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

mike@mikeshardware.co.uk

Q2 2004

Previous Roadmap   Next Roadmap

04

nVidia GeForce 6800 / Ultra (NV40) was announced on April 13th. NV40 is the successor to NV35 (the GeForce FX 5900 series) and offers a significant performance improvement over its predecessor, the FX 5950 Ultra. The NV40 is built on a 0.13micron process and consists of 222 million transistors, compared to 125 million in NV3x. NV40 is an AGP 8X part. The forthcoming NV45 will be identical to the NV40 but will have a PCI Express interface.

The NV40 core features 16 pixel pipelines and 6 vertex pipelines - quite an increase from NV35's 4x2 pixel pipelines (4 pipelines with 2 texture units per pipeline) and 4 vertex pipelines. NV40 will be available in a number of graphics card models with the GeForce 6800 Ultra offering the full compliment of resources (16 pixel and 6 vertex pipelines). The non-Ultra model disables four of the pixel pipelines, leaving a 12 pixel and . The GeForce 6800 Ultra runs at a clock speed of 400Mhz interfacing to 1100Mhz GDDR3 memory (GeForce FX 5950 Ultra had clocks of 475/950Mhz).

The NV40 is a completely new design with which nVidia tried to overcome all the shortcomings of the NV3x series. NV3x had a large performance hit when using FP32 (rather than FP16) due to a lack of register space, so the register setup has been completely redesigned and increased in NV40 which has reduced the performance hit when using FP32.
The flitering and AA system has also been redesigned for Intellisample 3. The NV3x series had excellent texture filtering algorithms, but they came with a significant performance reduction. The algorithms of NV40 are actually more simplistic and produce inferior output, but they do so with much less of a performance hit. The Anti aliasing of NV40 has also been modified to support a rotated grid arrangement for 4X AA (which was previously only present in 2X AA). This is an improvement, but not as good as ATI's programmable sample pattern. The performance of 8X AA has also been improved, by moving from a 4x super-sample + 2x multi-sample to a 2x super-sample + 4x multi-sample algorithm.
High dynamic range rendering has also been improved in NV40 by implementing a full FP16 frame buffer, compared to the 32-bit integer unit used by ATI, which should alow quality rendering for high-end applications. This gives nVidia full floating point precision throughout the entire pipeline.
nVidia have also improved their shadow rendering with their Ultra Shadow 2 technology. NV40 offers 4X the performance of NV35 when it comes to the real time shadow effect used in Doom 3 and other upcoming games.
Perhaps the most significant advancement in NV40 is it's full support of the Shader Model 3 specification in DirectX 9 - something ATI does not do with their R420. The key change in SM3 is support for larger vertex and pixel shader programs (65k in length compared to 256 instructions for vertex programs and 96 instructions for pixel shader programs) and a much better programming model including support for subroutines, loops and branches and dynamic flow control. SM3 does not add significant visual changes over SM2 (most SM3 effects can be done with SM2 and multipass). SM3 does, however, add geometry instancing (for replicating identical objects) and true displacement mapping.

On the 2D front, nVidia have included a dedicated programmable video processor which can be used to encode or decode MPEG 2, MPEG 4, DiVX, WMV9 or any other type of video file format as well as things such as scaling, gamma conversion etc. The video processor consists of an address, scalar, 16-way SIMD vector, and branch unit.

Initial benchmarks suggest that the GeForce 6800 Ultra offers slightly less performance than ATI's Radeon X800 XT. However, the R420 is simply a slightly enhanced R3xx so the performance increase through drivers is going to be minimal. The NV40 is a completely different architecture with a large amount of untapped performance when using the current drivers. I would expect future drivers revisions to allow NV40 to equal and probably surpass R420. The major downside for NV40 is it's power requirement - 2 molex connectors, a huge double height heatsink & fan and hefty PSU are all required.

05

IBM POWER5 CPU was released on May 3rd. The POWER5 features support for 2 way SMT, a 1.92Mb L2 cache an on-die memory controller and an improved L3 cache over the POWER4 series processors.

ATI Radeon 9100 IGP PRO (RS350) chipset for the Pentium 4 was released on May 3rd. The RS350 is the successor to RS300 (Radeon 9100 IGP, 'non-Pro'). Although the chipset is heavily based on the RS300 and features an identical graphics core, both the North and South bridges have been modified for the RS350. The North bridge features generally enhanced system performance, optimised AGP 8X performance, improved memory timing and increased DDR400 compatibility. The South Bridge (IXP320) features support for 8 x USB2 ports and most notably 2 x SATA ports.

Initial benchmarks show that the performance of the Radeon 9100 IGP Pro is noticably better than the previous generation Radeon 9100 IGP. This chipset works especially well with the Precott core Pentium 4, outperforming even Intel's i875 under many circumstances. The performance with Northwood is good, but does not beat either the i875 or i865. Of course, the Graphics performance is miles ahead of every other Pentium 4 chipset despite the fact that it is identical to the Radeon 9100 IGP.

ATI Radeon 9000 IGP Pro (RC350) chipset for the Pentium 4 was released on May 3rd. The RC350 is the single channel memory version of RS350.

ATI Radeon X800 Pro / XT (R420) GPU was released on May 4th. The R420 (AGP8X), and its brother the R423 (PCI Express), are the successors to the R360 core (Radeon 9800 XT) and feature 160 million transistors built on a 0.13 micron process. Unlike nVidia's NV40 core, the R420 is not a radical departure from ATI's previous generation product, the R360, and does not fully comply with the Pixel Shader 3.0 (DX9.0c) specification. The R420 core features 16 pixel pipelines and 6 vertex units (the same as NV40), although not all the pipelines will be enabled in ATIs lower prices offerings. The main changes over the R360 core - asides from the increased number of pixel and vertex units - are enhancements to the memory controller, support for longer shader programs, more internal registers, a few updated instructions (e.g. SINCOS in 1 clock) and a new compression scheme dubbed 3Dc for improving the quality of compression on Normal maps. One significant advantage the X800 series has over nVidia's competing GeForce 6800 series is the units lower power requirements. The X800 XT requires 67w, compared to the 6800 Ultra's 110w. This means that the cards only require one Molex connector for power, do not necessitate a 480w PSU and are all available in single slot format.

The R420 is expected to be available in a variety of forms, with the X800 Pro and X800 XT being released today. The specification of these cards are as follows:

ATI Radeon X800 Pro - 475Mhz core. 900Mhz memory. 12 pixel pipelines, 6 vertex units. Expected retail price $399.
ATI Radeon X800 XT - 525Mhz core. 1150Mhz memory. 16 pixel pipelines, 6 vertex units. Expected retail price $499.

Initial benchmarks show that the Radeon X800 series generally performs at a slightly higher level than the equivalent GeForce 6800 model (comparing the X800 Pro with the 6800 and the X800 XT with the 6800 Ultra) with today's drivers. However, NV40 is based on a fundamentally new architecture which has quite a lot of room for driver performance improvements, unlike R420, and it also has the advantage of full PS3 support. Having said that, the lower power requirements and slight initial performance lead should make the ATI cards the preferred option for many.

nVidia GeForce 6800 Ultra Extreme was released on May 4th. This is a higher clocked version of the GeForce 6800 Ultra and features a 450Mhz core and 1.2Ghz memory clock. The performance of this part is expected to be similar to, if not a little faster than, ATI's Radeon X800 XT with current drivers.

nVidia GeForce 6800 GT was released on May 4th, although it is not expected to ship until Mid June. This is a lower clocked version of the GeForce 6800 and is designed to fit in a single slot and require only one Molex power source. The GeForce 6800 GT features a 350Mhz core and 1Ghz memory clock speeds. It offers the full complement of 16 pixel shaders.

VIA K8T800 Pro chipset for the Athlon 64 is was released on May 6th. The K8T800 Pro is the successor to the K8T800, featuring support for a 1Ghz Hyperthreaded FSB and a PCI/AGP lock - the K8T800 Pro has an asynchronous bus design allowing PCI and AGP devices to be correctly clocked when overclocking the FSB/CPU. The South Bridge remains the same as that incorporated into the K8T800 chipset - VIA's 8237. The K8T800 Pro will be made available for Socket 754, Socket 940 and Socket 941.

nVidia nForce2 Ultra 400Gb was released on May 10th. The nForce2 400Gb is the successor to the nForce2 Ultra 400 and leverages the featureset of the nVidia nForce3 Pro 250 onto the Socket A platform. The Ultra 400Gb adds native Gigabit Ethernet support, a built in hardware firewall, 2xUATA133 channels and 2xSATA150 channels along with impressive RAID support (raid between Parallel and Serial ATA, RAID 0, 1 and 0+1, Spare drive support etc) to the featureset of the nForce2 Ultra 400.

The integration of the nForce3 Pro 250 featureset to the nForce2 has once again made nVidia's the premiere chipset for Socket A.

Intel Pentium-M (Dothan) mobile CPUs were released on May 10th. The Dothan core is the 0.09 micron successor to Banias. The initial members of the Dothan family are the 735 (1.7Ghz), 745 (1.8Ghz) and 755 (2Ghz). All three members of the Dothan family clock down to 600Mhz when using battery. Dothan contains 140 million transistors with 64Kb L1 and 2Mb L2 cache and features a 400Mhz FSB speed. Like Banias, Dothan will use Intel's Odem and Montara (855 series) chipsets.

Initial benchmarks show that Dothan scales up nicely from Banias, featuring better Office performance and considerably better Content Creation performance.

ATI Radeon 9550 was released worldwide on May 13th (previously it was only distributed in the Far East). The Radeon 9550 is a lower clocked version of the Radeon 9600, featuring a 250Mhz core and 400Mhz memory clock. The memory interface is 128-bit meaning that its performance is higher than the 64-bit Radeon 9600SE. Benchmark results also show it's performance is higher than it's main competitor, the GeForce FX 5500.

AMD Opteron 150 was released on May 18th. The Opteron 150 runs at a clock speed of 2.4Ghz.

AMD Opteron 250 was released on May 18th. The Opteron 250 runs at a clock speed of 2.4Ghz.

AMD Opteron 850 was released on May 18th. The Opteron 850 runs at a clock speed of 2.4Ghz.

Microsoft Mac Office 2004 was released on May 19th. Mac Office features significant performance improvements over its predecessor and makes more use of OS X's Quartz graphics. Max Office 2004 adds a navigation plane for Word (like that found in Acrobat), a compatibility checker to test for PC compliance and support for OS X's long filenames.

VIA C7 (Esther) CPU was announced on May 19th, with general availability expected in late June 2005. Esther is the successor to VIA's previous Nehemiah core. Esther is built on a 90ns process and features 26.2 million transistors with a 128Kb 32-way L2 cache. Esther is targeted at security applications and low power devices offering a power consumption of 3.5W at 1Ghz. Esther can operate at a clock speed of 2Ghz and above and interfaces over an 800Mhz FSB to either a Banias or VIA V4 bus. Esther supports SSE2 & SSE3 and VIA's PadLock hardware security which includes execution protection through NX, Montgomery multiplier support for RSA encryption and secure Hash algorithms (SHA-1 and SHA-256) in hardware.

Intel CPU Price cuts for Xeons occurred on May 30th. See the Intel CPU Prices page for more details.

06

ATI Radeon X800 Pro & XT for PCI Express (R423) GPU was released on June 1st. The R423 core is identical to the R420 apart from it's use of the PCI Express x 16 bus. Unlike NV45, which uses an AGP to PCI Express bridge, ATI's GPU is a native PCI Express solution. The clock speeds of R423 designs are expected to be identical to those of the R420.

ATI Radeon X600 Pro & XT (RV380) was released on June 1st for the PCI Express 16X format only. The RV380 core is the successor to the RV360 (Radeon 9600 series), aimed at the mid range market. RV380 is, in fact, a slightly modified version of RV360 with native PCI Express support. The X600 will be released in two formats - the X600 Pro and the X600 XT. Their specification is outlined below:

ATI Radeon X600 Pro - 400Mhz core. 600Mhz memory.
ATI Radeon X600 XT - 500Mhz core. 740Mhz memory.

Both varieties are identical, containing 4 pixel pipelines and 2 vertex units, apart from their clock speeds. The X600 Pro offers the same clock speeds as its AGP8X counterpart the Radeon 9600 Pro. The X600 XT offers the same core clock as the 9600XT but has a memory clocked at 740Mhz DDR rather than the 600Mhz DDR of the 9600XT.

ATI Radeon X300 (RV370) was released on June 1st for the PCI Express 16X format only. The RV370 core is the successor to the RV280 (Radeon 9200 series), aimed at the low end of the market and built on a 0.11 micron process. The RV370 offers the same 4 pixel pipelines / 2 vertex units, DirectX 9 compliant architecture as RV380, but runs at lower clock speeds (yet to be announced). The X300 will be available in the following configurations:

ATI Radeon X300 - 128-bit memory interface.
ATI Radeon X300 SE - 64-bit memory interface.

ATI Mobility Radeon X600 was released on June 1st. The Mobility X600 offers the same 4 pixel pipelines and 2 vertex units as the X600, but additionally contains advanced power saving features. New features for the Mobility X600 are PowerStep 5 and LCD-EE, a technology for saving energy on attached LCD panels. The Mobility X600 is based around a PCI Express bus, but interfaces through a modular package similar to nVidia's MXM called AXIOM.

Athlon 64 3800+ & 3500+, the first AMD CPUs on Socket 939, were released on June 1st.  Both processors are built around the Newcastle core (i.e. 512Kb L2 cache) and run at clock speeds of 2.4Ghz and 2.2Ghz respectively. Socket 939 processors have a dual channel DDR memory controller to unbuffered DDR400 memory, giving a theoretical bandwidth of 6.4GB/s - the same as the Dual Channel DDR chipsets for the Intel platform. However, as the memory controller is on-die the realised memory bandwidth is around 5.7GB/s - around 40% more than the 4.4GB/s of Intel's i875P.

Athlon FX-53 for Socket 939 was released on June 1st. This processor is identical to the FX-53 on Socket 940, but can use Unbuffered DDR.

Athlon 64 3700+ for Socket 754 was released on June 1st. This processor is based on the Clawhammer core (1Mb L2 cache) and runs at a clock speed of 2.4Ghz.

nVidia nForce3 Ultra MCP for Socket 939 was released on June 1st. In addition to the featureset of the Socket 754 nForce3 Pro 250, the Socket 939 nForce3 Pro 250 features a 1Ghz Hypertransport link.

AMD Sempron processor series was announced on June 6th. The Sempron brand name will replace Duron for AMDs value processor lineup. Sempron processors will be available for Socket A, Socket 754 and Socket 939, but will feature different cores depending on their interface. Sempron processors will be labeled as Value 2xx or Value 3xx processors, e.g. Value 320. A Value 320 processor would be equivalent to an Athlon XP 3200+ (i.e. divide the previous PR rating by 10).

AMD 8132 chipset for the Opteron was released in June 14th, with availability in Q4. The 8132 chip is part of AMD's 8000 series chipset and is a PCI-X 2.0 tunnel. PCI-X is higher speed version of PCI with PCI-X 2.0 supporting signalling speeds up to 533Mhz and offering ECC support for the PCI-X communication. The 8132 chip supports both the PCI-X 1.0 (66Mhz and 133Mhz) and PCI-X 2.0 (266Mhz and 533Mhz), offering up to 4.3GB/s of bandwidth for high-performance applications such as Gigabit Ethernet and high-end storage controllers.

Windows XP SP2 RC2 was released on June 15th.

Intel i915P (Grantsdale) chipset for both Prescott and Tejas is expected to be released on June 21st. Grantsdale will feature support for DDR333/400 and DDR-II 400/533 memory, a totally redesigned graphics core, "NewCard" expansion card format for wireless communications and support for PCI Express and Serial ATA through the ICH6 'South bridge'. Grantsdale will support Socket 775. It is expected to perform similarly to Intel's previous high-end chipset, the i875, but will be targeted in the mid-range segment, replacing the i865 series. It should be noted that Grantsdale does not support AGP.

Intel i915G chipset was released on June 21st. The i915G is the integrated graphics version of the i915P, featuring a DirectX 9 compliant graphics core. Grantsdale's integrated graphics run at a clock speed of 333Mhz and features 4 pixel pipelines, giving a theoretical 1.3Gpixels/sec fill rate. The pixel pipelines feature support for the Pixel Shader 2 standard of DirectX 9, with DX9 Vertex Shader support being provided by the CPU. Grantsdale integrated graphics also provides a 400MHz integrated RAMDAC, MPEG-2 decoding assistance (although IDCT is performed on the CPU), support for an external TDMS and dual monitor / TV Out support.

Intel i925X (Alderwood) chipset was released on June 21st. Alderwood is expected to be the performance version - containing Intel's PAT technology - of the Grantsdale chipset and the successor to the i875 Canterwood chipset. Alderwood is expected to be around 5% faster than the i875 and Grantsdale and will provide support for ECC memory (although reportedly not in the first stepping of the chipset). It should be noted that Alderwood does not support AGP.

Intel ICH6 south bridge was released on June21st. ICH6 is the successor to ICH5, featuring support for PCI Express (16x and 1x) ports, 4xSATA ports and 8xUSB2 ports. ICH6 will be available with optional on-board Wireless (/W) and on-board RAID (/R). The RAID version of ICH6 (ICH6/R) will provide a Matrix RAID mode allowing RAID 0+1 (Striping and Mirroring) with just 2 drives.

PCI Express, formally 3GIO and before that Arapahoe, is will be introduced in various chipsets - notably Intel's - in June. PCI Express is a Serial I/O point-to-point interconnect standard consisting of a number of 'lanes', each of which provides a bandwidth of 2.5Gbits/s (0.31GB/s). If more than one lane is present (i.e. 2x and above) the data stream is multiplexed across all the available lanes. PCI Express is fully scalable, with plans to produce 16x (5GB/s) and even 32x (10GB/s). PCI Express is expected to be used for the connection between the North and South Bridge chips, replace AGP and be a interconnection standard between the South Bridge and expansion devices. The most common devices will be 1x and 16x, with 1x busses being used for low bandwidth peripherals such as network cards, sound cards etc and 16x devices being used for high bandwidth devices such as graphics cards. A 1x card will have a 10W power limit, 8x and 16x cards can accept up to 40W and a 32x card can carry up to 100W (and will therefore replace AGP Pro). Initially we are expected to see a combination of standard PCI and PCI Express slots, on motherboards. A standard configuration would be a 16x slot, a 1x slot and a number of standard PCI slots. PCI Express devices also have the advantage of being hot-swappable, and external PCI Express ports will allow the connection of high-speed external peripherals such as external IDE/SCSI controllers or network cards. The "Big Water" form factor/platform standard will include a modular architecture, allowing PCI Express cards to be inserted in a similar way to Game Console cards.

Intel Pentium 4 Prescott 3.2, 3.4 & 3.6Ghz in the LGA-775 form factor was released on June 21st.

Intel CPU Price changes occurred on June 21st. See the Intel CPU Prices page for more details.

Intel Pentium M 715 & 725 were released on June 23rd.

Intel Celeron D was released on June 24th at clock speeds of 2.53Ghz (325), 2.66Ghz (330), and 2.8Ghz (335). Celeron D is based around a Prescott core with reduced L2 cache and features 16KB of L1 cache, 256Kb of L2 cache and SSE3. These processors will run on a 533Mhz FSB. Initial benchmarks show that Celeron D is, in fact, a very respectable performer - unlike the previous Northwood-based Celeron - providing good competition for AMD's Athlon XP.

Intel CPU Price changes occurred on June 24th. See the Intel CPU Prices page for more details.

Intel Xeon DP (Nocona) IA32 64-bit CPU was released on June 28th on a 0.09 micron process at clock speeds between 2.8 and 3.6Ghz. Nocona is the server version of Prescott targeted towards the high volume Dual Processor market. Nocona features an 800Mhz FSB, contains 16KB of L1 cache, 1Mb of L2 cache, improved branch prediction and Intel's HyperThreading II technology. Nocona also features Intel's Demand based Switching technology to dynamically adjust the processors power usage depending upon CPU demand. Nocona is Intel's first x86 processor to support Intel's EM64T (formally CT & Yamhill) 64-bit extensions. These extensions are compatible with AMD's 64-bit extensions.

Intel E7525 (Tumwater) chipset for Nocona was released on June 28th. Tumwater features support for an 800Mhz FSB speed, PC3200 DDR and PC2-3200 DDR-II and PCI Express x16 and x1. Tumwater also features a PCI Expressx8 port which can have a x16 interface to enable nVidia's SLI graphics card technology.

ATI Radeon X800 SE was released through OEMs, notably Dell, in Late June. The Radeon X800 SE is a cut down version of the Radeon X800 Pro / XT. Although it is expected to have the same RV420 core as the X800 Pro and X800 XT, it is expected that only half of the core's pixel pipelines will be used (i.e. 8). It is also expected to feature lower clock speeds and interface to 128Mb DDR1 memory.

Q2

Intel Mobile Prescott 2.8A, 3.06A, 2.2Ghz is expected to be released in Q2.

Intel Mobile Celeron (Prescott) 2.53, 2.66, 2.8, 3.06Ghz is expected to be released in Q2. This processor is expected to contain 256Kb of L2 cache.

Previous Roadmap   Next Roadmap