Dear Sirs, Has anybody here tested the MSP430 core performance ? I'm interested in DMIPS / 1 MHz . I saw TI Application Report, named MSP430 Competitive Benchmarking of William Goh, but not every test that is mentioned there is clear for me. For example, what does it mean Cycle Count for Dhrystone Analysis ? It appears there on the graph that MSP430 has "100 000 cycles". 100000 Cycles for what ? My purpose is to compare MSP430 performance vs Cortex M3 performance (1.25 DMIPS / 1 MHz) Thanks in advance, E.L.
MSP430 core performance
Started by ●November 24, 2009
Reply by ●November 24, 20092009-11-24
"elil" <elil1975@yahoo.com> wrote in message news:1518d4e9-b71b-4d5e-8353-cae08b55bb50@h2g2000vbd.googlegroups.com...> > For example, what does it mean Cycle Count for Dhrystone > Analysis ? It appears there on the graph that MSP430 has "100 000 > cycles". 100000 Cycles for what ? >For a reasonable overview of the Dhrystone benchmark see: http://en.wikipedia.org/wiki/Dhrystone -- Chris Burrows CFB Software http://www.cfbsoftware.com
Reply by ●November 24, 20092009-11-24
On Nov 24, 8:43=A0am, "Chris Burrows" <cfbsoftw...@hotmail.com> wrote:> "elil" <elil1...@yahoo.com> wrote in message > > news:1518d4e9-b71b-4d5e-8353-cae08b55bb50@h2g2000vbd.googlegroups.com... > > > > > For example, what does it mean Cycle Count for Dhrystone > > Analysis ? It appears there on the graph that MSP430 has "100 000 > > cycles". 100000 Cycles for what ? > > For a reasonable overview of the Dhrystone benchmark see: > > =A0http://en.wikipedia.org/wiki/Dhrystone > > -- > Chris Burrows > CFB Softwarehttp://www.cfbsoftware.comHello Chris Thank you for your answer. Actually, I've already traveled to Wiki, but since the issue had remained unclear for me, I turned to forum. BTW, I was sure that the 1st response I would get is something like "Go to Google/Wiki ... ". (-: Best Regards, E.L.
Reply by ●November 24, 20092009-11-24
elil wrote:> Dear Sirs, > > Has anybody here tested the MSP430 core performance ? I'm interested > in DMIPS / 1 MHz . > I saw TI Application Report, named MSP430 Competitive Benchmarking of > William Goh, but not every test that is mentioned there is clear for > me. For example, what does it mean Cycle Count for Dhrystone > Analysis ? It appears there on the graph that MSP430 has "100 000 > cycles". 100000 Cycles for what ? > > My purpose is to compare MSP430 performance vs Cortex M3 performance > (1.25 DMIPS / 1 MHz) >Since you don't want a "google" response, here's the other standard c.a.e. response - it depends. Running Dhrystone benchmarks is fine for comparing how fast two cpus are at running Dhyrstone code, for the particular compiler you choose, and using the particular (and unrealistic) compiler flags enforced by "official" Dhrystone tests. In other words, they are only useful within a very limited context - perhaps to get a rough idea of the /relative/ performances of two similar devices from the same company. For real code, it depends entirely on the application. For example, if your code involves a lot of 16-bit simple arithmetic, the msp430 is not bad. But if you switch to 32-bit values, the msp430 performance will half - while the M3 performance will be very similar. So you are stuck with either testing things yourself, reading about cycle counts for common operations in the data sheets, or very rough guestimation. The M3 will do more per MHz than the msp430, but the mps430 will do more per mW. That's about the best general answer you can get.
Reply by ●November 24, 20092009-11-24
On 24 =D7=A0=D7=95=D7=91=D7=9E=D7=91=D7=A8, 12:25, David Brown <da...@westc= ontrol.removethisbit.com> wrote:> elil wrote: > > Dear Sirs, > > > Has anybody here tested the MSP430 core performance ? I'm interested > > in =C2=A0DMIPS / 1 MHz . > > I saw TI Application Report, named MSP430 Competitive Benchmarking of > > William Goh, but not every test that is mentioned there is clear for > > me. For example, what does it mean Cycle Count for Dhrystone > > Analysis ? It appears there on the graph that MSP430 has "100 000 > > cycles". 100000 Cycles for what ? > > > My purpose is to compare MSP430 performance vs Cortex M3 performance > > (1.25 DMIPS / 1 MHz) > > Since you don't want a "google" response, here's the other standard > c.a.e. response - it depends. > > Running Dhrystone benchmarks is fine for comparing how fast two cpus are > at running Dhyrstone code, for the particular compiler you choose, and > using the particular (and unrealistic) compiler flags enforced by > "official" Dhrystone tests. =C2=A0In other words, they are only useful wi=thin> a very limited context - perhaps to get a rough idea of the /relative/ > performances of two similar devices from the same company. > > For real code, it depends entirely on the application. =C2=A0For example,=if> your code involves a lot of 16-bit simple arithmetic, the msp430 is not > bad. =C2=A0But if you switch to 32-bit values, the msp430 performance wil=l> half - while the M3 performance will be very similar. > > So you are stuck with either testing things yourself, reading about > cycle counts for common operations in the data sheets, or very rough > guestimation. > > The M3 will do more per MHz than the msp430, but the mps430 will do more > per mW. =C2=A0That's about the best general answer you can get.Dear David, thank you for your help. What does it mean"MSP430 will do more per mW"? Do you want to say that current (or power) consumption of MSP430 is less then Cortex M3? If so, this is a mistake, since I have a few CM3 that consume less then MSP430 in every Power Mode. For example, MSP430F5524 consumes 2.5uA in Standby mode with RTC. The 1st one of my CM3 consumes only 0.9uA and the second 1.2uA. In full run mode MSP430F under 3V, 8 MHz consumes more then 300uA per 1 MHz vs 200uA of Cortex M3. So it looks that I will have a benefit from CM3 performance and from its power consumption as well. Best Regards
Reply by ●November 25, 20092009-11-25
elil wrote:> On 24 נובמבר, 12:25, David Brown <da...@westcontrol.removethisbit.com> > wrote: >> elil wrote: >>> Dear Sirs, >>> Has anybody here tested the MSP430 core performance ? I'm interested >>> in DMIPS / 1 MHz . >>> I saw TI Application Report, named MSP430 Competitive Benchmarking of >>> William Goh, but not every test that is mentioned there is clear for >>> me. For example, what does it mean Cycle Count for Dhrystone >>> Analysis ? It appears there on the graph that MSP430 has "100 000 >>> cycles". 100000 Cycles for what ? >>> My purpose is to compare MSP430 performance vs Cortex M3 performance >>> (1.25 DMIPS / 1 MHz) >> Since you don't want a "google" response, here's the other standard >> c.a.e. response - it depends. >> >> Running Dhrystone benchmarks is fine for comparing how fast two cpus are >> at running Dhyrstone code, for the particular compiler you choose, and >> using the particular (and unrealistic) compiler flags enforced by >> "official" Dhrystone tests. In other words, they are only useful within >> a very limited context - perhaps to get a rough idea of the /relative/ >> performances of two similar devices from the same company. >> >> For real code, it depends entirely on the application. For example, if >> your code involves a lot of 16-bit simple arithmetic, the msp430 is not >> bad. But if you switch to 32-bit values, the msp430 performance will >> half - while the M3 performance will be very similar. >> >> So you are stuck with either testing things yourself, reading about >> cycle counts for common operations in the data sheets, or very rough >> guestimation. >> >> The M3 will do more per MHz than the msp430, but the mps430 will do more >> per mW. That's about the best general answer you can get. > Dear David, thank you for your help. What does it mean"MSP430 will do > more > per mW"? Do you want to say that current (or power) consumption of > MSP430 is less then Cortex M3? If so, this is a mistake, since I have > a few CM3 that consume less then MSP430 in every Power Mode. For > example, MSP430F5524 consumes 2.5uA in Standby mode with RTC. The 1st > one of my CM3 consumes only 0.9uA and the second 1.2uA. In full run > mode MSP430F under 3V, 8 MHz consumes more then 300uA per 1 MHz vs > 200uA of Cortex M3. So it looks that I will have a benefit from CM3 > performance and from its power consumption as well. > > Best RegardsWhen I said the msp430 will do more per mW than the M3, that is of course a generalisation - all such comparisons are just rough generalisations. It is highly dependent on the particular devices you pick, what you want to do with the core, what other peripherals you are using, and the complete system as a whole. There are certainly going to be circumstances where an M3 will save power compared to an msp430, but for simpler tasks it is more likely to be the opposite.
