provided by Slot1 or Slot2. Socket370 is supposed to save Intel a major amount of money in production costs over Slot1 and the motherboards for Socket370 will also be cheaper to produce than Slot1 boards. Once Intel is producing Pentium III in 0.18 micron technology, as planned for Q3 1999, this new high end CPU will also have on-die L2-cache (CPU code name ‘Coppermine’) and will most likely go back to a socket-solution as well. It turns out that Slot1 was only a transition for the time when Intel was not able to place CPU core and L2-cache onto the same piece of silicon.
The size of Celeron for Socket370 is the same as Pentium MMX’s, but you can see that Celeron has a lot more pins.
The MS-6905 from MSI let’s you run a Celeron for Socket370 in any Slot1-board.
Some of you may guess that Celeron’s big disadvantage against Pentium II is its 66 MHz bus clock. Pentium II 350, 400 and 450 run at 100 MHz bus clock or front side bus and we learned last year that this is what it’s supposed to take for high performance computing. Running normal software as e.g. office or 3D gaming applications does hardly show much of a difference between systems that run at 66 vs. 100 MHz FSB though. Most of the performance is still achieved by the L2-cache and although Celeron’s L2-cache is only 128 kB vs. 512 kB of the Pentium II, the Celeron L2-cache runs at CPU-core clock, the external L2-cache of Pentium II is only clocked at half the CPU core speed.
Celeron Now Threatening Pentium II Sales
So it turns out that Celeron can affect the sales of Pentium II considerably, because the performance difference between Celeron 400 and Pentium II 400 is pretty small, Celeron 366 is even almost identically as fast as the much more expensive Pentium II 350. Intel does of course not want that. The restriction of Celeron to the 66 MHz front side bus is already one action that Intel took to make Celeron look inferior. The next thing is the platform for Celeron. As long as you get the still available Celeron for Slot1, you can stick it into the same platform as Pentium II. However, this will make a Celeron system unnecessarily expensive, because after all you don’t need the 100 MHz FSB offered by a BX-board platform, unless you should try overclocking it. Intel wants us to stick Celeron into the Socket370-platforms and those platforms don’t really come with BX-chipset, or at least there are not supposed to. Socket370 boards are available in a lot of different flavors. The vast majority is using the good old 440LX AGP-chipset, which is cheap and offering a good performance for 66 MHz FSB-systems as well as the full range of features that are expected of a professional system. Intel wants us to use the new 440ZX chipset though, which is a castrated version of 440BX. It supports only 2 DIMM slots, less PCI slots and no ECC for the memory. Since it is the smaller brother of BX, it would be eligible to run at up to 100 MHz FSB, but Intel claims the opposite by marking the latest ZX-chips with ‘ZX-66’. The experienced of us know that in reality ZX can do 100 MHz FSB just as fine as any Celeron could, but Intel doesn’t want that, to save the Pentium II sales.
Overclock Protection
It seems as if it is becoming reality now, the new Celerons are supposed to be double overclock protected. It is already a matter of fact that Celeron will only ship with a multiplier lock, many of the new Socket370-boards don’t even have any jumpers, dip switches or BIOS settings for the clock multiplier anymore, because the new Celeron only runs with one. Additional to that, Intel is obviously now or very soon putting a PLL-circuit on the chip, which makes sure that Celerons do only run as 66 MHz front side bus. This will give you zero chance of overclocking anymore, so that all the nice 75 or 83 MHz FSB-settings on the new Socket370 boards are probably supposed to stay unused forever.
Multiprocessing
The upcoming Quake arena could change a lot, it will be the first 3D action game with multi-processing support. So far the vast majority of people couldn’t be bothered about dual-CPU systems, because only some specialist software as e.g. 3D rendering or CAD applications were able to take advantage of it. Multi-processing is also not supported by Windows 95 or Windows 98, so you would have to get Linux, BeOS or Windows NT to enjoy the blessings of parallel processing. Now, since Id announced that Quake Arena will support multi-CPU-systems, the interest in dual-processor-systems has grown considerably. Celeron is not supposed to support multi-processing, and whilst you can get around that in case of a Slot1-Celeron with some serious soldering and drilling, you can finally forget it with the Celeron for Socket370. We still don’t know how much advantage Quake Arena will take of dual-CPU systems, but this game could be one of the very few reasons why spending a lot more money for a Pentium II rather than a Celeron system.
Upgrade Path
The other reason why a Pentium II or at least a Slot1-system could be of advantage is the possible upgrade path to Pentium III. Intel will release Pentium III 500 at the end of February and the new ‘streaming SIMD instructions’ (former ‘KNI’) are supposed to give it an edge over Pentium II in 3D gaming, video processing and other nice areas. Even if ‘Coppermine’, the 0.18 micron Pentium III with on-die L2-cache that’s planned for Q3/1999, should go back to a socket-solution, it will hardly be Socket370. Intel won’t be crazy enough to do this, they will define a new socket, so that the Socket370-platform owners will only be able to upgrade to the then upcoming Celeron for 100 MHz FSB.
AMD
The new Celerons will give AMD a really hard time. K6-3 seems to be late, which isn’t really surprising if we look at what happened with K6-2 or K6 in the last two years. K6-2 has hardly got a chance against Celeron, or at least not at the price point it’s selling right now. The performance chart will show you that even Celeron 366 is already faster than K6-2 400. Celeron 366 sells for less than K6-2 400 right now, so that AMD will feel forced to adapt K6-2-prices to Intel’s new low-cost/high-performance CPUs.
The Test Systems
For the CPU comparison chart all Intel CPUs were tested in the following Slot1-system:
- Asus P2B motherboard with Intel 440BX chipset
- 128 MB PC100 DIMM (Kingston w/Micron chips)
- Hercules Dynamite TNT graphics adapter, 90/110 clock, NVIDIA reference driver 0048
- Adaptec 2940U2W PCI SCSI host controller
- IBM DGVS 09U UW-SCSI hard drive
- 3Com 3C905B-TX Ethernet adapter
All AMD CPUs were tested in the following Super7-system:
- Asus P5A motherboard with Ali’s Aladdin V chipset, revision E
- 128 MB PC100 DIMM (Kingston w/Micron chips)
- Hercules Dynamite TNT graphics adapter, 90/110 clock, NVIDIA reference driver 0048
- Adaptec 2940U2W PCI SCSI host controller
- IBM DGVS 09U UW-SCSI hard drive
- 3Com 3C905B-TX Ethernet adapter
Windows 98, 1024x768x16 resolution, 85 Hz refresh rate
For 3D StudioMax Windows NT4 SP3, 1024x768x16 resolution, 85 Hz refresh rate
For the Socket370 comparison we used the following system configuration:
- SuperPower SP-7XZA motherboard with onboard Creative ES1373 sound chip (disabled)
Intel 440LX chipset:
- SuperPower SP-7XLA motherboard with onboard Creative ES1373 sound chip (disabled)
VIA Apollo Pro Plus chipset:
- TMC TI7NBA motherboard
Intel 440ZX chipset:
All tests were performed with special Intel evaluation samples of the new Celeron processors, which were neither multiplier nor FSB-clock locked. This does not mean that the final Celeron processors will still be overclockable!
All K6-2 tests were using the new ‘CXT’ core of K6-2, including the new write-combining feature.
A – The Influence of the Front Side Bus
Intel wants to make us believe that 100 MHz FSB is significantly superior to 66 MHz FSB to justify the higher price of Pentium II over Celeron. With normal software I couldn’t find much though.
The Celeron 400/66 is about 3.7% slower than the Pentium II 400 in Winstone 99, in Quake 2 it’s about the same and in 3D StudioMax the Celeron is even a bit faster than Pentium II, due to its faster L2-cache.
B – Comparison of Socket370-Platforms
Intel wants to push the 440ZX chipset, but most of the Taiwanese motherboard manufacturers prefer 440LX. They have a good reason, because the LX-chipset is cheaper and a full blown PII-chipset, with support of five PCI bus masters, 4 DIMM slots and ECC. ZX is the stripped-down version of 440BX and thus not offering the same features as LX. The advantage of BX over LX is mainly the 100 MHz FSB-support, which isn’t used in ‘ZX-66’ anyway. VIA’s Apollo Pro Plus chipset is a new version of Apollo Pro for Slot1-systems. It is also fully equipped as Intel’s LX, but even offering 100 MHz FSB-clock. This won’t help much though if the new Celeron’s are all overclock-protected.
The BX-chipset on Slot1 has got a slight edge over the Socket370 boards, but this could be due to the board architecture. After all the Slot1 test board is the Asus P2B, known as a particularly good performer. ZX and LX don’t show any difference in Winstone 99, but ZX seems to have a better AGP-implementation, making it somewhat faster in Quake2. VIA’s Apollo Pro Plus is definitely slower than the Intel chipsets, but that’s what we already expected remembering the results of the Apollo Pro on Slot1.
C – CPU Office Performance under Windows 98
As in the years before, the new Winstone 99 shows different results to its predecessor when you compare different CPU-architectures. Whilst AMD’s K6-2 was performing very well vs. Pentium II in Winstone 98, it now dropped a bit in Winstone 99.
This is the chart containing only the ‘unoverclocked’ CPUs. You can see that Celeron sticks up very well.
Including all overclocked CPUs makes the chart a lot larger, but not really much different. AMD’s K6-2 400 is somewhere between Pentium II 333 and Celeron 366. Please realize that you might not be able to reach all those numbers anymore when Intel introduces the overclock-protection in all of their chips.
D – CPU 3D Gaming Performance
This time I used NVIDIA’s RIVA TNT instead of 3Dfx’s Voodoo2 for the Quake2-performance comparison chart. I did this because it is more valid when looking at other games than Quake 2. AMD has a special 3DNow!-patch for the combination of Quake2 and Voodoo2, but this patch does not exist for other games. Thus it’s not really representative using the most ideal environment for K6-2.
For Intel CPUs I ran Quake2 3.20, for AMD CPUs Quake2 3.19 with AMD’s Quake2 patch for 3.19, video mode ‘3Dnow! OpenGL’ selected.
The same with overclocked CPUs:
This chart shows that the current Intel CPUs do still have the clear lead in 3D-gaming and I doubt that any hardcore gamer is really using K6-2. However, K6-2 is still sticking up pretty fine and can certainly be used for most 3D games out there just fine. Quake 2 at 1024×768 without a Voodoo2 is pretty tough with the K6-2 though.
E – FPU Performance
It is pretty much tradition that Tom’s Hardware Guide is using 3D StudioMax for FPU-benchmarking. 3D rendering is one of the fields where some real FPU power is required unless you want to wait days for the rendering of your scenes.
This chart is not too different to the Quake2-chart. K6-2 has still got a hard time against any 6th-generation Intel CPU when it comes to pure FPU performance.
Summary
It is certainly true that most of us would have loved to see AMD’s K6-3 kicking Intel’s butt in this comparison and that includes me too. However, the K6-3 seems to be farer away than it seemed at the end of last year. So we should be fair and congratulate Intel on a job well done. The new Celerons are performing very well. We still have the bitter taste of Intel’s new overclock-protection and the cheesy ‘ZX-66’ chipset in our mouth though and this is certainly another reason why none of us will embrace Intel for the release of Celeron 366 and 400. Nevertheless are those new CPUs offering one highly important thing: very high computing power for very little money. And that brings me back to AMD. Without Advanced Micro Devices we would never have come to this situation, so let’s thank them for giving Intel a tough competition and let’s hope that K6-3 will be out soon.