PlanetZ forums - scopeusers.com

I have been begging a pro DAW builder to post some results using the i7 and Kontakt.
They had plenty of Core 2 Duo Quads, Dual Core's etc. which had the usual Cubendo, Sonar, etc stuff.
I wanted a sampling streaming app as these are just as popular and all I ever hear is cackling about how many loads I get w/ this and that but never any proof, and when proof is provided there's no information on libraries used, etc.

I finally got a direct answer to a direct question and they provided some nice information.

_______________________________________________________________________________

Kontakt samples playback

This test is more geared toward Composers and heavy samples users.
There has been alot of discussion about what is best for samples playback and if more cores are better.
Samples playback is very memory intensive, memory speed and bandwidth are effected by several things.
GHZ of the processor, Front side bus of the processor, memory type (DDR2/DDR3) and of course platform (chipset and processor architecture). Little or no improvement is found with Quad core vs dual core.
A faster GHZ dual core will out do a slower quad core for samples playback. DDR3 1600 is better than DDR2 800.
Ideally when considering an overall picture of a project a fast GHz quad is best as the Quad cores do have a great effect on effects usage, overall project playback and high track counts @ high rez.
Some thoughts: the AMD is a very affordable and well performing answer for a samples(slave) box or an overclocked dual core.
The Core i7 is outstanding as a main system, or the more afforable overclocked Core 2.

The test: (not downloadable but duplicatable)
all done with XP Pro SP3, Fireface 400, 4gig ram) (core i7 was 6gig ram, vista 64)
the samples set was over 2gig)
6 Instances of Kontakt3 in rack All using K3 included Vienna instruments Default dfd settings All running live, no freeze, no bounce MIDI tracks running to every patch 1 Violin ens14 (all) Viola ens10 (all) Cello ens8 (all) Bass ens6 (all) 2 Piccolo (all) Flute (all) Oboe (all) FrenchOboe (all) EnglHorn (all) Clarinet (all) 3 TrumpEns3 (all) FrnHornEns4 (all) Trombens3 (all) Tuba (all) 4 Violin ens14 (all) Viola ens10 (all) Cello ens8 (all) Bass ens6 (all) 5 Tenor Recorder (all) Piccolo Recorder (all) Low Whistle (all) 6 Timpani Hits.

_______________________________________________________________________________
This pretty much answers all of the questions I have been asking builders and developers months ago to do.
Even though they used the expensive 965, the 945 and 920 show equally impressive loads.
This summer when the 32nm's are released the 945 which is close to 3GHz w/ 4 cores/ 8 threads will be a great choice for Romplers.
We could really use an X58 motherboard w/ 3 x PCI's or possibly cards w/ the PCI-e adapters I keep hearing about.

Yes, waiting until 32nm for i7/i5 would be a VERY good idea. Performance is great right now but things will change from this first generation with what motherboards will support.

PCIe adaptors would have to account for the fact that Scope cards require 5v (not 3.3v or 12v). Also if someone is purchasing a new i7 (or AMD3+/PhenomII) motherboard it would be a good idea to check the keying on the PCI slot voltages, some boards are out now that are 3.3v only for even the 32bit PCI slots.

I'll get a faster Quad CPU when the price comes down.

The 32nm stepping (Westmere?) will address some of the L2 concerns trading off L3 space (512k L2, 6MB L3) and add a 6core variant to the mix (as well as some extensions meant for crypto which don't affect music afaik).

Also I agree about the SLI licensing issues (Nforce200 bridge isn't necessary), but the ddr3 voltage tolerances aren't that big of a deal (not being able to do 'high voltage'):

Before i7 was out, Intel made an announcement that overclocking would have to reduce the voltages, since the qpi (quickpath interconnect) now connects the ram & cpu directly together (same voltage). There was a lot of press about it, and Intel stated that they didn't recommend overvolting beyond stock voltages to avoid damage to the cpu (something AMD phenom I & II users are familiar with and it's not as much of an issue as Intel was making it out ot be). And interestingly, Intel is allowing multiple SPD profiles for the RAM vendor with user settable performance settings. This runs counter to their statement about operating voltages being even more of a concern. Also I've seen overclocks hitting 2.2v without harming the i7.

But my understanding is that you don't want high voltages anyway, you want parts that perform well at lower voltages as they are less flawed. So this is the key point, DDR3's lower voltages are ok because DDR3 has a 30% reduction in operating voltage just due to the process change, and you don't want to have overvolted parts anyway (unless you are interested in overclocking/gaming?).

The way I have read it, is that when parts are binned by the maker they're binned based on the amount of voltage required to function properl, with more voltage actually meaning lesser parts because the parts were failing validation at the lower voltages. During the DDR1 & DDR2 era, OCZ and other 'performance' part vendors began to aggressively market these 'higher voltage' parts--that are being sold off as churn by the original manufacturer--as being 'performance ram' because they are tagged with a higher voltage from the get-go. This may also help with the timings, so the parts vendors boast faster CAS latency timings to boot. Note that I'm suggesting these 3rd tier parts are probably not ideal for someone who prefers stability to performance.

So DDR2's higher JEDEC specifies a 1.9v cieling for stability, 2.3 volts before failure (although parts aren't required to operate correctly before failure), while JEDEC DDR3 spec states 1.575 max voltage "recommended" plus DDR3 memory modules must withstand up to 1.975 volts before incurring permanent damage (although they are not required to function correctly at that level). Note the 30% drop in specified voltage, due to process change

And so with DDR3 there are TONS of parts selling for 1.65v to even 1.9v as 'stock' aka default profile, but 1.5v parts would be the best for someone that wants longevity & stability. And if you care about timings, just don't max your ram out now as they will improve as a process ages and the process size anyway (meaning you can always buy lower voltage parts in more volume later that have even better timings).

That's a result of the immaturity of the processes used for DDR3 and the i7 platform. The i7 platform issues are why I think it's best to wait 6-9 months to move beyond socket 775. I would never buy a rev1 or rev2 motherboard for a brand new cpu architecture myself. Not to mention that this particular architecture change has caused rather large changes in the way memory traces are done on the boards. The i7 & i5 platform should mature soon, though in this case Intel is skipping the next "tock" which would normally just be a process shrink with only minor core changes. The next process shrink is a "tick" with the move to westmere, a new core again. Intel's investing big time in 32nm...

As for the RAM makers, the DDR3 issues are currently being compounded by the insolvency of the manufacuring side of that sector of the market. Make no mistake, some of this market is imploding both from overproduction last year followed by the recession affecting tech purchases currently. If Qimonda and one other maker go under, expect things to get REALLY slow in terms of production and costs to rise in sharp contrast to the sinking that occured in last year's glut of parts.

At least that's imo...

Exactamente,

That's why Intel and other tier 1 guys are trying to keep Samsung from releasing those 4GB DIMM's.
You obviously saw the DAW builders graph w/ 775 recommending OC'ing and using DDR3 @ 1600.
Supermicro is looking good for this move later.
2 of the big DAW builders are using Gigabyte for the early adopters and I bet they are tweaking the P45's they are pushing also.

Speaking of monopoly...........

http://www.xbitlabs.com/news/other/disp ... onday.html