VSX7 64 bit Rendering Test

[Davidk]

There seems to be a mis-perception abroad about just what a 64bit computer is, in terms of the processor and assuming otherwise like-for-like (number of cores, clock speed, hyperthreads or equivalent etc). So, some thoughts on the topic from one who - once upon a time - built computers for a living.

Simplified, 64bit simply means that it's data bus bandwidth - in the cpu, the motherboard and any memory and other devices attached to the bus - is 64bits (8bytes) wide, as compared to 32bits (4bytes) for x86. That means a RAM search/access for say, 4K bytes, will take half the time required by a 32bit machine, because each access gets or writes twice the amount of data for every access. It also allows the cpu to directly - ie in RAM - address considerably more memory than 4gb, which is the upper limit for any 32 bit machine (2 raised to the 32nd power = 4gb). More RAM means more data can be in fast memory, accessible much more quickly than if it was in virtual storage (pagefile on a disk) or otherwise in the data store, on a disk. And much more RAM, more so.

The cpu instruction decoder and arithmetic unit, although 64bits wide, are compatible with and run 32bit instructions and data. In effect, the lower 32bits of the 64bit bandwidth are identical to the 32bit item. A commercial necessity.

So the only advantage (over a 32bit unit) a 64bit cpu machine gets is in memory activity, intended to be RAM, or similar data accesses from a parallel operating input device. Serial items doing a conversion to 64bits, eg a SATA disk drive or a usb memory stick, probably won't get any advantage at all because of the serial to parallel conversion function. Actually, what happens is that the interface chips do the work and interrupt the cpu to come to get the results when available, and meanwhile the cpu can do other things; but this applies to 64bit and 32bit machines. If the data manipulation in a 64bit cpu involves register bit shifts, that specific operation may actually take longer (more clock cycles) than on a 32bit machine because there are potentially more bits (twice as many) to shift.

From this it follows that on a processor with limited RAM (say 4gb which seems to be common even on 64bit machines) any project that has a lot of data (robertoz reported 72 clips in 12 mins of video) on a disk probably won't see any advantage over a 32bit unit. OTOH, a project which is that long but with a limited number of files and can fit into the available RAM will no doubt see a significant increase in user-perceived processing speed. Bear in mind that rendering creates the final file small bits at a time (elements eg frames appended to the resulting file, building it up slowly until finished) so the available RAM would also need to be big enough to handle the output result as well as the input sources and workspace.

During beta tests on X7 there were tester comments made about the 'blinding speed' of the x64 version. That may well have reflected the results obtained using a small test file. I have the 64bit version of the released software installed on a laptop, and whilst its faster on small projects, and does reduce an annoying periodic pause (32bit - about 6 sec, 64bit about 2sec) when reading smart packaged files from a usb2 device, I would not call its performance improvement 'blinding'. Just better on what I have used it for so far.

Various articles I have seen lately have questioned the "more RAM is better" mantra that developed 20 years ago, citing faster and improved technology, both hardware and software. I doubt it, based on things those articles did NOT say about the test environment used. RAM is still orders of magnitude faster than any hard disk is. Therefore, if a video project uses a lot of data files it necessarily has a lot of storage i/o, and just that element alone will probably overwhelm the benefits of a speedy processor - unless it does not have to perform much or any disk i/o. 64bits and lots of RAM are a given for gaming machines and video is not far short of game requirements.

So, for those who are 'testing' 64bits on practical data, look to the numbers and size of that data and how much memory you are using when comparing results.

FYI

Davidk

[aussiejoe]

HI

thanks for the detailed explanation.

So would you say that Corel's press statements such as:
"With new 64-bit power that delivers more speed, more quality and superior performance across the board"

these are more marketing spin than real world benefits?
cheers

[Davidk]

Mostly.
Look at the words . . . .

Speed and performance are both words for the same thing. What's been discussed so far is the relative speed of rendering, 32bit versus 64bit versions of the software, on the same project files. And its a toss up - faster and slower instances have been cited. What I tried to do was remove the "mystique" of 64bit - that it must always be faster or better, because that seemed to be the expectation from posts on the board. If the process involved only memory to memory operations - 64bit would blow it away. But the real world isn't like that . . . Further (something I didn't cover earlier), rendering in HD WILL take longer than rendering in standard DVD, because altho the number of frames per sec is the same the amount of data per frame is exponentially rising (double the resolution of DVD means 4 times the amount of data per frame). It's why HD files are so much larger than DVD (mpg) files. But on a comparison of render performance of the same vsp 32bit vs 64bit, this should not matter. However if you are used to render times for DVD and are now trying 64bit and HD, the HD part alone will make it longer.

Quality - is a function of how good the programming is, got nothing to do with the data bandwidth of the processor or system electronics. If the result you see is good quality, it should be the same for both versions. There's been mention of poor coding, left-over debug tools etc for the 64bit version in a prior post - may have an effect, but it seems unlikely. These days, coding is done using higher level languages, with the compiler managing the conversion of the language to executable code for a specific processor: and there are not that many compilers around. And compilations would start with the same source code with switches for 32bit or 64bit processors. If there's any poor coding, it should show in both the 32bit and 64bit versions. Compilers are not efficient code producers - typically, a compiled hi-level language version of a conditional statement might take 100 times as many instructions as a directly written assembler sequence. It's why there are 'efficiency' routines to improve the result, and you will often see them directly referenced when installing or upgrading Adobe software like Acrobat reader or Flash. But those efficient coding methods take a lot longer to write and debug, and thus overall higher level languages are cheaper to write. The cost in terms of hardware (memory) to run them is yours, and obviously it takes longer to execute 100 instructions than to execute 3 or 4. Extrapolate that over a whole OS and package like VS and you can see how the end result can be slow . . . . .

Davidk

[apostlejohnau]

Hi Guys..

Re rendering speed. I have pro x6 32 bit and went for the 64 bit pro x7.. I found the difference to be amazing! It's smoother and a LOT faster/stable.. ( I'm rendering some 10 -12 layers deep sometimes as by the way) Basically the rest of pro x7 is pretty/less than average but the speed difference is impressive. Obviously a lot of you are not experiencing this.. I have a top end machine and maybe its something to do with the program being able to hook in to the processor? I'm no IT guru by the way . I'm using the Ivy bridge i7 with 32 gig ram on a 64 bit Win 7 Pro.

PS: Rendering from AVCHD from Sony camera and outputting usually to mpeg HD 720 or 1440
Cheers
John

[Davidk]

And I think this result from apostlejohnau is an obvious benefit of having a lot more RAM that the processor can directly address.

Davidk

[Davidk]

The purpose of this post is to illustrate, via a standard memory test, the difference more (64bit) memory makes. I'd been pondering a memory upgrade of my 64bit laptop for sometime. And after participating in this thread I decided it was time to put money where mouth was. So, ordered an upgrade to max of machine, which was really just from 4gb to 8gb, and what I wanted was an upgrade that matched the existing hardware, ie was not slower or introduce wait states that reduce overall performance.

First item was to run a windows experience test (control panel, performance information and tools) to get a reference number of the current system performance, which includes the processor, memory graphics and hard disk. My interest here was in the memory results.

This next part is FYI mainly make the point that vendor memory recommendations need some verification if you have the means to do it. Finding the right memory cards for this upgrade was an exercise in itself, and in the process found that HP and the memory distributors often don't know what is right. Being led astray by that is discomforting. In the process of this, most memory distributors want a model number, which isn't obvious unless one knows or is advised to look under the battery, to recommend upgrades. One dist then offered memory that was 20% slower (PCL3-10600 instead of -12800) and almost all of them offered memory for machines I had never heard of nor were listed in websites, citing them as equivalents. By now very wary, I asked a computer designer friend for his choice of maker, and then opened up the laptop and looked on the current 2x2gb modules and found micron part numbers. A website search turned up this was a member of a family of memory that had slow, fast and faster modules in 1, 2 and 4gb sizes. Datasheets were available, that was where the speed variation was evident in detail - this one just happened to have a speed table for memory family members in it.. Picked the 4gb modules in the same speed family and tried to order them. More roadblocks. This time, Micron only sells to manufacturers, but Crucial was their retail arm. Asked crucial for equivalent, then ordered the upgrade (2x4gb modules) locally (but not without the dist again trying to sell something different), and it arrived today.

The upgrade installed OK, machine booted and ran without errors and I then adjusted the page file (virtual memory) to windows recommendation (1.5 times installed RAM), re-booted and repeated the windows experience test. The result - RAM performance increased from 5.9 to 7.2 just as a result of that memory increase alone. Graphics performance also affected by memory increased over the prior test results.

Davidk

[lata]

Hi David

Thanks for the information, I assume that by increasing your ram to 8 Gb that Video Studio has also improved in performance.
I have 4 Gb installed at the moment and have been thinking of upgrading to 8Gb.

Just need the time and money to do that, a recent purchase of a bluray writer has emptied my pocket for the time being. But as I say I have been considering upgrading for some time.
I know when I went from 2 Gb to 4 Gb that I did notice the difference in response time using Video Studio, render times did not change, at least not enough to get excited.

[Davidk]

Hi Trevor,

For my specific machine, VS X7 starts and runs fine on the enlarged memory. I won't know just how much difference more memory has made to VS X7 until I try to run some student clips - smart packaged projects loaded via a usb3 port which generally have been an issue - next term (starts 23 April), but while I expect it to improve the amount of extra memory (8gb is the max the machine is designed to have installed) probably won't be enough to offset the limits imposed by the hard drive or the external memory interface where the video clips/audio/images/text files are stored. A point that I've been making - expressed differently - is that the hard drive storage device for programs and data is usually the slowest thing in any machine. More usable RAM is only possible in a 64bit computer and then for improvement to be really noticeable there has to be enough of it to overcome HDD delays.

In terms of VSX7 performance versus RAM, the post by apostlejohnau on 29 Mar in this thread is illustrative of the sort of effect one can get if the machine is designed for a lot of RAM, whereby the limits of data storage devices can be largely overcome.

Based on my experience, identifying the 'right' 64bit memory modules for your machine I would suggest that you:
1. note what your manual says about RAM. For example, mine says SO DIMM (the size/form factor) and in two places "DDR3-1333Mhz dual channel support (DDR3-1600 Mhz downgrade to DDR3-1333Mhz)" and "PC3-12800Mhz shared for each of 2gb and 4gb modules" and these statements identify the interface spec (DDR3 or PC3), clock speed and bandwidth ( -12800) the memory is designed to run at. (this was why - altho I didn't have part numbers at the time - I got suspicious of vendors offering lower clock speeds and lower bandwidth modules mentioned in the earlier post). Note also that dual channel support (summary from a link supplied by the manufacturer's user forum) is in the baseboard hardware, not the memory itself and is a means of designing the memory to work with 2 sets of controllers - one is active doing a read/write while the other is setting up for the next access. Supposedly this method gets a memory performance benefit of about 10%. Given that the baseboard is designed for operation at a particular speed - evidenced by the part that is already there - putting a slower module in is only going to stress it because the baseboard will be running it faster than it was designed to be. And of course slower is . . . slower.
2. remove the battery (to minimise accidents)
3. open the memory compartment and (without removing anything if possible, but you may have to unlatch a module to allow it spring up to a point where the part number can be seen) identify the module part numbers and if possible the manufacturer. If the manufacturer isn't stated, note the logo on the sticker. Make sure you get it all - mine for example was MT8KTF25664Hz-1G6 and the first sequence identified the capacity and the second the speed and bandwidth required. The logo identified the manufacturer as micron - something I only confirmed when having searched the part number I arrived at the micron website. Knowing I was finally on the right road helped the attitude a lot.
4. web search for the memory part number, identify the manufacturer and if possible the manufacturers datasheet (pdf doc, a link with the part number). The datasheet will tell you whether there are other members of that memory family (generally, a datasheet covers several members eg 1gb, 2gb and 4gb in slow, fast and faster versions, with part numbers for each one given). If the datasheet isn't available or there's not a larger capacity module shown, ask the manufacturer for an equivalent in the size you want. eg, 4gb modules equivalent to the current 2 gb modules part xxxxx and the current interface spec (the PC3 numbers I mentioned above that are stated for your machine).

Expect that you might have a lot of email to-and-fro to get to the point where you know with some certainty what the right upgrade parts are,and who to order them from. I was warned about this but really didn't expect that I'd have over 2 dozen emails and about 6 webforum Q&A's to arrive at an order point.

Davidk