Rise mP6 Reviewed

Rise mP6 - A New PR Processor

  The Rise™ mP6™ processor is new performance rated. sixth generation microprocessor designed to exceed the Basic PC needs of low-power, performance multimedia Windows applications at an exceptionally realistic price. The architecture of the Rise mP6™ processor is the first superscalar, superpipelined, Pentium MMX-compatible design that features 3 integer units, 3 MMX units, and a fully pipelined floating point unit . Its innovative circuitry maximizes processing per clock cycle at a minimum power consumption level, Rise hopes the mP6™ processor will be your next choice for cost-effective, power-efficient desktop Windows systems.

Providing software for and compatibility with Microsoft Windows, Windows NT, MS-DOS, OS/2 and UNIX along with the large installed base of x86 operating environments and applications; the  mP6™ processor is completely hardware compatible with existing Pentium processor systems, and with Socket 7 or Super7 based systems.

Rise™ mP6™ Processor Architecture

mp6Block2.gif (31310 bytes)

The 3-way superscalar architecture of the mP6™ processor, utilizing three integer and MMX pipelines, enables instruction tripling. With instruction tripling, three instructions can be issued and completed in a single clock cycle, greatly impacting system throughput. Aggressive instruction pairing and tripling rules, combined with advanced branch prediction algorithms, ensure that three pipelines of the processor are utilized to their fullest. Superpipelined, each integer pipeline contains six stages. Supporting the integer pipelines is a three stage non-interlocking instruction fetch pipeline and a two stage non-interlocking data cache pipeline. These supporting pipelines and the six-stage integer pipelines provide superpipelined efficiency to reduce pipeline stalls, and increase the aggregate instruction completion rate. To effectively manage its pipeline architecture, the mP6™ uses innovative branch prediction and instruction decode techniques. It's large 512 entry Branch Target Buffer (BTB), and an 8-entry call-return stack provide the base from which advanced branch prediction algorithms operate. By storing return addresses in its deep call-return stack, the mP6™ processor is able to more accurately predict return targets, even from nested subroutines.

The mP6™ implements Dual Instruction Stream Execution (DISE), a unique feature that allows instructions to be decoded from three streams and issued from two different instruction streams simultaneously. Up to eight unresolved branches may simultaneously be operated upon by instruction fetch and integer pipelines. Pairing and tripling rules combined with the ability to decode from multiple streams also helps to increase the total instruction throughput.
To reduce pipeline stalls and to improve data movement efficiency, the Rise™ mP6™ processor resolves data dependency conflicts with the following features:

Each of these features reduces typical read after write, write after read, and write after write data dependencies. This allows the mP6™ to simultaneously execute pairs of instructions that other processors cannot.

The integrated Floating Point Unit  of the mP6™ is designed as a four stage parallel operation pipeline. The Rise mP6™ processor can execute up to two floating point operations in the same cycle and the three pipelines in the MMX Unit can execute up to three instructions per cycle in parallel and can resolve data dependency (twinning) to increase performance. The MMX Unit is geared for high performance multimedia such as image processing, video conferencing, and soft modem.

The mP6™'s 16 KB L1 split cache design improves instruction and data handling, based on  individual needs. The instruction cache, uses a split-line access mechanism to ensure that up to 16 bytes of data can be accessed with each cycle, even if the data spans multiple lines.
The data cache utilizes eight interleaved banks to minimize conflicts between accesses. The "D" cache is dual ported (it's tags are triple ported) to allow two read accesses, two write accesses, and a snoop access - all in the same cycle. The "D" cache also contains a filtered tag prefetching mechanism to prefetch data while minimizing bus traffic and cache pollution. There is no misalignment penalty - an important factor for multimedia (e.g., MPEG playback).

The  mP6™ processor is instruction and pin compatible with the Pentium processor implementation of the System Management Mode. The mP6™ is also compliant with the (ACPI) Advanced Configuration and Power Interface specification.
In addition to system level power saving facilities, the mP6™ processor has the following internal architectural mechanisms that minimize power dissipation:

No action is required by the system designer or the end user to realize the power saving benefits of these architectural features.


Windows 98 Performance

The test sample, provided by Specialty Tech (a web-based hardware reseller), the mP6™ has been produced and released on a .25 micron die using a 2.8v core. Judged by it's heat production, the mP6™ is indeed an extremely power efficient processor. Extremely easy to keep cool, the mP6™ unfortunately doesn't seem to overclock well.  Although it purpotedly supports clock multipliers of  2x, 2.5x, 3x & 3.5x, its recommended clock multiplier setting is 2.0x100MHz for the PR266 processor, 2.0x95MHz for the PR233 processor and 2.0x83MHz for the PR166 version. (Identified by their PR rating on the POST screen on mainboards with the proper BIOS update)  Although I did manage to boot cleanly at 3.0x75MHz, (225MHz) the system quickly became unstable and locked up frequently.

Testing Configuration for Rise mP6™ PR266 Processor

Mainboard EFA Viking mS3
L2 Cache 1024KB
Processor(s) Rise mP6 PR266 (2.0x100MHz)
Intel P233MMX (2.5x100MHz & 3.5x66MHz)
Cyrix MII 300 (3.0x75MHz)
AMD K6-2 266MHz (4.0x66MHz)
Memory 1x64MB AOpen PC100/
73.08864.P7L DIMM
Hard Drive Maxtor Diamond MAX 8.4G Ultra ATA/33
Graphics Adapter Diamond Viper 330 4MB AGP
Matrox Millennium G200 8MB AGP
Creative Blaster Banshee 16MB AGP

ZDBOp's Winbench 99 and Winstone 99 Benchmark applications were all carried out with the Viper 330 4MB AGP graphics adapter and the scores below reflect an average of three runs of each application under each individual processor...

I have to admit to being a bit disappointed in the floating point performance of the mP6™ as one of it's marketing highlights is its pipelined FPU.  It did however, blow past the Cyrix MII 300 but I was hoping to see FPU Winmarks at least matching the Intel P233MMX.  Let me point out however that the P233 was overclocked on the FSB to 100MHz.

RiseFPU.gif (13788 bytes)

While Winstone 99 performance shows the mP6™ at the bottom of the heap, I think that the processor has possibilities. The mP6™ roadmap alluded to by the folks at Rise and at Specialty Tech   ".18 micron very soon with a core voltage of 2.0v & a PR rating of 300 or better. Power draw could go as low as 4w (!) - would only need a heatsink & no cooling fan. mP6 II with 256k on die L2 cache running at CPU speed. A socket 370 version of the mP6 II (think about it - double the L2 cache of a Celeron, most likely no multiplier lock)" creates a lot to think about if Rise can get them on the market by summer.

RiseWinstone.gif (16052 bytes)

SiSoft's Sandra 98 Standard also points out the mP6™'s failings in implementing a strong FPU.

mp6Sandra.gif (23045 bytes)


3D Graphics and Gaming Performance

Again, nothing special here. The mP6 PR266, while thoroughly adequate for the market at which it is currently aimed - cost efficient low power multimedia SOHO and Home PCs - depending on your graphics card the mP6 matches up consistent with  many of the currently available socket 7 CPUs. Scores below reflect performance under Creative's Graphics Blaster Banshee, and the Quake II marks were acquired using the 3Dfx OpenGL Glide Engine...

As you can see by the marks below, even in D3D applications, the mP6 holds it own...

RiseFsaken.gif (12899 bytes)

mp6graphics1.gif (22890 bytes)

The mP6™ does provide better support for MPEG video than Cyrix MII.   Both video clips and graphics editing seem well handled my the processor at its current PR rating and if Rise does manage to develop the 0.18 micron die at decent yeilds, we will no doubt be hearing a whole lot more about them in the near future.

Conclusion

The mP6™ PR266 is well up to the challenge if you're considering an extra low-cost mainboard/CPU combo for a second home PC or SOHO workstation. But as it stands now the processor has more of a novelty market than anything.

It's low power consumption and easy cooling and genuine support of the 100MHz front side bus make it a better buy than Cyrix' current MII offerings but I would look carefully at the price/performance ratio of the Intel P233MMX before deciding which CPU to purchase for a low-end super7 system.  The mP6™ displays seemingly flawless stability at it rated speed and because of that we happily endow the mP6™ PR266 with our Super7 Award but as I mentioned before, it almost uniquely un-overclockable although I tend to bet that if you are seriously looking at this particular version, you have no affectation for overclocking anyway. awardgold.gif
        (6926 bytes)

The mP6 PR266 is still at limited availability but if you're interested you can find it at truly reasonable prices at Specialty Tech as well as a few other web based hardware vendors.  And I'd expect to hear a great deal more from Rise in the future...

Last Update - March 14, 1999

© 1998/99 2000 MediaTek