I understand that with Dragon the processing power is dependent on L2/L3 cache, the number of cores, the total amount of RAM and RAM speed. In terms of price vs usability is it worth spending an extra £80 for the 8MB L3 cache in an i7-2600k compared to the 6MB L3 cache in an i5-2500k? What sort of improvement will I see with the extra 2MB?

Thanks in advance.

Although an i5 is adequate, we no longer recommend anything under an i7. We're just hearing too many bottleneck complaints with lower end computers. We also prefer quad cores to dual cores.

-------------------------

Click KB 2012 REV D to Download a 30 Day Evaluation of KnowBrainer 2012 

 

 

 

 You said: [QUOTE=Agent 9;290972] (...) at most get one of the higher clocked i5's [the 'benefits' of the i7 really are not there, especially if you have to pay for it].][/QUOTE]

I thought to buy a Tablet PC Fujitsu T902 (http://store.shopfujitsu.com/ca/EcomCA/mobileconfigurator.do?model=T902&partNumber=BTCK410000AAAAAI&lino=&cacfg=1&GPID=&SMC=259) to use DNS V 12 (and much other demanding softwares) at my work , but the highest option is an Intel® Core i7-3520M processors with 2 cores. (http://en.wikipedia.org/wiki/Ivy_Bridge_(microarchitecture)#Mobile_processors). Is this processor, with 16GB of RAM, can all the same do the job without bottleneck?

Thanks!

Unfortunately, tablet computer technology typically falls behind notebook technology because of limitations in size and weight so you're not going to experience quite the same performance with a tablet that you would a 7 pound response notebook. All we can say is make sure you have a 30 day return on your tablet and see how it works for you. A lot of people are able to successfully use a dual core i5 but it's not our 1st choice.

-------------------------

Click KB 2012 REV D to Download a 30 Day Evaluation of KnowBrainer 2012 

 

 

 

A machine with the Intel® Core i7-3520M processor and 16 GB of memory should work just fine as a dictation machine with Dragon NaturallySpeaking and the Microsoft office suite. Try to get an SSD to improve overall machine snappyness. I own a machine with the previous Sandy-bridge equivalent Intel® core i7-2640m and I use it extensively as my travel machine for email, documents and presentations.

or straight dictation with reasonably sized individual documents, unless you have an extremely fast machine for your desktop (I have an overclocked hex core with 15 MB of cache) you will probably not notice that there is a barely perceptible increase in the lag before the words hit the screen (I'm literally talking eye blinks level of elapsed time),

Just don't try to do any kind of heavyweight multi-tasking like video processing, encryption, compression etc. while you are dictating on that machine because the response time to the commands or the dictation will drop off dramatically.

I recently posted a bunch of quantitative Intel processor comparisons using Rüdiger's Dragon NaturallySpeaking performance tool. If you poke around the forum and take a look at my posts you will find half a dozen posts where I make quantitative comparisons.

on the other hand, if you just buy the machine, use it for dictation of reasonably sized individual documents and don't expect heavy duty multiprocessing you should be okay.

Phil Schaadt

 

-------------------------
DNS12 Pro BM V Large Vocabulary plus KB or Voice Computer running Win7 64 bit machines with i7-2640M to i7-3960x and i7-3770K processors plus a Sony VAIO Windows 8 machine. DBX Audio Gate with Allen&Heath mixer/USB Audio; Andrea PureAudio USB  usually with Airline 77 or Audio-Technica but also Sennheiser MD431 II, theBoom, et. al.

Someone (Stonekeeper) on another forum wrote than the expected difference of power between the Intel i5 and i7 in the Fujitsu T902 is of 10%.

 

Another (Agent 9) wrote: "Well, the base processor in the T902 is a Full Voltage i5, so it is really not going to have the issues that the ULV processors will and as the ULV processors are very prevalent in computers I would think that the comment made about needing the power of the i7 is to 'easily' get the ULV equipped people to get LV or Full Voltage equipped devices [though they wouldn't know it. I don't know if that makes sense...]... Also, the "10%" that stoneseeker is talking about will equate to about zero real world difference, the only real world difference you are likely to notice in Tablet PC's is the difference between a Full Voltage equipped device and a ULV equipped device [that is coming from my personal experience and what others say]

 Ultimately it is up to you. But the 'basic' i5 processor in the T902 will be more than you will likely need in the next 3-4 years. Your number one limiting factor will be the mobile GPU [as it is in any mobile computer] and raw CPU power should not be something you will be lusting after because of poor performance on the T902. The 'i7' thing is more of a marketing/ bragging thing than anything else.

 The base i5 paired with plenty of ram, and a SSD will be able to handle anything you might throw at it and more, albeit not as well as a Ivybridge desktop with a dedicated GPU and extras could, but being stationary is not the point of a Tablet PC so the convince of mobility greatly outweighs the performance loss [if that isn't true for you then a very competent desktop with a Bamboo/ Intuos/ Cintiq might serve you better]

 As for overclocking, you should go into it expecting to 100% for sure not be able to overclock the processor/ GPU, they pretty much designed them this way unfortunately. But if you did manage to OC I imagine heat and thermal output would exceed the capabilities of the heatpipe/ fan/ case design and result in a very uncomfortable Tablet PC [the only time a OC is needed is when you are pushing it, and that is also when a OC'd system will produce the most heat]"

 

I don't understand every technical thing write by Agent 9, but it seems like I can avoid frustrating bottlenecks with the Intel i5 processor of the T902 when running DNS 12, and than it's not necessary to get all the power of an i7 processor to avoid bottlenecks. What do you think?

Your quote above makes your request below virtually impossible to fufill. Plus there is no device optimal for "any situation". I have been very clear about measured specs for machines suitable for DNS, DNS and office/personal productivity and DNS with heavy duty development. From me that is the extent of my interest and measurements

Sorry

Phil Schaadt

-------------------------
DNS12 Pro BM V Large Vocabulary plus KB or Voice Computer running Win7 64 bit machines with i7-2640M to i7-3960x and i7-3770K processors plus a Sony VAIO Windows 8 machine. DBX Audio Gate with Allen&Heath mixer/USB Audio; Andrea PureAudio USB  usually with Airline 77 or Audio-Technica but also Sennheiser MD431 II, theBoom, et. al.

Thanks. I also understand 8GB RAM is recommended, is there any point in terms of Dragon performance in upgrading to 16GB? Cheers.

We are in the process of upgrading our desktop computer to 24 Gb of RAM because we edit video. Our notebook computer had 8 Gb of RAM and we upgraded it to 16 Gb of RAM because quite frankly, it was only $79 and we felt silly not doing it BUT we rarely if ever go over 5 Gb of RAM in our notebook computer or 6.5 Gb of RAM in our desktop computer (when we're not doing video). For 99% of the population, 8 Gb of RAM should be more than adequate. However, this is just our $.02

-------------------------

Click KB 2012 REV D to Download a 30 Day Evaluation of KnowBrainer 2012 

 

 

 

 what type of video editing software do you use? I've used Vegas and pro-cut. I like Vegas better.

Thanks for the info.

What do you mean by 'bottleneck complaints'? Where would there be a bottleneck when using an i5 that could be avoided when using an i7?

Thanks.

Veng3r,

To answer your basic question, the Core™ i7's are generally quad core, with the exception of the Core™ i7 980X's and 990X's which are, ofcourse, six core processors. The Core™ i5's and below are generally dual core processors. Where the bottleneck is in the Core™ i5's is that DNS 11-11.5 wants to use two cores simultaneously and in parallel (multithreading) to process you are dictation using two distinct acoustic models. Under this condition what do you think happens if another applications, such as Microsoft Word or Excel, or your antivirus software, yada yada yada, needs and wants access to one of the cores, gains it, and then Dragon wants to use two cores. Windows isn't going to give up the ghost to NaturallySpeaking by kicking any of the other applications out and letting Dragon in to do its thing just because Dragon requires two cores, so Dragon is going to have to wait, and wait, and wait under some conditions. I don't know about you but I'd call that a bottleneck.

It's like knowing that the drive from Los Angeles to Las Vegas requires at least 14 gallons of gas, but you only fill your tank up to 7 gallons. What happens when you get stuck in the middle of the desert after you run out of gas

Remember, your operating system has absolute control over your hardware. When an application wants something it asks Windows, and once Windows has made its decision and assigned priorities and placed everything in the queue, you stay there until your time is called. BestMatch III mitigates this to a certain extent, but if you're going to use BestMatch IV you're better off going with at least 4 cores or more. This is why we recommend a Core™ i7 over a Core™ i5. It's not just the CPU model number, it's the number of cores coupled with the amount of L3 cache and the total amount of RAM.

Otherwise, the basic difference between 8 Mb of L3 cache and 6 Mb of L3 cache is not that significant in the big picture, but the number of cores in this case can be significant based on the above.

Hope that makes it a little clearer.

Chuck Runquist
Technical Project Manager
VoiceTeach LLC
Home of VoicePower®: We don't make Dragon NaturallySpeaking, We make it better!

"Aiming for the moon and missing it is better than aiming for the ditch and hitting it."   - Author Unknown

-------------------------

Thanks for the clarification. I didn't think DNS was multithreading capable? Someone quoted this to me elsewhere:

 

 So where is the confusion arising?

We don't know where the previous quote came from or when it was made. Hopefully, whoever made that statement, was referring to NaturallySpeaking Ver. 10. If not, they are clearly unqualified to be commenting on the inner workings of NaturallySpeaking.

Beginning with Ver. 11, Nuance doubled NaturallySpeaking's sampling rate and introduced Best Match IV which feeds dictation through 9 algorithms simultaneously through 2 cores on your computer. We know this to be a fact because if you have an older single core processor, NaturallySpeaking will not allow you to create a Best Match IV profile. If you attempt to move a Best Match IV user profile to a single core processor with an installation of NaturallySpeaking 11, you will not be able to open it. The only possible confusion we see is with the idiot you quoted. Sorry to be blunt but the Internet is full of good intentioned unqualified advice.

You can install NaturallySpeaking on an i3 computer but we recommend against it. Although we frequently state that an i5 is adequate, if you were a professional mechanic, would you want to purchase adequate knuckle busting wrenches from Kmart or professional tools? We have probably received at least 100 phone calls about DNS latency issues. On further investigation, we nearly always find that our customers decided to save $80 by purchasing an i5. To us, the price difference just isn't worth the drop in efficiency if NaturallySpeaking plays a major role in your workflow. Please feel free to quote us

-------------------------

Click KB 2012 REV D to Download a 30 Day Evaluation of KnowBrainer 2012 

 

 

 

The quote is from Chuck way back in April 2009 and seems to be re DNS9 maybe 10

http://www.speechcomputing.com/node/2475

BTW I have a wide range of processor configurations and I dictate with DNS for 8 plus hours per day. Based on my experience a quad core i7 with 8 MB cache is probably the best price/performance processor specs if you earn your living with DNS. 

Phil Schaadt

-------------------------
DNS12 Pro BM V Large Vocabulary plus KB or Voice Computer running Win7 64 bit machines with i7-2640M to i7-3960x and i7-3770K processors plus a Sony VAIO Windows 8 machine. DBX Audio Gate with Allen&Heath mixer/USB Audio; Andrea PureAudio USB  usually with Airline 77 or Audio-Technica but also Sennheiser MD431 II, theBoom, et. al.

The i5 has 4 cores http://en.wikipedia.org/wiki/List_of_Intel_Core_i5_microprocessors#Based_on_Sandy_Bridge_microarchitecture and so does the i7 http://en.wikipedia.org/wiki/List_of_Intel_Core_i7_microprocessors#Based_on_Sandy_Bridge_microarchitecture The i5 2500k is the same as the i7 but minus hyperthreading and 2MB less of L3 cache. Therefore I still cannot see the benefit of going for the i7 unless 2MB of L3 cache if worth £80 as hyperthreading has not been mentioned. Does dragon use hyperthreading? Thanks.

the way Intel names their processors can be very confusing. it is unbelievable that the different processors have the same name; even though some of them have two cores while others have four cores. there are quad core core i5 processor and there are those with only two cores. Here is a comparison chart from Intel itself:

http://www.intel.com/content/www/us/en/processors/core/core-i5-processor-office-use/Corei5Business.html

 

-------------------------

Dell Inspiron 7520 SE,  2ghz core i7(Ivy bridge), 8gb ram, 750gb Seagate Momentus XT SSD/HD Hybrid hard disk. SpeechWare 3 in 1, Sennhesser ME3, Buddy USB 6G, Windows 7 Ultimate, Office 2007.

It is hard to make a quantitative judgements when you have not described your specific usage scenarios. If you are simply using DNS as a "typewriter replacement" to write unformatted prose you will not need much processing power. On the road I use a i7-2640M dual core 4MB cache machine for email and basic text processing and it is serviceable but with noticeable lags.

If you have complex workflows or use many programs at the same time, plus live dictating a demo script while presenting WebEx simultaneously, while emailing and processing large encrypted files it's difficult to have to much processing power or too much network bandwidth. However most folks peak usage scenario is no where near as resource intensive as mine.

Finally if you dictate as many hours per day as I do, the cost differential in your example is less than two cents per hour of dictation for one year.

Perceived value is a personal judgement.

Phil Schaadt

-------------------------
DNS12 Pro BM V Large Vocabulary plus KB or Voice Computer running Win7 64 bit machines with i7-2640M to i7-3960x and i7-3770K processors plus a Sony VAIO Windows 8 machine. DBX Audio Gate with Allen&Heath mixer/USB Audio; Andrea PureAudio USB  usually with Airline 77 or Audio-Technica but also Sennheiser MD431 II, theBoom, et. al.

Thanks for the replies everyone.

I'm comparing the i5 2500k (4 cores, 6MB L3 cache) to the i7 2600k (4 cores, 8MB L3 cache). I'll be using Dragon Professional as I need to control the computer with my voice and use my hands as little as possible due to a physical disability. I'm currently not working as I am unable to use my hands or a computer without significant pain so I'll be learning how to use the software so I can apply it in a work environment, possibly learning web development, but I need to understand the software and its limitations before I can decide how I can use it to enable me get back to work.

I therefore envisage that I'll be running web development software, web browsing and using it as a home theater PC. In addition to the programs listed and the obvious anti-virus/firewall etc I may have another 2 or 3 simple programs running in the background that run fine on an old Pentium 3 so there shouldn't be too many intensive programs competing for processing power.

I want the software to be responsive and usable, as if I was using a mouse and keyboard IE. I want to be able to tell it to start a program and then switch to and use another while its starting up, tell it to click on something and have it respond without any real noticeable delay compared to a mouse. I don't know if the i7 will make any real world difference in my situation, if it just knocks a few milliseconds off the response time and its not noticeable then its not worth it but if it makes a difference to how usable the software is then it is.

So in my circumstance is the 2MB of L3 cache going to make any noticeable difference? Thanks for helping me clear this up.

I am partially disabled with multiple pinched nerves.  I must take anti-inflammatory medication continually to control the pain. Even with medication I can perform only limited amounts of intermittent typing before my pain levels increase.

Without Dragon NaturallySpeaking I would not be able to work. With Dragon NaturallySpeaking (plus VoicePower and KnowBrainer) sitting in a special ergonomic chair I can be productive for hours and run a boutique systems integration firm.

It takes a real commitment to learn how to do virtually all of your computer work by voice. If you are actually going to earn your living using Dragon NaturallySpeaking you should make sure that your microphone, sound card and PC do not get in your way.

Combining the workload you described with your performance goal of:

I strongly recommend a machine with a processor like the i7-2600k and 8mb of cache as well as much memory as your machine can support.

I also strongly recommend that you consider a product such as VoicePower both for structured training in the proper use of DNS as well as additional commands to better control your machine. As you seem to have programming skills, you might also want to get a copy of KnowBrainer to facilitate your creation of custom voice commands. If you are actually fully disabled KnowBrainer is free.

Finally, spending 30 minutes a day reading this forum will provide you with all sorts of useful tips and techniques for making productive use of Dragon NaturallySpeaking.

Philip Schaadt

 

 

-------------------------
DNS12 Pro BM V Large Vocabulary plus KB or Voice Computer running Win7 64 bit machines with i7-2640M to i7-3960x and i7-3770K processors plus a Sony VAIO Windows 8 machine. DBX Audio Gate with Allen&Heath mixer/USB Audio; Andrea PureAudio USB  usually with Airline 77 or Audio-Technica but also Sennheiser MD431 II, theBoom, et. al.

Veng3r,

That's an extremely old posts that predates DNS 11-11.5.  In fact, it's one of mine.  In short, it's out of time out of place.  The basic concept of multithreading and what can and can't be done via parallel processing, which is multithreading, is correct.  However, Dragon has found a way via BestMatch IV to multithread the recognition process by processing your speech simultaneously using two cores, which amounts to two threads (without hyperthreading) and four threads (with hyperthreading), at the same time using two distinct and unique Acoustic Models.  Once the processing of both threads is complete, then Dragon merges the results to produce the "BestMatch".  In other words, instead of nine possible matches, BestMatch IV provides 18 possible matches, from which it chooses the nine best or most probable and selects the top match (BestMatch).  That is multithreading.

Phil is basically correct. The original post from which you quoted was back when questions were being asked about multiprocessing/multithreading and multicore usage by applications towards the end of DNS 10.1. DNS 11-11.5 and not been released at that point. So, while the concept is correct, its relationship to DNS 11-11.5 is basically a non sequitur. That is, it doesn't apply.

Regardless, that is the only place in Dragon where multithreading can be used effectively.  Otherwise, what I said in the original post from which you quoted still holds.  That is, any linear process that requires the results of the previous process in order to continue will not benefit from multithreading.  In addition, every application uses whatever core(s) is/are available unless you use the "affinity" setting to restrict access to specific cores.

What I said originally was that Core™ i7 processes are generally quad core and Core™ i5 processors are generally dual core.  By that I mean that the majority of processors included under the Core™ i7 heading use four cores and hyperthreading.  On the other hand, the majority of cores under the Core™ i5 processor list are generally dual core, but at least a dual core processors have hyperthreading, the quad core processors do not.  And yes, hyperthreading is important for DNS performance.

However, hyperthreading is a processor specific function.  The only thing that hyperthreading means is that each core in the processor is capable of processing
two sets of instructions simultaneously.  This is application independent.  If a processor supports hyperthreading all applications benefit, but the performance of Dragon NaturallySpeaking benefits significantly. Without hyperthreading, Dragon performance takes a hit.

Some while back, we conducted a raw performance test based on audio files provided by Rüdiger. If you click on the following link you will see just how processors make a difference in terms of pure performance with DNS 11-11.5.

DNS raw performance testing

Chuck Runquist
Technical Project Manager
VoiceTeach LLC
Home of VoicePower®: We don't make Dragon NaturallySpeaking, We make it better!

"What you are aware of you are in control of; what you are not aware of is in control of you." - Anthony de Mello

-------------------------

Chuck,

9 is just the maximum number that the display in the correction menu is set to. Effectively, it will come up with 30 or more very often, and you can see them all when using my tool, by clicking the "Phrases" button after recognition has been performed. Additionally, this is nothing new in version 11 (BestMatch IV), it is just the same in 10 or with BestMatch III in 11, therefore, multithreading is not involved with the number of alternatives being evaluated.

Rüdiger

 

-------------------------

Well, it's past the point where we can make any changes in the code, but we can still make changes to the Easter Egg!

Rüdiger,

OK. I see your point and I buy into it up to a point. However, there are too many variables to make the absolute statement that in many cases you can get "30 or more" results. In fact, in 15 min. of testing I never got more than 22 and in most cases I got 12 or less.

Here's one to scratch your head over. I have never dictated anything out of Lincoln's Gettysburg address. However, I dictated the first sentence in DNS 11.5 Professional with the Speed vs. Accuracy slider set to 100%.

Number of Last Result Phrases: 1

1: four score and seven years ago our fathers brought forth on this continent ,\comma\comma a\determiner new nation ,\comma\comma conceived in liberty ,\comma\comma and dedicated to the proposition that all men are created equal .\period\period

Then I set the Speed vs. Accuracy slider to 0% and repeated the same sentence.

Number of Last Result Phrases: 1

1: four score and seven years ago our fathers brought forth on this continent ,\comma\comma a\determiner new nation ,\comma\comma conceived in liberty ,\comma\comma and dedicated to the proposition that all men are created equal .\period\period

Just to make sure that it wasn't a quirk, I dictated the second sentence with the exact same results:

With the speed vs. accuracy slider set to both 0% and then to 100%, I got the following in both cases.

Number of Last Result Phrases: 1

1: now we are engaged in a\determiner great civil war ,\comma\comma testing whether that nation ,\comma\comma or any nation so conceived and so dedicated ,\comma\comma can long endure .\period\period

So, while I concede that you can get more than 18 possible results and that my original post stands corrected on that basis, that doesn't present an argument against multithreading in DNS 11-11.5. I should have perhaps separated the definition of multithreading in BestMatch IV from the results issue. Regardless, BestMatch IV uses two distinct Acoustic Models (i.e., you can tell that they are distinct because they have different sizes). Whether they are unique from one another can't really be determined. Nevertheless, using two cores and two different Acoustic Models to simultaneously (i.e., in parallel) analyze each utterance is multithreading independent of the results simply because you're using two different threads (cores) to process the same utterance using two different Acoustic Models.

It doesn't get any more multithreading than that.

Chuck Runquist
Technical Project Manager
VoiceTeach LLC
Home of VoicePower®: We don't make Dragon NaturallySpeaking, We make it better!

When you have eliminated the impossible, whatever remains, however improbable, must be the truth. - Sir Arthur Conan Doyle

-------------------------

Chuck,

you might just take my word for it, and you were just wrong stating it that way. It is not that I go looking ever so often at it, but many many times I have seen 30 or more alternatives showing up altogether, and I guess the highest amount amount of alternatives that I have ever noted using my tool was 36. Besides, the maximum amount to display is defined by a setting related to the SDK, but this is no restriction imposed on the recognizer obviously which I have only become aware of by looking into the SDK, and it has been that way since when I did the program which was while there was no version 11 around.

Since I don't have the time, nor the interest, in providing current material to give you some evidence, I just looked at one of the performance log files (automatically generated by my program on demand) of which I have numerous having received them from various people helping me with it by providing me with material. So here is an excerpt of one from Mark (MDH) running 10.1 Medical.

Testing Performance Tool, Copyright R.Wilke 2010
Log created: 7/11/2010, 12:09 PM
Currently loaded user profile: MARK DANIEL HAIMES
User training time: 128 minutes, 30 seconds

Processor:
IntelR CoreTM2 Duo CPU     E4500  @ 2.20GHz
Architecture: 32-bit
L2-Cache Size: 2048 KB
Physical Memory: 2044 MB

Operating System:
Microsoft Windows XP Professional SP3

+++ Clear Box has been pressed +++
-----------------------------------------------
*** Starting logging utterance, with Speed vs. Accuracy set to 100 % ***

DgnEngineControl_event: Utterance begin  started 12:8:49:224
DgnEngineControl_event: Engine recogniton  started 0 msecs total plus
DgnDictEdit_event: Dictation Recognition  started 0 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 1927 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 2118 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 2357 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 2500 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 2883 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 3185 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 4475 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 4890 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 5113 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 5320 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 6148 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 6244 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 6992 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 7056 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 8266 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 8951 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 9127 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 9254 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 9334 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 9445 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 9557 msecs total plus
DgnDictEdit_event: Text changed   started 9764 msecs total plus
DgnEngineControl_event: Utterance end  started 9780 msecs total plus

Utterance Length:     9019 msecs
Overall processing time:     9780 msecs
Ratio processing time vs. utterance length:  1.08

Time between utt start and text display:  9764 msecs
Time between engine rec start and text display:  9764 msecs
Time between text rec start and text display:  9764 msecs

30 phrase(s) in collection

Number of words in utterance: 30
Length of entire utterance: 9019 msecs

 

Hope this will make it clearer now.

Rüdiger

 

-------------------------

Well, it's past the point where we can make any changes in the code, but we can still make changes to the Easter Egg!

Rüdiger,

I concede your point. I learned something I didn't know before. Regardless, what you're saying makes perfect sense.

Perhaps it's because I'm using my Core™ i7 980X Extreme. Or, perhaps it's the phrases that I'm using. Nonetheless, I just found it tending not to produce that many results. I did do so, but they were rare on my system. Go figure!!! While I would concede that you might be a better judge of this than I, would it not the possible that this is another area where processor type, L3 cache, etc. might have an impact?

Let's not make mountains out of mole hills.

Chuck Runquist
Technical Project Manager
VoiceTeach LLC
Home of VoicePower®: We don't make Dragon NaturallySpeaking, We make it better!

"There are painters who transform the sun to a yellow spot, but there are others who with the help of their art and their intelligence, transform a yellow spot into sun." - Pablo Picasso

-------------------------

Chuck,

no making any mountains intended, but here is the first performance log file (an excerpt from it) that I received from you:

 

Testing Performance Tool, Copyright R.Wilke 2010
Log created: 10/1/2011, 5:35 AM
Currently loaded user profile: Rudiger BMIII
User training time: 11 minutes, 33 seconds

Processor:
IntelR CoreTM i7 CPU       X 980  @ 3.33GHz
Architecture: 64-bit
L2-Cache Size: 1536 KB
L3-Cache Size: 12288 KB
Physical Memory: 12279 MB

Operating System:
Microsoft Windows 7 Ultimate  64-bit
 SP1

+++ File transcription invoked +++
-----------------------------------------------
*** Starting logging utterance, with Speed vs. Accuracy set to 50 % ***

DgnEngineControl_event: Utterance begin  started 5:34:57:797
DgnEngineControl_event: Engine recogniton  started 3 msecs total plus
DgnDictEdit_event: Dictation Recognition  started 7 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 1624 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 1688 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 1806 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 1874 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 1933 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 2031 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 2088 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 2151 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 2254 msecs total plus
DgnDictEdit_event: Phrase hypothesis   started 2307 msecs total plus
DgnDictEdit_event: Text changed   started 2381 msecs total plus
DgnEngineControl_event: Utterance end  started 2404 msecs total plus

Utterance Length:     8870 msecs
Overall processing time:     2404 msecs
Ratio processing time vs. utterance length:  0.27

Time between utt start and text display:  2381 msecs
Time between engine rec start and text display:  2378 msecs
Time between text rec start and text display:  2374 msecs

30 phrase(s) in collection

Number of words in utterance: 19
Length of entire utterance: 8870 msecs

-------------------------

Well, it's past the point where we can make any changes in the code, but we can still make changes to the Easter Egg!

I full support the views expressed by Philip Schaadt in his response to your forum question. Get a computer with as much processing grunt as you can afford. Personally, I would get Core i7 with 16 GB of RAM, because it's just nice to have. That being said: the Core i5 2500K, if being primarily used for speech recognition, with 8 GB of RAM will be more than adequate. Perhaps a far more important consideration would be your purchase of a soundcard and microphone that is optimised for speech recognition; and I would suggest that you spend much more time concentrating on that aspect. Unless the Dragon NaturallySpeaking speech engine can 'hear you' clearly; no matter how much processing power you have DNS is not going perform adequately. As you use £80 in your original posting, I take it you are in the UK. KnowBrainer/Lunis does ship to the UK. In fact, I took delivery of a package from him yesterday. Alternatively, you may wish to purchase from monkey8 a prolific contributor to this forum who is located in West Sussex.



-------------------------
DNS 12.0 Professional, Windows 7, Intel Core i7 2630QM, 16GB of RAM. Second-Generation SpeechWare 6-in-1.

I have not been able to get my hands on an i5 machine to test with Rüdiger's performance tool. One of the other posters on this forum was unhappy with their i5 performance but I can't remember what processor they were using. I strongly believe that trying scrimp $150 on a $2000 Fujitsu T902 machine is very economically shortsighted.

The quote above from Agent 9 has does not directly address straight DNS recognition speed as "raw CPU power" is the first order, most important factor in DNS performance and GPU has nothing to do with DNS. BTW, OC on a laptop is only for extreme gamer hobbyists.
Phil Schaadt


-------------------------
DNS12 Pro BM V Large Vocabulary plus KB or Voice Computer running Win7 64 bit machines with i7-2640M to i7-3960x and i7-3770K processors plus a Sony VAIO Windows 8 machine. DBX Audio Gate with Allen&Heath mixer/USB Audio; Andrea PureAudio USB  usually with Airline 77 or Audio-Technica but also Sennheiser MD431 II, theBoom, et. al.

 

So, if I understand you, get 10% more power with a i7 processor option on the Fujitsu T902 is worthing the $150 demanded when the goal is to use DNS 12 without experiment bottleneck. Is-it that?

To be completely accurate someone would need to measure both processors with Rüdiger's DNS performance  measuring tool. I don't have access to the i5 processor so I need to go on public benchmarks.  If you think saving $150 on a $2000 notebook by going with a lower speed processor go right ahead but I sure wouldn't do it.

 

Here's why, first when you go from the i5 to a i7 the increase in cache is from 3 MB to 4 MB or 33%. by the way the Fujitsu site has the wrong cache sigh specified for the i7. The amount of on processor cache is one of the primary determinants of performance, particularly with respect to any kind of multitasking which could be something as simple as encrypting/decrypting or compressing/ decompressing an email.

 

On these relatively low-power processors when you're executing these kinds of tasks on the machine it will dramatically slow down Dragon NaturallySpeaking which can be very annoying if you are using Dragon to control your computer while you're waiting for one of these background tasks complete. This is not a theoretical guess on my part. I measured a variety of typical background tasks'  impact on DNS performance on the machines I mentioned above.

 

The published Passmark CPU benchmarks (Google it) give the i7- processor a roughly 20% higher score than the i7 even though the difference in effective processor speed is about 5%. Passmark score is all about CPU processing capability and it is influenced by the amount of onboard cache. I found passmark scores a reasonable first-order proxy rejecting relative Dragon NaturallySpeaking performance

 

 My position is that these small machines are okay for simple dictation, for example emails or smallish documents. I use a similarly powered machine on the road. cutting back on the processor capabilities will very likely mean you will not be able to use BestMatch V because with limited resources is to be quite slow.  There is a step function below which best match V  just becomes unacceptable. As I said above I have not been able to measure a lot of i5 machines but you are likely at that threshold.

 

 It's your money but again I think that you risk wasting $2000 on a machine that you have a decent probability of not being happy with.

 

I'm a very twitchy guy and use an overclocked hex core machine as my daily workstation and I'm okay (just barely)  with my low powered machine for doing lightweight work on the road.

 

Plus, if you can take a heavier form factor, $2000,  which is the price of the Fujitsu machine linked above, will buy a very fast machine that will use BestMatch V with virtually instantaneous dictation speed and will be able to handle some medium duty multi-tasking.

 

Why anyone would blow $2000 on an undepowered machine and try to scrimp on the processor is way beyond me but it's your money.

 

Phil Schaadt

 

-------------------------
DNS12 Pro BM V Large Vocabulary plus KB or Voice Computer running Win7 64 bit machines with i7-2640M to i7-3960x and i7-3770K processors plus a Sony VAIO Windows 8 machine. DBX Audio Gate with Allen&Heath mixer/USB Audio; Andrea PureAudio USB  usually with Airline 77 or Audio-Technica but also Sennheiser MD431 II, theBoom, et. al.

It's because I am also a twitchy guy than I take my time before to buy another computer to understand these computer concepts that make the difference between satisfaction or insatisfaction in running DNS. When you say Phils:

 

you think maybe to a big desktop computer, but since clients at my work must fill and sign forms (my client's folders are all computerized), I didn't find any other means more useful and simple than to pass my Tablet PC to my clients in my office when I am in front of them. If there is an also or more useful and simple way to do that and to use DNS efficiently, I am very open to know.

I needed a fast computer to dictate a 300 page book in a few weeks. After reviewing comments on this forum, I chose a stable HP Z420 engineering workstation with an E5-1620 Xeon processor with a 10mb L3 cache running at 3.6 Ghz. It runs 4 cores however Xeon processors do hyperthreading so there are 8 effective cores. During optimization and profiling, DNS uses 4 processors. This basic machine has 6 GB ram operating at 1.6 Ghz which leaves about 2 GB free during operation with DNS. The hard drive is SATA 3 running at 6 Gb/s. I chose a Rode Podcaster USB microphone mounted on a cheap mic stand to my left adjusted 2 inches from my mouth and to the side. This mic has a head phone jack for monitoring of voice quality and level with ear buds. It works flawlessly. Latency is under 1 second no matter how long a sentence. I can dictate 12 pages per hour in my first week after training.

I use 2 1600x1200 NEC LCD2090UXI 20" graphics monitors. The system works with DNS with a minimum of errors. Some pages have no errors. Even complex medical text jargon and rare words are normally returned without error. Voice quality is near broadcast level and lower frequencies lost in headset mics are in full power. The computer system ran about $1,200 and the used monitors were purchased on Ebay for $30-60 each. The microphone is $200. I appreciate very much the forum information on technical details that I was able to use in selecting this setup. Thanks.

By the way, The Rode Podcaster mic is a dynamic microphone which is rare among USB devices. Dynamic microphones have much lower sensitivity to background noise. They need to be used close to the mouth. The primary DNS benefit is all you get is just the spoken word. No computer noise, air conditioner sounds, and background noise in the building. You can hear the background if you turn up the monitoring knob to maximum, however in operation, DNS does not hear it.

-------------------------
Tech Writer

Michel,

Pay attention to what Phil is saying. The bottom line is that most of the reviews of processors on other forums are based on raw performance as Phil notes. Most of these people wouldn't know Dragon NaturallySpeaking if it jumped up and bit them in the ass. So, take what they say with a grain of salt. The reviews that you were posting are like reading Hamlet. Full of sound and fury, signifying nothing!!!

DNS 12 BestMatch IV and BestMatch V use multicore/multithreading, which simply means that they process your speech using two Acoustic Models either in sequence (BestMatch IV using two cores sequentially) or in parallel (BestMatch V using two cores simultaneously). How well a particular processor is able to manage these, as well as being able to make the best use of the cache available and having sufficient RAM, is more important than just raw processing power. How well Dragon performs relative to any particular processor is more dependent upon the number of cores, the amount of L2/L3 cache, and the total amount of RAM than anything else. Hyperthreading also significantly benefits DNS in terms of overall performance.

The bottom line is if you want real performance out of DNS, don't cut corners. As cheap as computers have become over the years trying to save a couple of hundred bucks relative to DNS, it just isn't worth it.

Chuck

"You seldom improve quality by cutting costs, but you can often cut costs by improving quality." - Karl Albrecht

-------------------------

Chucker and Phils, you are probably right. But when Agent 9 (TabletPC Review Forum) write:

"If the full voltage i5 that is stock is 'not enough' or 'right on the verge' of "not usable" as he is implying then moving up to the i7 that is ~10% better in benchmarks [and about zero % difference in real use] then it will be just as unusable.
Neither of those processors are going to compare to a desktop processor, and while a desktop processor is 'significantly' faster/ stronger, it is really not that much faster/ stronger and pretty much anything that can be done on a powerful desktop can be done on a slower mobile device [like Bronsky said with people using Atom devices and having DNS reportedly running well enough for them]

The i5 3320M Intel Core i5 3320M Notebook Processor - Notebookcheck.net Tech and the i7 3520M Intel Core i7 3520M Notebook Processor - Notebookcheck.net Tech They both have two physical cores and 2 'virtual' cores [Hyper threading], same amount of L1 and L2 cache and only a difference in L3 cache [you would never know it unless they told you], and they are both Full Voltage 35W processors, also from the same generation of processors and all that... If one of them couldn't do something then the other cannot either... If someone tells you otherwise they are either lying to you to 'sell' you on something, or they are mis-informed and ignorant to facts.


If [and that is a big if] the T902 came with a Low or Ultra Low Voltage processor as the base option, and a Full Voltage processor option then it would make sense to upgrade; but with the difference being between a very high end Full Voltage dual core to a barely better/ different Full Voltage dual core processor there is no actual reason to upgrade other than for the sake of having it... $150 can buy a lot of extra stuff like a second AC adapter, a nicer pen like the Axiotron Studio pen or Morion LE series pen, laptop sleeve, portable speakers/ headphones, BT mouse, backpack, ect... and those will make use of the computer a lot better especially when on the go compared to a ~10% benchmark increase..."

 

then I must understand than everything said by him about performance of a laptop and Voltage is right for any situation where DNS isn't use. (?)

But same there, according your explanations, buying a Fujitsu Lifebook T902 with a i7 processor 3520M (2 cores, 4 threads, 2.9 GHz, L 4MB Cache, Bus/Core Ratio: 29) with 16GB of RAM sould cause significant latency (bottlenecks) if I use DNS 12, less than with a i5 processor 3320M (2 cores, 4 threads, 2.6 GHz, L 3MB Cache, Bus/Core Ratio: 26) and 8GB of RAM, but latency all the same. Isn't?

Everyone,
There are lots of dual core i7 processors out there with 4 MB of L3 cache so take care if you really want a quad core to google it and see the specs on ark.intel.com

-------------------------
DNS 11 Premium, Core2 Quad Q9400 4 GB Ram