PlanetZ forums - scopeusers.com

Among the three..Xite1,XiteD and 14dspPCI,
what is the exact difference according to DSP power(number of synths or other modules simultaneously)?
Have users of Xite reported lower ully latencies?

ARCADIOS wrote:Among the three..Xite1,XiteD and 14dspPCI,
what is the exact difference according to DSP power(number of synths or other modules simultaneously)?
Have users of Xite reported lower ully latencies?

In term of pure DSP power, the Xite-1 was innitially announced to provide 10x the 14 DSP board power.
In term of pure DSP power the Xite-1D provide slightly more than 1/3rd of what the Xite-1 provide.

Due to some optimization requirement that would be necessary (especially for 3rd party devices), this figures are still partially theoritical. The 14 DSP board used more slower DSP. The Xite series use a mix 6 old and "slow" DSP (mainly dedicated to IO routing and host PC interaction) and a set of 12 new high-speed DSP for the Xite-1 (4 for the Xite-1D).

The last estimate I remember was that you should expect the Xite-1D to provide you 2.5x what you could expect from a 14 DSP board, in other words, the Xite-1D could be seen as the equivalent power of a PCI system using to 14/15 DSP Scope Pro and 1 Pulsar II/Project 6 DSP. There had been some hints that the difference between a 45 (3x15) DSP system setup and a 36 (2x15 + 6) DSP system setup was not that obvious (probably due to the use of DSP ressource used for data exchange between the DSP).

So the figure for the Xite-1 should be in the range of 6x to 8x what you could achieve with a 14 DSP board, so roughly the fusion of 3 of the previously mentionned PCI setup.

My 2 cents

HUROLURA wrote: in other words, the Xite-1D could be seen as the equivalent power of a PCI system using to 14/15 DSP Scope Pro and 1 Pulsar II/Project 6 DSP.

2,5x is equivalent to 2 pro + 1 pulsar.... if i am not wrong? ... or you meant "1,5x" of a scope PCI system?

+also, if we make the multiply of 330mhz x 12 + 6x60mhz = 3960+360 = 4320(xite-1)
and 14x60 = 840(pci14dsp) .......... this is about 5 times the power of a PCI card and not 10. am i correct??

on the second part, no you're not correct.

for sheer horsepower, the XITE-1 can do the same work as 10 14dsp cards. in real life, it's more like 6 to 8 of the 14dsp cards, because of how complicated loading that many devices on that many dsps becomes. i don't think clock speed works quite like that.

on the first part, you're correct. the XITE-D does the job of 35 of the old dsps or about two 14dsp cards and a 6 dsp card. if you do the same math that you did for the XITE-1, you get (6x60= 360) + (4x330= 1320) = 1680 which is considerably lower than 35x60 = 2100

think of it this way, the XITE-1 has the same number of 6omhz dsp chips, but 4 times the number of 330mhz chips. if the XITE-D is 2 1/2 14dsp cards, then the XITE-1 is almost 4 times that power or 10 14dsp cards.

what is the point?

i just want to know how far i am(14+6dsps) from the xite-D
and also if there is ully latency improvement over the PCI cards

DSP rating alone wont give you a fully accurate power comparison. On the XITE-1D my experience is that other resourses will max out before DSP is anywhere near full. And it depends on what types of devices and how you connect them. For example, some async devices may max out the bus to the host, some devices loaded inside mixers instead of outside may max out sat connections.

Best for XITE1-D is to use a mix of different devices, and you have to find the right mix, to get the maximum out of the available resources (DSP, inter DSP connects, DSP to host, onboard RAM, etc).

Having said that, I can do way more on the XITE1-D at 48Khz than I could on 12DSP PCI at 44.1Khz.

On 12DSP PCI I could run a mix and mastering plugs at 44.1 Khz. On XITE-1D I can run the same mix/master at 48Khz, and throw in a couple of instances of Pro Wave, a side-chain compressor and a couple more DAS EQ's.

latency really can't go lower without an unreasonable increase in the use of CPU resources. latency is all about giving the cpu time to complete all of it's tasks. the D is quite a better deal than the old pci cards were, dollar vs processing power, but the real plus is that the XITEs also sound quite a bit better than the old cards.

Let's provide some numbers ...

As Garyb explained, the rough calculation MHz need some caution.

On one side the core of the new DSP has been improved compared to the old one, so the performance of the new one should not only be seen from the MHz ratio. Checking Analog Devices figures in MFlops for example tend to show it.
The old ones (ADSP21065L) provide 198 MFlops for 60 MHz.
The new one (ADSP21369) provide 2 400 MFlops for 400 MHz (so 1 980 MFlops at 330 MHz).
The MHz ratio 330/60 = 5,5
The MFlops ratio (with same cMHz) 1 980/198 = 10 !!!
If you use this last figure, one Scope Pro with (14)/15 old DSP provide 15 x 198 = 2 970 MFlops
So your setup with 1 Scope Pro + 1 Pulsar II setup brings 4 158 MFlops
One Xite-1D then provides 6 x 198 + 4 x 1 980 = 9 108 MFlops
One Xite-1 provides 6 x 198 + 12 x 1 980 = 24 948 MFlops
With such figures 1 Xite-1D = 3 x Scope Pro = 2.2 x your setup
On the same level, 1 Xite-1 = 2.7 x Xite-1D = 8.3 x Scope Pro = 6 x your setup

Besides making calculation inside the same chip should be more efficient as it can spare on cycles needed to allow inter DSP chip data exchange on old cards. But as soon as you need to exchange data between DSP (to bring the data to the IO for example which muight sometimes be necessary, or to access to host RAM for delay and reverbs), then a bottleneck could also occure.

On the over side, some optimization might still be needed to achieve such result.
So maybe it would be reasonnable to lower the theorical ratio from 20% which brings think back to the 1 Xite-1D = 2.5 x 1 Scope Pro.

Then, 1 Xite-1D would be equivalent to 1.8 x your setup.

Then the I/O available should also be taken into account.
If your setup includes 1 SRB board (with no I/O), the Xite-1D can bring you what you had before with nearly twice the DSP power.

These figures are still theory as there is probably several parameter which could rise or lower the capabilities of the system depending on what you try to do, the plugs choosen, the way they are mapped to the DSP.

But to have a simple rough idea, 1 Xite-1D should provide the power of the most powerfull Scope PCI system.
And the Xite-1 should provide 2 to 3 times more power.

PS: that said, the Xite-1 still is the most economical solution in term of price per DSP power, as soon as you can afford it.

ARCADIOS wrote:i just want to know how far i am(14+6dsps) from the xite-D
and also if there is ully latency improvement over the PCI cards

There is also two speeds for PCIe to run at, for PCIe1.0 and PCIe2.0.

PCI = 133 - 266 MByte/s (shared for all cards on bus, including non-SCOPE cards)
PCIe1.0 = 250MByte/s
PCIe2.0 = 500MByte/s

From my experience, the XITE PCIe and ExpressCard both support up to PCIe 2.0 speeds. PCIe is also a "point to point" communication system, so there is no bus contention, nor sharing of bandwidth between devices. So for most systems, PCIe is faster than PCI is capable of. If you get a motherboard or laptop that supports PCIe 2.0, that would be even better. For ULLI this means S|C has provided many metric shit-tons of bandwidth, so if you use many channels of ASIO, or devices that need system RAM, or XTC mode, you are less likely to have to increase the ULLI latency to accomodate that much data.

My laptop is based on Sandy Bridge (core i7), it supports PCIe 2.0 ExpressCard, and I leave ULLI at the lowest setting. I use Ableton Live, and with 16 stereo channels of ASIO in each direction, and with the simulated CPU utilization turned up to 80%, I get no stutters/glitches/clicks. All is well.

Metric Shit Tons = the new industry standard

Greg

HUROLURA wrote:The old ones (ADSP21065L) provide 198 MFlops for 60 MHz.
The new one (ADSP21369) provide 2 400 MFlops for 400 MHz (so 1 980 MFlops at 330 MHz).
The MHz ratio 330/60 = 5,5
The MFlops ratio (with same cMHz) 1 980/198 = 10 !!!

The ADSP21369 can only achieve 10 times the performance on specific workloads with specific instructions (SIMD). The most realistic performace figues would be between 5x (frequency difference) and 8x (frequency difference and heavy optimization of the code, which might never happens).

SC have reduced the intercommunication speed of the new ADSP21369 to be compatible with the old ADSP21065L to simplify the circuit of XITE which means that the communication between the new DSPs run way slower than specified. The Solaris synth can do things that the XITE can't in that aspect. I can run 3-4 Plate reverbs on my Scope board and I can run 7-8 Plate reverbs on XITE-1 before the internal communication is saturated.

Theres a PDF here explaining the internals - not that I understand it much but it does look like the SIMD Core (Single Instruction Multiple Data) is a jump over the first gen SISD (Single Instruction Single Data).

http://www.analog.com/static/imported-f ... _21369.pdf

see what you started?

garyb wrote:see what you started?

garyb wrote:XITEs also sound quite a bit better than the old cards.

why? because of better materials at some stages? or even because it stands away from the pc box??


can xites be bridged??
lets say 2 xites together?(which means also 2 times the ios as well)

XBoxes are noticably better than the cards. I use to notice the RME's playing my same GVI sampled instruments had better clarity than my cards.
Then I noticed the ASB's and Klangboxes sounded better too, possibly from hotter outputs.
But more recently the Kontakt instruments I used coming from my XBox totally overwhelm the Muse Receptors, which were also slightly better than my cards.
Most guys I know really want an XITE-1 but whine like children when they hear the price.
I often reflect back to 1992 when I had Roland S760 smaplers, an Oberheim Matrix 12/ XPander combo. Just that gear was 15,000 USD, and my XITE-1 easily out performs that combination now. But I have since added the hardware Solaris synth, and can say the Oberheims and Rolands could never do what these 2 units do.

It's also why I hate listening to children whine about price and bleed all over the developers for bringing them a quality product.
A Muse Receptor w/ NI Komplete will run you 27-2800 USD.
The XBox w/ Kontakt 5 runs around 34-3500 USD.
S|C blows away the Receptors, but OMG the extra 600 bucks..................
I am actually glad they choose to save a few bucks, as then I can enjoying beating them unmercifully in public as they try and get their VST Kazoos to sound as good as the DSP synths.

I encourage them to continue buying the Spuse Rejectors.....
I'd hate to think that others I work oppostie of would have such quality realtime sounds. I'm use to punsihing them, and hope to continue doing so.

the better sound is a combination of things including better components and the dsps and i/os being out of the computer.

no, there's no way to combine units or add i/os at the moment. there is a bus that allows more i/os and all kinds of other cool ideas. if enough units sell to keep S|C in business and with a little bit extra money to pay developers, ALL the cool things that can be imagined will happen. the guys at S|C are not the kind that sit around on their butts when there's work to be done.

xite-1 is worth it analogically to power and money!
anyway..... i feel fine at the moment with my old 20dsps.....
"at the moment"