1 person found this helpful
First a link to some background reading: How to get the best from a PC? Some guides...
You did not mention a budget, but the most economical route is an i7-920 overclocked. Look here for some system configurations: http://ppbm4.com/Benchmark.html Have a look how some dual W55x0 systems perform in comparison to some overclocked i7's.
AVCHD is a beast to edit, so it needs a beast of a machine. Probably the best you can buy/build now if cost is no object, is a dual W5590, but if you wait a couple of weeks the much more economical route is to get the i7-980X Gulftown hexacore on a X58 mobo.
Personally, I would not spend the money on an i7 Extreme version, but rather overclock a 920 and invest the savings in a better disk setup, which propelled my score to the top.
Assuming you go for a X58 mobo, 12 GB memory is the lower limit. 24 GB may be more future proof with CS5 around the corner, but carries a hefty price tag.
For the video card one can only guess, since we have nothing more than a 'sneek preview' of what MPE can do. However, the GTX285 or the new GTX480 seem much more attractive than any Quadro card IMO from a price/performance point.
Testing has not shown any advantage in using SSD's over conventional hard disks, apart from loading times for applications. For editing performance the test results have been utterly disappointing. If you need storage space and performance, and again if money is no object, you could consider 4 x 80 GB Postville SSD's in raid10 for OS & programs and 24 x 600 GB Seagate Cheetah 15K7 disks in raid30 on an Areca ARC 1680-iX-24 with 4 GB cache and BBM. Mind you this would set you back around $ 14K+.
You need not go to this extreme, but some more disks in your raid would be beneficial. Do invest in a proper raid controller like Areca. Also have a look here: http://forums.adobe.com/thread/525263
Thanks - immensely helpful. At the risk of pushing my luck I've a couple of follow-ups if you don't mind:
[ BTW - budget = should not be the limiting factor but clearly is limited! Ideally I want to build a mega machine for £5-6k, British Pounds, excl. monitor etc but I would happily spend less if I don't need to spend this much! ]
1. Over-clocking. How "serious" is this? By which I mean, if I want a robust reliable workstation am I putting stability or longevity at risk by over-clocking? Maybe better to buy more processor power and clock at the manufacturer-designed rate? What is the downside, I assume there is one or everyone would do it?
2. Timing wise I can wait - is the i7-980X worth waiting for? Would a single cpu system with this processor be comparable in processing power to the current top-of-the-line dual Xeon systems?
3. Memory. Sounds like I should just bite the bullet and go for 24GB?
4. Any idea why Premiere Pro system requirements would state that it can't be installed on a flash memory disk? [ BTW - my interest in the solid state disk was partly speed - you told me not to worry about this! - partly about noise, I was told that 10'000rpm disks are very noisy ]
5. Video card. Is there an easy answer as to why a powerful graphics card seems to make almost NO DIFFERENCE at all in this scenario which, you would think, was an incredibly graphics-intensive application? Is there something wrong with my logic here? Also, is the MPE stuff worth thinking about? (i.e. making sure my new machine is future-proofed for it)?
Lastly - your note and your speed table has got me seriously thinking of building a machine myself. I am an original techie but these days a little rusty. Assuming I can cope with system-level technical stuff (i.e. I can't write code any more but I can figure out if components are compatible and plug them in!), is this a route you would recommend? I have been put off in the past by things like:
- Most basic, how to choose a case which fits my chosen motherboard!
- Since cooling seems to be a very big deal, how is an amateur supposed to calculate power and cooling requirements and get the right case with the right power and the right cooling?
- Does everything "just fit" or will I end up drilling holes in metal and fashioning elaborate bits of bent aluminium to hold things in place?
- How do basic things like clock speed work - do I need to buy an oscillator myself or is the clock / overclock something which is on the mobo and selectable with a link?
- What does one do about things like bios, drivers for all the new hardware, etc?
Not expecting your help with all of this but a steer would be brilliant.
Some further reading, this time on overclocking: http://forums.adobe.com/thread/529916
At current prices my system is less than € 3 K and despite being about a year old still performs nicely. At my most used OC of 3.7 GHz it still is performing nicely with a PPBM4 score of 38 secs and has some serious storage capacity. It is explicitly shown with all components in the PC Guide, near the bottom of the article, including the CPU cooler, which, if I had to do it again today would be the Noctua NH-D14.
The i7-980X Gulftown is around 40-50% faster than a similarly clocked Bloomfield because of the hexacore architecture versus the quad core structure of the Bloomfield. This also applies to all Adobe programs because they do support multi-threading greatly. The drawback is that it is expected to be around $ 1 K, which is around 4-5 times more expensive than the i7-920.
Memory is a personal choice but keep in mind that going from 12 GB (around € 300) to 24 GB carries a hefty price tag (around € 1400) at current prices and is only worth it's extra cost if you do a lot of Dynamic Linking between AE, EN, PS and PR or have these applications open at the same time. Also keep in mind that if you choose a server type mobo with dual sockets for Xeons, your choice of memory is pretty much limited and the price is much higher, because of the ECC requirements and overclocking is not realistically possible due to BIOS limitations.
The noise of 10K RPM disks is higher than SSD's and 7200 RPM disks, but I do not consider them really noisy, although with only one 10K disk and 16 7200 disks internally, I am not surprised that I do not notice the noise of this one disk. Why Adobe suggests that SSD's are not supported, I just don't know.
PR does not currently use the video card to accelerate anything apart from some very limited number of effects and transitions, so effectively it is not used for performance gains. That is the reason that a $ 30 card is not noticeably slower than a $ 3K video card. With MPE this may be a completely different matter, if things are really as good as the previews suggest. To be sure this new system is future proof, make sure that you use Win7-64 Pro as OS and that you get a supported CUDA capable video card. The list is currently pretty short, but expected to grow as market introduction comes nearer.
For building your own system not much is needed. In the PC Guide I mentioned earlier I give a number of suggestions on cases (the one I use is again in my system overview, a Lian Li PC-A77B) and there are links to a PSU calculator, and a number of other relevant items on cooling, so I suggest you read that article carefully.
There is no need to start drilling holes or any other mods. Luckily Win7 comes with about all the drivers you may need and a mobo comes with the BIOS, so there is hardly anything left to do apart from installing the OS and programs. The third guide about tuning Vista largely applies to Win7 as well.
I know these articles will take some time to digest, but a lot of people found it worthwhile to read and the posts there contain valuable info as well.
Last one - I am probably going mad but you make reference to "the PC Guide" but I'm not sure I have the link to that guide?
In post #1 I gave you a link in the first line. If you follow it, there are three more links, a PC guide, a storage guide and a tuning guide. This is the direct link to the PC guide: http://forums.adobe.com/thread/433549
Take a look at the results on my Premiere Pro Benchmarks (PPBM4) and see that generally moderate overclocking on a single processor can readily be accomplished relatively easily and safely today whereas a few years back it was much more difficult and risky. Also your Xeon "workstations" cannot be overclocked with any simple adjustments.
Adobe has never had any GPU assisted effects execpt a few rarely used ones and they currently in CS4 do not have any GPU assisted encoding, except expensive third party encoding. That all should change when CS5 is released maybe even in March???
The WD VelociRaptor 10,000 rpm drives are not noisy. Actually they are 2.5 inch drives in a 3.5 inch holder. Actually I have several SAS 15,000 rpm drives and I do not even consider them noisy.
I have an array of 8 SSD's and in RAID 0 (for testing) they do not even come close to the performance of 8 1TB drives in RAID 0. SSD's cost much more and provide much, much less capacity. I also experimented with SSD's for the OS/Applications with no positive results over the VelociRaptor.
I'm in much the same postion as Alisair, I want a workstation for AVCHD files. We differ in one respect, however, in that I have neither the knowledge or inclination to build a workstation myself. Nor do I even understand most of the jargon used in these discussions. I do have budget of +,- $4,500 to have a workstation built by a professional.
I create videos for Internet consumption. Most are of less than 20 minutes duration, some as long as 90 minutes. Most have fewer than five layers of video and two of audio. I currently use PPro CS4.
After having read most of the discussion topics referred to herein, I would like to ask of you:
- Should I invest in a system now, or wait for another couple of months until new components and/or CS5 is released?
- If you think it's appropriate now, would you please be so kind as to detail the brand and model numbers of the components, including storage, you would recommend within my budget that I can understand and pass on to a builder. Should you have a recommendation on a builder as well, I would be pleased to know that as well.
to the OP.
Dual Xeon i7 would be better than a single core i7 by far, providing you buy the X5550 or better as these are the first of the 6.4 QPI series.
otherwise the 975 (or over clocked 920 )would be on par with say X5530 or better (as you lower the GHz of the Xeons)
Xeon particularly if you have long play
make sure it has DDR 1333 not 1066.
no reason to wait. there is nothing coming out from Intel thats better other than in April or so the 6 core processors. these will be very expsensive and not worth the $. prices wont drop for awhile either.
present X58 (single or dual) will allow for you to drop in the 6 cores later when they have reasonable pricing.
how are you encoding for the internet? flash, H264, (sorensen?) Quicktime GPP or Wirecast streaming?
You have made these statements before, but no benchmark test seems to support that statement. OK, one can say all benchmarks stink, but how else can one compare systems if you can't rely on benchmarks? I know that my middle-of-the-road system outperforms dual W5590 systems on the PassMark test and on the PPBM4 test for a fraction of the cost of a dual W5590 system. I am not convinced of the reality of your statement, but maybe you can convince me with either PassMark results or PPBM4 results from a dual W5590 system. Once that is done, the next step is to convince me that it really is worth the quadruple investment or even more.
this is what i do for a living. we benchmark every system before it ever leaves. our tests are difficult for the end user to duplicate (too big to download)
1) Passmark is a joke it is in no way a good example of how video editing will perform
just like it would not indicate how Audio would perform.
2) as i have already stated the presently used benchmark (bills) is antiquted and in no way is representative of how HD (not HDV which is not HD espeically if ingested via firewire its as compressed as DV)
will work on todays systems.
while AVCHD is not that heavy of an HD file (compared to say 2k) it requires far more than footage from a firewire source.
FYI most of the guys i talk with on the phone i usually talk out of buying a xeon so this is not an "upsell" thing.
its just the facts.
That is a nice way out, but what if I called your benchmark a joke, because.... and you can't use it outside of your office, so it has no value whatsoever for comparison purposes. So we just have to believe your statement which can not be substantiated. And you still do not get the purpose of the PPBM4 benchmark, despite Bill's explanation. It has nothing to do with ingest, over firewire or otherwise.
Sorry, your arguments are pretty lame IMO.
LOL. its very evident you are not working with HD material if you were you would completely get it.
the advice you give is based on your non HD experiance. you can not give the best advice to someone working with HD if you dont know from experiance yourself.
the same applys for those who transcode to H264. or other formats like it.
another example was the thread with the guy who has the RTX2. if you are going to DVDs the RTX2 makes alot of sense.
but you poo pooed it becasue its not your work flow.
next you are going to say Decklink and Aja are not worth it.
Firewire ingest is dead man.
i will have a test thats duplicatable by an end user within a week or so
assuming you have an ingest card that has HDMI/component and a blueray player.
"So we just have to believe your statement which can not be substantiated. And you still do not get the purpose of the PPBM4 benchmark, despite Bill's explanation"
And yet you dont feel the same way about this statement from PPBM4 home page: "There is no HDV, AVCHD or any other format version experience has shown that for hardware evaluation that plain old SD is the best format."
This is most definitely unsubstantiated since AVCHD is exactly what everyone is having issues with editing and playing back even on single I7 desktops. Why do you think that is? Could it be due to the compression level? Could it also have something to do with the large increase in pixel data from the 1920x1080 resolution scans instead of the old SD. Maybe it also has to due with the increase in data that Progressive frames have over interlaced. Maybe also to due to the CPU's having to shelve allot of data to cache and ram during the decoding phase to process effects on the video stream and then continue decoding after each frame completion. Maybe it has something to due with the Data process through the CPU pipeline multiple times before it outputs on the bus to the video due to all of this. Lets not also forget the compression algorithm is allot more complex for AVCHD at 1080 30P than DV NTSC 29.97i and that has to be processed through the CPU most likely Floating Point processing the majority.
So based on this how can you concur that DV in your experience is best for hardware evaluation? DV has stopped stressing the CPU/RAM/Subsystem over 3 years ago. Your dealing with a test that is so light on a current workstation that the OS is trying to put half the hardware in C state or park cores and that is the test you want to use for hardware evaluation? Lets not even get into the fact that a laptop from 2 or 3 years ago can edit DV and people are not having issues editing DV because everyone is transitioning to HD. Right now the Predominant aquisition codec is AVCHD and people would like to edit that natively rather than transcode to save time. Since that is exactly what is causing most people their problems forcing them to search for new hardware, then why wouldn't you use that to hardware eval systems. Maybe then you might see Xeon Workstations outperforming I7 Desktops at the higher end as they should. I honestly can't believe you can even consider DV SD a better eval tool than AVCHD just on the principles of the Codecs and computers. That is a stretch no offence.
Another newbie that has his opinions.
I have comfort in the many compliments that users of PPBM that I have help tune and speed up their systems so they can comfortably edit the formats of their choice.
So dismiss and insult rather actually counter with evidence as to why DV SD is still a better Benchmark. Your benchmark was great a few years ago. Like everything else in the Computer and Video industry though technology evolves and benchmarks have to evolve to really test the improvements. Using DV now to benchmark is nothing more than a Ghz and Drive test because that is all you are really testing. The rest of the system is really not being used to it's potential since the processing is so light. It's no different than benchmakring a game from 5 years ago on systems then and today. You are going to see very little improvements on the current system in peformance due to the point of diminishing returns since the software doesn't make use of the system beyond what it needs to run at best peformance for the code. Beyond that, the system runs idle. Does that mean a system now is no faster than a system from 3 years ago? No of course not. It means the software just doesn't use the hardware that is available. Compare your car and a Nascar. Both of you drive on a highway and the speed limit is 65. So since that is the fastest you can drive, does that mean the Nascar doesn't out perform yours. Of course not. That just means the Nascar is not using it's potential beyond what it needs to go 65MPH. You can be confident as much as you want in what others say about your test but the facts are technology has changed and your benchmark no longer really tests the hardware potential in a system to it's full potential. Like it or not that is the case.
This is not a race it is a tool! If you do not like it don't use it.
I like your benchmark allot, just not the codec material/project used anymore. I just dont like the fact people are using it to say this configuration is better and this is what you should build and not the obviously more powerfull workstation configuration because it doesn't perform better. That is not true. The benchmark just doesn't make use of the hardware. I get calls all the time, I was sold this I7 desktop to edit AVCHD and it still bogs down for what I do. I shake my head at it because Adobe multithreads so well and can use allot more than 8 cores unlike FCP and other editing platforms. If people only have a budget to build a single I7 desktop themselves then that is great. If they are looking to do more and need greater performance because of AVCHD then they should look at building workstation Xeon I7's until CS5 releases and parallel processing changes everything for video.
Goodness I had no idea there was so much pent-up frustration on such an apparently benign topic! All the posts have been immensely interesting to me (genuinely), seeing people who actually know what they are talking about debate the issues.
Meanwhile per my original post:
1. Harm has absolutely convinced me to build a workstation - thanks, Harm, I'm looking forward to it.
2. I now no longer have a clue whether an overclocked single quad-core cpu, a normally-clocked six-core cpu, or a dual cpu with a couple of top-of-the-line Xeons is either a) the best overall, b) the best price/power combination.
Can I ask you professionals for one more "vote" on this one? There's no doubt when you start pricing up a dual Xeon system the bucks get quite scary quite quickly and I also seem to note the maximum FSB speeds are lower which presumably is a big factor in memory-intensive applications? On the other hand, I am going to build with CS5 and MPE in mind, so I would be gutted if I spend lots of £££'s on a single i7 machine and it turns out not to have the horsepower to take full advantage of the new tech.
So! Back to my original main point: which CPU(s) should I opt for if I have a good budget, not unlimited but good, want phenomenal performance, to take max value from the developments in CS5, and I am editing (at present only) AVCHD video?
Promise I will sign off after this and leave you guys to it!
Yet again a lot of assumptions that are all wrong. Your credibility would be much improved if you were not making false assumptions and false statements. One thing is right, I don't use a crappy consumer format like AVCHD. For the rest you are wrong.
[irrelevant inflammatory comment deleted]
For your info, I have been using HDV, XDCAM EX and XDCAM HD for years. Not only DV. And exporting to H.264 BR and MPEG2-BR.
In the future, get your facts straight.
I didn't realise AVCHD was crappy..?
What should I be using to be current with my HD these days?
To be more accurate, AVCHD is only used on consumer cameras with lousy ergonomics (too small, no shoulder cameras), no XLR, no exchangeable lenses, too small sensors, no manual focus, zoom or iris, etc. There is no decent camera that uses AVCHD. If it were any good, we would have seen professional cameras with that format, but good cameras only use XDCAM EX, XDCAM HD or P2, if we leave out Red for the moment.
i dont use it.
but for you or others to continue to point to it like its valid for HD and MISLEAD people into thinking xx system is better than yy system has got to stop.
this is how i steer people generally. and while its usees a very wide brush its usually very acurate.
Hobbyiest Core i7 1156(P55) platform 3 drives no raid
Weekend warrior (1-2 or less per month jobs) usually weddings has a day job. Core i7 X58
3 drives. OS, and 2 sets raid 0 (and sometimes no raid if thier format or time does not need)
Pro: Dual Xeon 2.66GHz and higher serious storage.
Broadcast or 2k: gets even higher spec with crazy raid arrays (not unlike what Harm has)
(and thats not counting audio drives if using a pro audio interface)
while its a broad brush its acurate 99% of the time. we talk with ever client and match the need to the system with some room to grow.
and what cams they use (format) can change it quickly.
so sorry guys to state a Core i7 is better than a Xeon for every guy that comes to this forum is completely MISLEADING based on an old SD test.
there are way too many variables.
to see how benchmark progression should work see here: while its for pro audio the tests needed to change over the yrs as the older tests we no longer valid. (goes back to 03 or 04)
Mods this ok?
oddly the newest test for audio bears out the same fruit for HD video
only used on consumer cameras with lousy ergonomics (too small, no shoulder cameras), no XLR, no exchangeable lenses, too small sensors, no manual focus, zoom or iris
I can find at least 3 AVCHD cameras by 2 different manufacturers (Panasonic and Sony) that have all of those features except interchangeable lenses.
There is no decent camera that uses AVCHD.
Care to share the model and type numbers?
Eric, If you have any better suggestions on how to spot CPU intensive problems and/or Disk intensive problems while running Premiere please let me know as I shortly will have to architect a new CS5 compatible test and tuning tool when it is released. Maybe I should not call PPBM a benchmark, but instead just call it a application specific test and tuning tool which has results that can be used for designing new systems. I already have a personal version timing H.264 encoding but again with CS4 it does not produce any quantitative data (except bragging rights) that the CPU intensive MPEG2-DVD does not already provide. So why complicate a simple solution?
the methodology you use is GREAT.
just need to replace DV with real HD add heavier effects and then have several output option 1 of course being H264.
i will have more on it later.
Also you might as well throw in the Canon Mark 2 series since the video recording codec is a variant of AVCHD.
Oh and you might want to read up on your codecs some. I put the link above regarding XDCam EX codec. XDCam EX is definitely no better than AVCHD and is actually lower end since it's based of the MPEG2 layers Spec instead of the MPEG 4 Layer Spec. Read up on your MPEG Layer specifications and you will see why. XDCAM HD 422 was Sony's desperate response to Panasonics AVCIntra codec. Problem is AVCIntra is still a better codec once again due to the MPEG 4 layer specification and higher bit rate which allows for more data at less compression. The Sony XDCAM HD 422 cameras also cost allot more than Panasonic's AG-HPX300 which is probably the hottest camcorder out right now for Broadcast work. That is the only P2 Codec that is really better than AVCHD at this point since DVCPRO HD is Scaled and Old tech unless your doing 720P work.
As to the AVCHD is consumer and a horrible codec, I am sorry that is absolutely false. It is a wonderfull acquisition codec and a good storage codec. It's just hard to edit with due to the compression level. If you need better color sampling or lower compression for editing with effects, compositing, or green screen then you can transocde it over to Cineforms codec with the cheap software Neo Scene. The results are absolutely beautiful. If you dont believe me then try it with their free trial. I really am not sure where you are getting your information from, but you really need to check it.
BTW AVCIntra is the Flavor of the year right now and the best codec out. R3D is great and has some serious advanatages. The Problem is the export options and the camcorder cost.
Mister "You know it all" as you announce yourself on your website.
Still very mediocre cameras that I would not contemplate for serious work. Sensors are too small, most are handheld, have no interchangeable lenses. Since you know it all, you ought to consider that there is not a single AVCHD camera that is silver rated by Discovery, BBC, NGC and the like, not even bronze, wheras all XDCAM-EX and HD cameras are silver rated. That ought to tell you something, just read up on that between your and Scott's fries.
While you are at it, read up on benchmarking. You don't seem to understand that the PPBM test is not about ingest methods and you seem to parrot Scott quite a lot, but then he is paying you.
BTW, cost is never an issue for you, since Scott advertises heavily and creates a lot of confusion about your $ 20K machines (Juggernaut) when the OP asked about a $ 4-5 K machine. You act very mysterious about your 'benchmarking' and have not given any proof to the contrary that my opinions are wrong. I consider both your approaches as commercial hype. Maybe you can register a couple more of your employees to enhance your parroting.
you are becoming unprofessional, personal and insulting now.
try debating on an intellectual level and not like a teenager.
And stop beating around the bush, and keep away from blatant lies.
It does not matter what ingest you use when you download a benchmark and run it.
It does not matter if you run it 10 times or 100 or 1000 times.
It does not matter when you use your CPU load at 90-100% whatever material is used.
It does not matter what format is used, when you use the disk thruput to the fullest.
The results when using AVCHD, HDV, HD, HDCAM-SR, P2, XDCAM, apart from absolute figures will gave around the same ranking. Testing has shown this to be the case. Benchmarking is never about absolute figures, in case you did not know, but about relativities, ranking.
Enjoy your fries...
i am bowing out of this thread as it has become pointless and
Now that is just silly. Do you believe every salesman that comes to your door? Did you think to ask what Cameras were used in the AVCHD material comparison they received. Did you possiblely consider the date they made that choice and what cameras were out at that time. Did you consider that they likely are a complete avid production enviroment which supports XDCAM and not AVCHD. Did you consider that they might have a deal with Sony for equipment? There are so many factors involved here and none of which have to do with the specifications of the codecs and quality of encoding. Also just because the Digital components of the camera handle the final product doesn't mean the analog components doesn't have a considerable effect on the over quality of the final product of material. You cannot tell me that the HPX 300 is a mediocre camera. The panasonic 150 has basically the same analog components as the 200. How is that a mediocre camera. The handheld comment is just plain false and completely outside the argument of quality. How does holding a camera with your hands or shoulders change the quality of encoding? That is just a blanket statement due to the handicams type series for consumers. Mind you Survivor Man used those and the quality was great so once again that is false.
Cost is always an issue especially now in this economy. That is just being blind to reality. Ask why many TV affiliates or movie studios have not updated their HD Cam equipment or Avid Studios? Answer is simple. Cost of replacing equipment, software, and training of staff. Many are still upside down on their investment and can't justify the loss of reinvesting in newer, better equipment when their old equipment ahs not been paid for. Once again a silly argument
Your comments on the benchmarks and how the material does not effect the results just shows you have a lack of understanding on how the encoding decoding process works or the entire point of benchmarks of anything. Test FCP on a Mac Pro with Pro Res and see how well it Threads. Test Adobe on a Windows and a Mac and see how well it threads on one compared to the other. Test the Threading performance difference between codecs in Adobe on the PC platform. Test how each effect threads differently in Adobe. Avid, or Vegas. Then finally Test how transcoding threads differently based on what codec you are coming from and what codec you are going to. Then add to all of this the current C state ACPI standards used by system board manufacturers and in Win 7 as well as Core Parking and how that changes the numbers for all of the previous tests. This is due to how the CPU and OS determine system load/useage and start putting hardware in what is called C States based on that. Core Parking was used to help alleviate certian issues with the previous Intel release of Hyper Threading but is causing some performance issues now with some software.
Benchmarks are exactly for getting as close to accurate logic numbers on hardware, software, or equipment that is otherwise hard to quantify. This is done to attempt to determine performance differences between hardware, software, or equipment. People use those logic numbers to determine what is best for their requirements. This makes the goal of benchmarks to be the most accurate logic numbers you can get otherwise any descision made from the skewed numbers is questionable. This is not much different than Forensic Technicians useing comparisons to determine and evaluate evidence. They are not looking for generalized numbers. They are looking for absolute logic and numbers. Otherwise their findings are completely inaccurate and pointless to those who require it to make a judgement on something. That is no different than using computer benchmarks to evaluate performance for different uses like video and how changes in technology or software effects the logic.
I did not select my nickname just as I didn't when my parents gave me one:)
I'm bowing out to this blah-blah as well unless you are willing to submit your results and show me wrong. Long winded posts do not give any insight. Tangible results may, but I'm betting you are afraid of showing your meagre $ 20 K results in comparison to some $ 3 K systems.
thanks guys interesting stuff
I apologize your questions were lost in this debate. If you are planning your system around CS5 then Video card replaces your CPU as the most important component. You need to make sure you buy a video card on Adobe's List. Most of the 200 series cards support Cuda but Adobe does not enable all of those card in their software currently. Only the ones on the list will allow the software to thread through the GPU instead of the CPU. With that being the case then your focus should be on GPU, RAM, HDD performance in that order. The GPU can be the same on either system so that doesn't decide. The ram performance is better on the I7 Xeon in raw bandwidth but you can get lower latency ram on the single I7 currently due to the Motherboard manufacturers bios on Xeon boards. For video playback latency is important and for transcoding bandwidth is king. Also lower compression codecs use far more memory bandwidth than higher compression codecs. With Higher Compression codecs latency is important due to the CPU caching so much during the Decode phase and effect processing phase. The Transcode though is mainly CPU(soon to be GPU), Ram bandwidth, and HDD throughput/latency though that changes based on the the codec your transcoding to. H264 has a much lower bitrate than uncompressed so the point of diminishing returns on HDD performance is much earlier. That is something Bill's benchmark shows very well. Keep in mind though that number will change based on the codec bit rate. Since your building for CS5 and not CS4, I would say the I7 Desktop will be the best choice and then put your extra budget into Ram and HDD config. Bills comments on HDD are accurate up to the SSD comments. SSD drives had a known Random write latency issue accept for the Intel drives due to the controller on the drive. If you buy an Intel SSD drive, you dont have the random write latency issue which means they out perform every mechanical drive. The draw back is the cost and volume size. That is something you have to measure. Since Adobe caches to the OS drive when you Dynamic link from PP to Media Encoder on Export, having a Intel SSD 160 drive is a great performance increase when encoding to H264. That is something I would consider in your config that you are building. The other thing is High performance DDR3 1600 with CL6 latency. That would be a big boost in performance when CS5 releases since the Video card caches and shadows to the Ram so much. If Xeon board manufacturers would change their Bios to emulate Single Desktop I7 board bios design then the Dual Xeon I7 would out perform here as well. Mind you this changes if AE in CS5 does not use the Mercury Cuda engine like PP will. I dont know for sure on that. Also if you use Fusion or another 3D application then you will want a Dual Xeon system since they wont use the GPU and CPU threading is the most important performance item with those applications. There are allot of things to consider here as you can see. It will not just be PP in CS5 to build around and there are still many questions on CS5 that have not been answered.
Once again see the posting on video game benchmarks. How would showing any
posting on how a system performs with an old benchmark that doesn't use all of the system prove anything. This has nothing to do with a 20K system and I can't understand why you even bring that up. You can build a Dual Xeon yourself for under 5K or even 4K. Please bother to read the posts before you bother to reply. I can take a game from 5 years ago and show you that your system doesn't outperform a laptop. Does that mean your system is no better? No it just means the software doesn't use the full scope of your system. This entire concept seems to escape you for some reason and I am unsure why.
Do you really think the Xeon 5590's are the only Xeons to buy??
I apologize it took so long to respond to your post. Others have kept me busy on posting. I will answer your last question first since timeline wise that deals with the current CS4 and the other is based on what CS5 shows. H264 most definitely would show a difference in CPU power since it offers HD resolutions making the compression level far greater. I am really unsure why your personal test would not show that if you came from HD material. Mind you AVCHD is an MPEG4 layer codec like H264 and very close on the Mpeg layer chart. Going down or up a layer with the same MPEG codec will be less intensive than going from say Motion Jpeg or uncompressed since the algorithms are very similar. This makes the data on output change far less than the original data. Put effects in however and that changes again since it now has to process most of the data completely anyway. The difference in resolution data though will still be significant as well as the CPU must decode the AVCHD in order to transcode. This should make AVCHD to H264 at HD res the best benchmark currently though going from MJPEG to H264 would really test the system. Another reason it may not show up is threading. Take a new Mac Pro for example. Put OS10.6 on it and boot camp Win 7. Then install Adobe on both. Render out a video on both and watch the threading and performance difference. You would be amazed at how poorly OS10.6 threads. Now throw on FCP on OS10.6 and test a Pro Res Transcode or render. Once again watch the threading. The performance is absolutely horrible compared to Adobe on the same Mac Pro with Win 7. What this means is software threads differently as do codecs based on how they are written and designed. Effects as well thread differently for the same reason. Read the Avid forums and people are all over wondering why Avid is not Multi Threading hardly at all at times. This also is the case in Adobe with some effects. Also Codecs thread differently. Test AVCHD in CS4 with 4.0 and then each update to CS4. You will see AVCHD threading far better in the current 4.2.1 update as well as some other codecs. Then add to all of this the Powermanagement settings in the Bios and Windows due to the current ACPI standard and Core Parking in Win 7. There are so many variables in performance just from the ACPI that could also cause what you are seeing with your personal edition. What codec are you using in your personal tests? What resolution and frame rate is the original material. That information would help me explain better as well.
Now as to your CS5 question. Adobe is breaking out into the new frontier ahead of all others in the NLE arena with their new Cuda based Mercury engine. Parallel Processing is where video will be going after everyone see's Adobe's release. Why is that is the biggest question and what does it show. If you watch the video for CS5, you see multiple highly compressed video files in multiple layers playing back realtime. How is it this can be the case when the current version has a hard time with 1 or 2 layers of R3D or AVCHD. The answer is due to how video is constant data streaming with multiple layers of information such as color, brightness, and resolution/pixel data. Then you add in the motion detection data that current compression uses and you have allot of threads of data that need to process simultaniously. Now current CPU's handle 2 to 8 threads simultaniously per CPU at up to 3.4GHz or overclocked to over 4GHz. Cuda based Nvidia GPU's however handle 192 Threads for the Quadro cards up to the FX4800 and 240 Threads for the GTX 285 or FX5800. These GPU's however only run at 576MHz to 700MHz depending on the card. So when you compare the CPU to the GPU you see that the CPU's even at 4GHz are no match for the GPU's running at 576 to 700 MHz. The reason is Threading is king for editing HD video because of all of the layers of information that have to process just to reproduce 1 frame. Now multiply that by 24 to 30 frames a second and multiply by each layer of video and it makes perfect sense.
So what does this mean for your CS5 test. You would likely want to build a test designed around adding layers and effects to an HD project that has a high compression codec. Then a second test to transcode or export that sequence out to H264 once you get to the point of bogging the system down. Verify the time taken to transcode that sequence that finally pushed the system. The only gotcha in this is the effects used. They must be effects that thread well so be careful on your selection. To me that would be an ideal test for CS5 when comparing Hardware. When and if Adobe supports Multi-GPU configs in CS5 then it will be alittle crazy but fun:)