CPU QuickLinks
Chipset QuickLinks
Video QuickLinks
Software QuickLinks
20XX
Q4 2008
Q3 2008
Q2 2008
Q1 2008
Q4 2007
Q3 2007
Q2 2007
Q1 2007
Q4 2006
Q3 2006
Q2 2006
Q1 2006
Q4 2005
Q3 2005
Q2 2005
Q1 2005
Q4 2004
Q3 2004
Q2 2004
Q1 2004
Q4 2003
Q3 2003
Q2 2003
Q1 2003
Q4 2002
Q3 2002
Q2 2002
Q1 2002
Q4 2001
Q3 2001
Q2 2001
Q1 2001
Q4 2000
Q3 2000
Q2 2000
Q1 2000
Q4 1999
Q3 1999
Q2 1999
Q1 1999
Q2 2000
4/00
Netscape Communicator 6 Beta 1 was released on April 5th.
Internet Explorer 5.5 Beta was released on April 6th. Enhancements include a print preview facility and improved printing along with additional DHTML functionality.
Windows Millennium Edition Beta 3 was released on April 7th. This release contains Microsoft Movie Maker, Media Player 7, USB updates and a number of bug fixes.
AMD K6-2+ Mobile was released on April 18th at speeds of 450, 475 and 500Mhz. This processor is identical to the K6-3+, but only 128Kb of the 256Kb on-die L2 cache will be enabled. The reason for utilising the same cores for both the K6-2+ and K6-3+ is to improve overall chip yield. Although the vast majority of cores will yield the full 256Kb of L2 cache, those that don't can be used as K6-2+'s by mapping out the damaged portions of cache. Although it was rumoured that the full 256Kb L2 cache could be enabled via a BIOS setting on K6-2+ machines, this rumour has proved to be untrue - the cache of K6-2+ CPU's will be hardwired to 128Kb.
AMD K6-3+ Mobile was released in April 18th at speeds of 450, 475 and 500Mhz. This processor is based around an enhanced K6-3 core, containing the additional 5 DSP (but not the 19 SIMD) 3DNow! instructions introduced with the Athlon as well as 256Kb on-die L2 Cache and AMD's PowerNow! (previously known as Gemini) mobile technology. PowerNow! is a similar technology to Intel's SpeedStep, providing a lower overall power consumption as well as the option to sacrifice performance (i.e. clock speed) for battery life. Unlike Intel's SpeedStep, this tradeoff is a BIOS-only function so power requirements cannot be changed on the fly.
Intel Price Cuts for the Pentium III and Celeron series occurred on April 23rd. See the CPU Prices page for additional information.
Mobile Celeron 550 was released on April 24th.
Mobile Pentium III 700 was released on April 24th.
ATI Radeon 256 (RV100 core), ATI's next generation graphics chipset, was unveiled on April 24th, although initial shipments were not made until July 19th. The Radeon 256 is built on a 0.18micron process and features a 183Mhz core and memory clock (Retail versions only - OEM versions and the SDR version run with a 166Mhz core and memory clock), 2 pipelines with 3 texture units per pipeline and support for up to 128Mb of DDR memory. The core frequency gives a fill rate of 366Mpixels/s for single, dual or triple texturing operations (i.e. a maximum of 1.1Mtexels/s). The memory frequency of 183Mhz DDR gives an available memory bandwidth of 5.8GB/s, compared to the 5.2GB/s of DDR GeForce256 and 5.8GB/s of GeForce2. The Radeon's feature set is very complete, featuring support for environmental bumpmapping (like the G400), Cubic environmental bumpmapping (like the GeForce), T-buffer style effects including Full Scene Anti-Aliasing, Motion blur and Depth of field effects (like the Voodoo 5), shadow mapping, range based fog and 3D texture support (a DirectX 8 feature).
Like nVidia's GeForce series, the Radeon has an on-board hardware transformation and lighting engine which will be even more powerful than those found in the GeForce 256 and GeForce 2. Whilst the GeForce 256 is capable of processing around 10M triangles/s and the GeForce2 around 20M, Radeon can process 27.5 million triangles/s. Unlike nVidia's parts, the hardware T&L unit is able to perform hardware clipping, vertex skinning and keyframe interpolation.
One of the main problems facing today's graphics accelerators is the available memory bandwidth. In order to reduce memory bandwidth, the Radeon features HyperZ technology. Essentially this technology reduces the memory bandwidth taken up by accessing the Z buffer. Bandwidth is reduced by compressing Z buffer data and also by performing early culling of hidden polygons. This technology increases the potential fill rate of the card from around 1.1Gtexels/s to around 1.4Gtexels/s but, more importantly, it helps ease the memory bandwidth bottleneck in a 32bit, high-resolution environment.
Like all recent ATI products, the Radeon has good support for video output. DVI flat panels, hardware HDTV decoding and adaptive de-interlacing are supported by the Radeon 256.
Initial benchmarks have shown that the Radeon 256 is a good competitor for nVidia's GeForce2 GTS, although it is slightly slower in the majority of circumstances. Whilst offering a lower maximum fillrate, the Radeon has a superior featureset and an architecture optimised for 3 textures per pixel graphics rather than 2 textures per pixel. Whilst most current games are dual-textured, future, tri-textured games - particularly when played at high resolutions at a 32bit colour depth - may enable the Radeon to overtake the GeForce2. Unfortunately by the time this becomes a significant issue superior cards such as nVidia's NV20 (GeForce 3) will be available.
Matrox G450 was unveiled on April 25th, although it will not be released until later in the quarter. The G450 is an improved version of the G400, built on a 0.18 micron process and running at an estimated core speed of 200Mhz. The memory speed is likely to be either 166 or 200Mhz. In addition to the die-size reduction and subsequent core speed increase, the G450 also features native support for DDR memory, 256bit DualBus architecture (rather than the 128bit DualBus of the G400), a second RAMDAC for increased resolution and refresh rate support for the secondary display (the G400's secondary display went through the analogue MGA-TVO chip limiting the maximum resolution to 1280x1024), DirectX Environmental Bumpmapping support and DVI Flat panel support.
nVidia GeForce2 GTS (NV15) was announced on April 26th on a 0.18 micron process. Volume shipments of GeForce2-based boards will start at the beginning of May. The successor to the NV10 (GeForce 256), the NV15 will feature a significantly redesigned and enhanced core as well as a faster 200Mhz clock speed. The most notable feature added to the core of the NV15 is the addition of a second texturing block to each of the four rendering pipelines. This allows two textures to be applied on each pixel per clock cycle, giving up to twice the multitexturing performance of a similarly clocked NV10. The fill rate of a 200Mhz NV15 is 800Mpixels/s, whether performing a single-texturing operation or a dual-texturing operation - i.e. it has a peak fillrate of 1.6Gtexels/s. This compares to the current NV10 GeForce 256's fill rate of 480MPixels/s for single-texturing and 240MPixels/s for dual-texturing.
In addition to the enhanced graphics pipeline, the NV15 will also feature an enhanced Graphics Processing Unit (GPU). As well as supporting Hardware Transformation and Lighting, the GPU of the NV15 will also incorporate hardware Clipping and Vertex blending & Skinning. The GPU of the NV15 is around 45-50% faster than that integrated into the NV10, with a peak throughput of 25M triangles/s.
One of the other advances made to the NV15 core is the introduction of the NSR (nVidia Shading Rasteriser) processor, which allows a number of per-pixel effects to be performed in hardware. NSR is similar to ATI's Pixel Tapestry architecutre introduced with the Radeon 256, allowing shadow masks, a number of types of bumpmapping (EMBM, Dot Product 3, Embossed), shadow volumes, volumetric explosions, elevation maps, vertex blending, waves, refraction, specular lighting etc.
The NV15 will also feature a 350Mhz integrated RAMDAC, an integrated HDTV processor, a TMDS transmitter for outputting to DVI Flat Panel displays and extended ACPI support. In addition to the 0.18 micron die-shrink, the ACPI support allows the NV15 will dissipate around 50% of the power of the NV10 (i.e. 8-9W rather than the 16W required for the NV10). The NV11 will dissipate 30% less power than the NV15, and the mobile version of NV11 even less than that.
It should be noted that the NV15 will be even more bandwidth limited than the current NV10. Although it is expected that cards based around the NV15 will feature 183-200Mhz SDR SDRAM or 166Mhz DDR SDRAM, the maximum bandwidth requirement is 3.4 times that of an NV10 (given a 200Mhz core clock). Assuming a memory clock of 200Mhz, the bandwidth provision is only 1.5 times that of the NV10. This will mean that the purchase of a DDR SDRAM based card is even more essential than it was for the NV10.
Cards based around the GeForce 2 will be initially only available with 166Mhz DDR memory, although SDR versions are expected to be released for the OEM market. Unlike the GeForce, the GeForce2 will be available as a PCI card as well as an AGP card.
Initial benchmarks have proved to be extremely promising, with performance increases of around 66% for 16bit and 33% for 32bit when comparing the GeForce2 to a DDR GeForce 256. When comparing the GeForce 2 to the other recently announced graphics processors, notably ATI's Radeon and 3dfx's Voodoo 5 5500, benchmarks show the GeForce2 to be the leader, both in terms of raw speed and driver maturity. Only the forthcoming V5-6000 will match the GeForce2 until the release of the next generation cards in the Autumn. It should be noted, however, that the GeForce2 is significantly more expensive than the V5-5500, retailing at around $349 as opposed to $270.
AMD Duron, formally Spitfire, was officially announced on April 27th.
3dfx VSA-100 provides the 3D engine for the Voodoo 4 and Voodoo 5 series of Graphics cards which will debut on April 28th. VSA-100 is the first implementation of 3dfx's Voodoo Scalable Architecture series featuring 14 million transistors on a 0.25 micron process. Each chip consumes around 12-13W of power (note that the non-pro AGP bus can supply around 18-20W) and can access up to 64Mb of memory. The memory and core clockspeed of the VSA-100 will be 166Mhz. The features of this chipset include AGP 4X support, 32bit colour rendering, large texture support (2048x2048), texture compression via DXTC or FX1, a two pixels per clock raster engine and single-cycle trilinear mipmapping. In addition to this, the VSA-100 architecture has support for between 2 and 32 chips running in parallel using 3dfx's Scan Line Interleaving (SLI) technology. The VSA-100's SLI implementation allows the driver to spread the load over all available processors by splitting each frame into sections.
The VSA-100's major new feature is known as a 'T-Buffer', which is only available when running in a multi-chip configuration (currently the Voodoo 5 series). The 'T-buffer' is a cut-down and modified version of the Accumulation buffer used in ray tracing. An Accumulation buffer is used to add effects by rendering the scene multiple times and combining them together. The T-Buffer technology of 3dfx's next generation part allows full-scene anti-aliasing, motion blur, depth of field (i.e. focus) bluring and soft shadows or reflections. 3dfx's new Texture compression scheme, known as FXT1, is similar to but more efficient than S3TC, giving superior compression rations and image quality. FXT1 is also backwardly compatible with S3TC, which should be useful for those games that utilise S3TC, such as Unreal Tournament and Quake III Arena.
The VSA-100 will provide better performance than an equivalently clocked Voodoo 3 for two main reasons. Firstly, the rendering architecture has been significantly optimised, resulting in a performance increase of around 20-25% in multitexturing applications. Secondly, and most importantly, the VSA-100 has a raster engine capable of processing two pixels per clock cycle. The Voodoo 3 can process only one pixel per clock cycle, although two textures can be applied to that pixel. A 183Mhz V3 has a fill rate of 183Mpixels/s and 366Mtexels/s. A (theoretical) 183Mhz VSA-100 would have a fill rate of 366Mpixels/s and 366Mtexels/s. The ability to process two pixels per clock can make the VSA-100 up to twice as fast as an equivalently clocked V3 when performing single-texturing operations.
In the professional market, Quantum 3D (now a 3dfx subsidiary) will provide solutions based around 8 to 32 chips working in SLI with between 128Mb and 2Gb of on-board memory.
3dfx Voodoo 5 series feature dual and quad-chip implementations of 3dfx's VSA-100 architecture and are aimed at the high-end gaming market. The Voodoo 5 series will feature T-buffer effects, a 350Mhz RAMDAC and will require an external power source. The Voodoo 5 series will consist of the following products, with the Voodoo 5 5500 being released on April 28th with shipments starting on June 9th (AGP):
The Voodoo 5 5000 features 32Mb of SDR memory and is a dual-chip PCI-only solution, delivering 667Mtexels/s and priced at $229. The V5-5000 will ship with a power connector which hooks into the internal power supply of the PC.
The Voodoo 5 5500 features 64Mb of SDR memory and is a dual-chip AGP-only solution, delivering 667Mtexels/s and priced at $299. The V5-5000 will ship with a power connector which hooks into the internal power supply of the PC.
The Voodoo 5 6000 is the flagship of the range and features 128Mb of SDR memory and a quad-chip AGP-only implementation of the VSA-100 architecture. The V5-6000 will deliver 1.33Gtexels/s and will be priced at $599. The V5-6000 will ship with a separate PSU, named 'Voodoo Volts', to provide the card with 100W of clean power.
Initial benchmarking of the V5-5500 has shown that when driver issues are eliminated the performance of the V5-5500 is around 15-25% faster than a DDR GeForce 256 when running in 16bit colour and around 10-15% faster when running at a 32bit colour depth. At this stage of driver development it is important to make comparisons by matching the cards with a high speed processor running at high resolutions. This is due to the fact that the drivers are currently very inefficient due to their beta nature, and in order to reduce driver issues to a minimum we need to push the fill rate heavily. The V5 series FSAA (Full Scene Anti-Aliasing) is a very interesting feature. While its effect on First Person Shooter games such as Quake III Arena will be minimal due to the significant performance hit, it will be a welcome addition to Flight sims and 3D Racers where fill rate is not as highly taxed and image quality is more important.
5/00
Windows Me RC0 was released on May 9th.
Microsoft Office 2000 Service Pack 1a was released on May 15th. This release was primarily made for users who had problems installing Office 2000 SP1 onto PC's which had been upgraded from NT4 to Windows 2000. Users who have not had problems installing SR1 need not download this version.
Pentium III Xeon 700 (1Mb, 2Mb L2) was released on May 22nd.
Windows Me RC1 was released on May 23rd.
Windows 2000 Datacenter Beta 2 was released on May 23rd.
Pentium III 933 was released on May 24th.
Intel Price Cuts for the Pentium III series occurred on May 28th. See the CPU Prices page for additional information.
6/00
VIA PM133 chipset started volume shipments on June 3rd. The PM133 is a version of the Apollo Pro 133A, for Intel processors, containing in integrated Savage 4 graphics core. Features include an external AGP 4X port for an optional external graphics card, integrated AC-97, MC-97 audio-modem, ATA-100 support and 4 USB ports.
VIA KT133 chipset was launched on June 5th and is the Athlon Socket A capable version of the KX133. The only difference between this chipset and the KX-133 are the slight timing differences needed to support Socket A processors. See the KX133 Roadmap entry for the full specification of this chipset.
AMD Athlon Thunderbird was released on June 5th at speeds ranging from 750Mhz to 1Ghz in 50Mhz increments. The Thunderbird core will initially be available in both Slot-A and Socket-A format, although the Slot-A processor will be for OEM use only and production will die out when sufficient Socket A motherboards become available. Thunderbird processors will be manufactured on both the 0.18 micron Aluminium process of AMD's FAB25 (Austin) plant and on the 0.18 micron Copper process of FAB30 (Dresden).
Intended as the successor to the current K75 Athlon core, the Thunderbird will contain 256Kb of on-die, exclusive, 16-way set associative L2 cache running over a 64-bit bus. This compares with the Althon Classic's 512Kb external, inclusive, 2-way set associative L2 cache running over a 64bit bus and Coppermine's 256Kb, on-die, inclusive, 8-way set associative L2 cache running on a 256bit bus. The exclusive cache of the Thunderbird differs from that found in the previous Athlon core (K75), as well as Intel's Coppermine, in that the data present in the L1 cache is not duplicated in the L2 cache. When comparing the Thunderbird's cache architecture with that of Intel's Coppermine the most notable aspects are the exclusive cache of the Thunderbird (a win for AMD) and the (significantly) wider bus of the Coppermine (a win for Intel). The Thunderbird core has support for 2-way SMP, although the production of SMP Athlon systems must wait for chipset support to arrive later in the year.
In general the Thunderbird Athlon core has resulted in a performance gain of around 10% (dependant on the application) by moving the cache on-die, and initial benchmark comparisons tend to show that the Athlon has caught up with, and in some cases surpassed, Intel's Coppermine. Like Intel's Pentium II/III and Celeron series, the Thunderbird is clock locked, so overclockers will only be able to overclock by adjusting the FSB speed (this has never been particularly easy on Athlon motherboards as even mild FSB overclocking can result in instability). See the CPU Prices page for additional information.
VIA Cyrix III was officially released on June 6th at clock speeds of 500 to 600Mhz. The Cyrix III processor, previously codenamed Samuel, features a 128Kb L1 cache along with AMD's 3DNow! instruction set and is available in Socket 370 format at 100 and 133Mhz FSB speeds. Initial tests have proved this CPU to be a very poor performer - significantly slower than AMD's K6-II / K6-III series and Intel's Celeron. VIA's clock speed ramping difficulties have not helped - the Cyrix III was initially 'released' at clock speeds of 533 to 667Mhz, but poor yields have postponed the higher clocked parts until later in the year.
Windows Me RC2 was released on June 6th.
Internet Explorer 5.5 went gold on June 10th.
Matrox G450 was released on June 14th, with part availability expected in July. See the April 2000 Roadmap entry for additional information.
AMD Duron, formally known as Spitfire, was released on June 19th at clock speeds of 600, 650 and 700Mhz. These processors will be manufactured on a 0.18 micron Aluminium process at AMD's FAB25 (Austin). This processor is aimed at the budget market, competing with Intel's Celeron (Coppermine 128) series and replacing AMD's K6 series on the desktop. It should be noted that the K6 mobile processors are likely to continue for some time due to their small die size and low power consumption. Duron contains 64Kb of on-die L2 cache along with 128Kb of L1 cache and is produced in the 462-pin Socket A format. Due to its focus on the budget sector, Spitfire will lag behind the other members of the Athlon series in terms of clock speed. Overclocking the Duron will only be possible by adjusting the FSB speed of the CPU. Unfortunately even minor changes to the FSB speed can cause system instability on EV6 (Slot A/Socket A) motherboards. See the CPU Prices page for additional information.
Windows Me (Millennium Edition) was released to manufacturing on June 19th.
i815 chipset, previously known as Solano, was released on June 19th. It should be noted, however, that boards based on the i815/i815E are not expected to be widely available for retail purchase until late summer (September). The i815 and i815E chipsets are likely to become Intel's mainstream chipset for the Pentium III/Celeron II series, superceding the ageing 440BX.
The i815's main feature is the support for PC133 SDRAM, and therefore 133Mhz FSB processors, although it also incorporates the ICH I/O hub which provides support for ATA-66 and 2 channel audio. In addition to the core chipset, an optional Memory Translator Hub will be available for Rambus DRAM support. It should be noted that, unlike the BX chipset, the SDRAM memory clock is asynchronous to the FSB, allowing the use of PC100 SDRAM on 66Mhz FSB Celeron processors, 100Mhz PIIIE's and 133Mhz PIIIEB's.
Although the i815 has chipset support for graphics, it will also support an AGP 4X socket allowing the use of a high-end graphics card. If only basic graphic support is required, the graphics functions can be catered for by populating an AIMM socket (AGP Inline Memory Module) and making use of the integrated graphics core of the i815. Plugging in an AGP card will automatically disable the chipset graphics. The integrated i754 graphics enhance the 3D capabilities found in the i752 (built into the i810 chipset) by providing support for trilinear filtering, texture transparancy and other DirectX 7 features.
See the Tips page for a comparison of Intel's chipsets.
i815E chipset, previously known as Solano-2, was released on June 19th. The i815E is identical to the i815 but with the incorporation of the new ICH2 I/O hub. This provides support for ATA-100, four USB ports (two hubs), integrated 5.1 surround-sound audio and LAN/System manageability functions.
The ATA-100 specification is the successor to ATA-66, allowing read transfer rates of 100MB/s and write transfer rates of 88.9MB/s. ATA-100 will use the same 80 wire cable design as ATA-66. It should be noted that modern IDE hard drives are only just starting to run up against ATA-33's bandwidth limitation of 33MB/s, so ATA-100 will have no real-world advantages over ATA-66 for some time to come. ICH2 also contains two USB hubs, providing a total of 24Mbps (12Mbps per hub) over four ports.
See the Tips page for a comparison of Intel's chipsets.
i820E chipset, also known as Camino-2, was released on June 19th. This chipset is the successor of the i820 (Camino), and will include the ICH2 I/O hub also present in the i815E. ICH2 provides ATA-100, Dual USB hubs, integrated 5.1 surround-sound audio support and LAN/System manageability functions. It should be noted that the i820E will not be shipped with an accompanying MTH (due to the widely publicised problems with this chipset) and will therefore support RDRAM only.
See the Tips page for a comparison of Intel's chipsets.
Mobile Pentium III 600 & 750 was released on June 19th.
Mobile Celeron 600 & 650 was released on June 19th.
Intel Celeron 633, 667 and 700 were released on June 26th. The core voltage of these processors has been raised from the 1.5V of the Celeron II 533-600 to 1.65V - the same as similarly clocked Pentium III Coppermine's.
Mobile K6-2+ 533 and 550 were released on June 26th.
nVidia GeForce2MX (NV11) was released on June 28th on a 0.18 micron process. The NV11 is the cheaper 128bit, two pipeline version of the 256bit, four pipeline NV15. The NV11 is clocked at 175Mhz and hence will have 88% of the T&L performance (around 20Mtrianges/s), but only half the fill rate, of an equivalently clocked NV15 (350MPixels/s). Note that whilst the current 120Mhz NV10 GeForce256 will outperform a 175Mhz NV11 under single-texturing operations, due to its 4 pixel pipelines and hence 480MPixels/s fill rate, the NV11 has a significantly higher multi-texturing fill-rate (700MTexels/s as opposed to 480MTexels/s), due to its second texturing block per pipeline and increased clock speed. This means that modern multi-texturing games such as Quake 3 Arena and Unreal Tournament should perform better on an NV11 than on an NV10 GeForce 256. It should also be noted that the NV11's 128bit SDR interface will mean that memory bandwidth issues are even more serious than they are with the NV15 - already a bandwidth starved chip. The GeForce2 MX core also supports DDR memory, but with a 64-bit memory interface only! Current GeForce2 MX boards contain either 166Mhz SDR memory (giving a memory bandwidth of 2.6Gb/s) or 143Mhz DDR memory (giving a memory bandwidth of 2.3Gb/s). Confusingly, the DDR varieties may perform slightly worse than their SDR counterparts!
The NV11 will also feature nVidia's TwinVision technology. Similar to Matrox's DualHead technology, Dual Display will allow the connection of multiple monitors, or a combination of Digital Flat Panels, TV's and Monitors, to a single graphics card.
A mobile version of the NV11 is also expected to be released, running at a lower clock speed but consuming only 3W. Neither the desktop or mobile versions are required to have a Heatsink.
Initial benchmarks have shown that the GeForce2MX generally performs slightly better than an SDR GeForce, but worse than a DDR GeForce - especially when rendering at a 32bit colour depth - due to memory bandwidth restrictions. It is interesting to note that although the GeForce2MX has a 45% greater peak fillrate then the GeForce it only has the same 2.7Gb/s memory bandwidth of the GeForce SDR. A 32Mb GeForce2MX is expected to be available for around $119. Both AGP and PCI cards are likely to be made available.
nVidia GeForce2BR was released on June 28th. The GeForce2 Blade Runner contains an NV11 core but uses a narrower 64bit (as opposed to 128bit) memory bus to reduce costs - much the same a TNT2 Vanta.
Visual Studio 6 SP4 was released on June 28th. Service Pack 4 includes the latest bug fixes for Microsoft's suite of developer tools, which includes VC++, VB, J++, FoxPro and SourceSafe, as well as new versions of MDAC, HTML Help, MS Scripting and OLE Automation.
To obtain SP4 go to the Visual Studio Web site to download (58-129 MB) or order a CD. MSDN users will obtain SP4 in their August shipment.
