Lucid Hydra 200: Vendor Agnostic Multi-GPU, Available in 30 Days

[munim]

?

5% per year, is gain of 5%...($100>$105>$110) <-- thats not 100% scaling, that just linear progression.

Yeah...that's what his point was. Some people are still thinking that linear scaling means this:

1 card: 1000 3dmarks
2 card: 2000 3dmarks
3 card: 3000 3dmarks

When in fact it can mean this:
1 card: 1000 3dmarks
2 card: 1500 3dmarks
3 card: 2000 3dmarks.

Or anything else for that matter, as long as adding another card adds exactly the same amount every time.

[Kot_original]

When in fact it can mean this:
1 card: 1000 3dmarks
2 card: 1500 3dmarks
3 card: 2000 3dmarks.

Or anything else for that matter, as long as adding another card adds exactly the same amount every time.

No.
0 card: 0 3dmarks
1 card: 1000 3dmarks
2 card: 1500 3dmarks
3 card: 2000 3dmarks

Adding the first card adds 1000 points adding the second one only adds 500 points, so it's not linear.

[Salavat23]

?

5% per year, is gain of 5%...($100>$105>$110) <-- thats not 100% scaling, that just linear progression.

That's my point, lol. You guys are the ones arguing that a 5% gain each year is not a linear relationship.

Don't roll your eyes at me. Roll them on the guys who don't know what a linear relation is and are trying to pass their bs as fact (m0da).

123

Wtf?

If you do not have the educational background to comprehend these formulas and properties, please do not insult them, calling them bs. I can scan these exact formulas from a textbook, but then again, you wouldn't understand what you are seeing, hence there is no point.

I don't have the educational background to comprehend those formulas? Since when do you need an educations background to comprehend formulas that have nothing to do with your point of 100% scaling?

Linear is not always 100% scaling. Linear is constant scaling.

Remember first, second, third differences way back from High School? In fact. Search it up if you can.

Contradictory to your point of view, linear scaling does in fact mean that scaling your number of cards, scales your output by the same amount.
See the scaling property of linearity, referenced in my earlier post.

Do you know what linear is? In all of your examples, you are assuming a slope of 1. Slopes can be other numbers too. If your slope, or m-value is 1, then you get 100% scaling. In all other linear cases, that is not the case. And that's exactly why your property of linearity is bs. And since when is it "YOUR" property, albeit it being a property or theory that doesn't make sense.

Salavat, if you would like to learn, with proof, more of what you may think you know, feel free to send me a PM.
Let's not bicker any further.
Moreover, I'd appreciate if you would refrain from further insults to the property of linearity.

First of all, linearity is not a property by mathematical terms. And second of all, Linearity is not always 100% increments.

Here are some hints that Linear does not always mean a slope of 1:

y=1/2x + 5 ------------ Whats that? Not 100% increases in the variables, yet its a linear equation?

y=0.0888x+3 ------------ Whats that? Not 100% increases in the variables, yet its a linear equation?

Get it?

[Manicdan]

No.
0 card: 0 3dmarks
1 card: 1000 3dmarks
2 card: 1500 3dmarks
3 card: 2000 3dmarks

Adding the first card adds 1000 points adding the second one only adds 500 points, so it's not linear.

i dont think theres any written rule for reviewers to compare xfire/sli scaling to reflect changes based on a 0 starting point

[Salavat23]

No.
0 card: 0 3dmarks
1 card: 1000 3dmarks
2 card: 1500 3dmarks
3 card: 2000 3dmarks

Adding the first card adds 1000 points adding the second one only adds 500 points, so it's not linear.

Linear lines don't always have to start at (0,0). It can start at (0,500), like in that case. However, it still makes no sense. You would just have to put a restriction on the line.

[ABXG]

No.
0 card: 0 3dmarks
1 card: 1000 3dmarks
2 card: 1500 3dmarks
3 card: 2000 3dmarks

Adding the first card adds 1000 points adding the second one only adds 500 points, so it's not linear.

This is correct, Hydra could only offer linear scaling if the slope from 1 to 2 cards was equal to the slope of going from 0 to 1 cards.

The form of a linear equation is: y = ax + b, and b would be nearly 0 (not exactly because you could technically run a graphics test on your CPU, but I'll say 0 for simplicities sakes).

For linear performance you would end up with something like:

[Assuming each card added is identical]

cards : 3dmarks
0 : 0
1 : 500
2 : 1000
3 : 1500
4 : 2000

In this case you would have a slope of 500 (slope does not need to be 1 in a linear equation, it just needs to be constant).

Near linear would look something like this:
0 : 0
1 : 500
2 : 999
3: 1498
4 : 1997



Hydra will not offer linear scaling unless it is offering 100% scaling.


*Note - In the following:
0 : 0
1 : 1000
2 : 1500
3 : 2000
4 : 2500

It is a composition of two linear equations:

y = { 0 <= x <= 1; 1000x } { 1 < x < inf; 500x + 500 }

It is linear from 0->1 and linear from 1->inf, but it is not linear from 0->inf.


EDIT screwed up the last equations signs.

[trinibwoy]

If you chart a linear relationship between two variables and then connect the data values, you will have a straight line. This does not necessarily mean that as one variable increases by one unit, that the other variable also increases by one unit. It simply means that as one variable is increased by some amount, that the other variable also increases by some amount, every time. Don't even begin to argue this. Google = friend.

I think this has been addressed already but if you need google to understand simple math then you're out of your depth here. Linearity doesn't restrict you to increases of "one unit" - it implies a constant ratio. i.e performance delta from 1->2 gpus = perf delta from 0->1 gpu. Since the starting ratio is 1 (or 100) then that must be maintained in order to claim linearity.

Linear lines don't always have to start at (0,0). It can start at (0,500), like in that case. However, it still makes no sense. You would just have to put a restriction on the line.

So you're going to completely ignore the contribution of the first GPU and delete that data point? So if GPU 2, 3 and 4 each only add 5% of GPU 1's performance you're going to claim that is linear scaling?

[Salavat23]

So you're going to completely ignore the contribution of the first GPU and delete that data point? So if GPU 2, 3 and 4 each only add 5% of GPU 1's performance you're going to claim that is linear scaling?

No. I would delete the point of 0 GPUs, 0 Score, because that is not a realistic scenario. You can't have any score at all if you don't have a GPU. 0 is still a score.

So according to you we should keep negative values as well? I don't think so. Restrictions on variables like these should always be put into effect.

[Aberration]

Just draw it out, you will see that a constant increase by % is not linear.

[Salavat23]

Just draw it out, you will see that a constant increase by % is not linear.

Compounding percentage, no. But percentage increases based on the initial amount, yes.

Compounding percentages are based on exponential relations, not linear relations.

[trinibwoy]

No. I would delete the point of 0 GPUs, 0 Score, because that is not a realistic scenario. You can't have any score at all if you don't have a GPU. 0 is still a score.

Ah I see, so in your world you'll only have data for 1 and 2 GPUs. So how do you define linear scaling in that scenario?

[Salavat23]

Ah I see, so in your world you'll only have data for 1 and 2 GPUs. So how do you define linear scaling in that scenario?

In an ideal scenario, you want 1+ GPUs. I want to include data that makes sense. 1, 2, 3, 4 GPU set-ups.

You are the one who wants to include 0 GPU data. Which makes no sense.

[m0da]

Since you don't want to do this over PM, I think we can do this here, without much damage.

First of all, refrain from name-calling.
Second of all, I will address your points with concrete data, not the ones where you attack words.

Let's begin:

Do you know what linear is? In all of your examples, you are assuming a slope of 1. Slopes can be other numbers too. If your slope, or m-value is 1, then you get 100% scaling. In all other linear cases, that is not the case. And that's exactly why your property of linearity is bs. And since when is it "YOUR" property, albeit it being a property or theory that doesn't make sense.

Your slope is determined by deltaY/deltaX. The properties of linearity do not put a limit on slope, hence you are assuming some incorrect information.

Remember (simply put):
-Property 1: if y = x, then Ay = Ax. This is called scaling. If you multiply your input by A, your Y must be amplified A times. So, if for an input of 1 (x = 1), your output is 10 (y = 10), then with an input of 3 (x = 3), your output should be 30 (y = 30).
*y = x -> x can be the # of video cards, and y can be your 3DMark score. This means with 0 cards you have 0 points. With 1 card, you theoretically have a score of 20000, and with 2 cards you have a score of 40000. Your slope here is 20000. Right? Don't bother replying to this, I know it is correct.
-Property 2: if y1 = x1 and y2 = x2, then y1 + y2 = x1 + x2. Simply put, if you add input x1 and x2, then your output is y1 + y2, the sum of their outputs.


Second:

First of all, linearity is not a property by mathematical terms. And second of all, Linearity is not always 100% increments.

Here are some hints that Linear does not always mean a slope of 1:
y=1/2x + 5 ------------ Whats that? Not 100% increases in the variables, yet its a linear equation?
y=0.0888x+3 ------------ Whats that? Not 100% increases in the variables, yet its a linear equation?
Get it?

Linearity is not a property by mathematical terms.
Is this a serious statement, or are you just playing with me? That's not funny.

Your two equations, they show a linear progression (read: increase linearly), but are not linear systems. Those y-value offsets (+5 and +3, respectively) make it non-linear. Try going over the first property of linearity I typed up for you with one of those equations.

*Hint: Wikipedia tells us the only case in which an equation of your forms satisfies linearity. It's at the end of the Linear Polynomials (the equations you typed are deemed "linear polynomials") section. It states "linear polynomials over the real numbers do not in general satisfy either additivity or homogeneity. In fact, they do so if and only if b = 0." Additivity and homogeneity are the two properties of linearity. Additivity is the 2nd property I listed, while Homogeneity is the 1st property I listed.

[p8ntslinger676]

regardless of if it does or doesnt provide linear scaling, im excited to see hydra finally come to fruition, when i first heard of it i never thought it was actually going to make it to market.

[Decami]

I have been following it since the start as well, here are a couple more sources that mention above linear:

Lucid
Cnet
X-bit
guru3d

Found those with a quick google search.

This whole "meaning of linear" debate is very entertaining!

debate will be ended by this. Your totally right with those links, me and everyone are confused.

The percentage comment and linear comment from Lucid are 2 totally different comments meaning separate thing.

Lucid said.

"above linear performance"

and "Near 100% scaling"

we are thinking the "scaling" and "linear" were in the same sentence, and they were not. LOL complete argument for nothing.
You can blame me for the whole thing if you please.

Linear is a constant path of increase. near 100% scaling means each object added will be worth its complete weight alone "almost".

So Lucid is saying its a linear gain with each GFX added of near 100% performance gains. Which is, by the way, freaking amazing if true.

If lucid isnt spitting out lies and bullcrap. Obviously the poster who said worse scaling that SLI/Xfire will look like an imbecile. (no offense of coarse)

[m0da]

regardless of if it does or doesnt provide linear scaling, im excited to see hydra finally come to fruition, when i first heard of it i never thought it was actually going to make it to market.

Let's get back to this.
These are the same thoughts I have about Hydra. Can't wait to see this supposed-bad-boy in action.

[ryan92084]

m0da i think the diconnect is happening when you assume the second and third card added are the same as the first (making it a simple y=mx) they are trying to say that this doesn't have to be the case (as you can mix different cards)

moral of the story is they never should have said linear when you can mix cards.

need more hydra benches

[railer]

Frkn ubber geeks took over this board.
Run for the hills!!!

[Hu1kamania]

Linear is a constant path of increase. near 100% scaling means each object added will be worth its complete weight alone "almost".

So Lucid is saying its a linear gain with each GFX added of near 100% performance gains. Which is, by the way, freaking amazing if true.

If lucid isnt spitting out lies and bullcrap.

Yup, that is the million dollar question. It seems too good to be true, but it sure is nice to think about and fun to talk about. Let's hope that at some point, sooner rather than later, Lucid can really pull this off.

[kadozer]

Okaaaaay...glad that we got junior high algebra out of the way. Who's still looking forward to/hopeful Fuzion comes through?

[Chaserjzx100]

goodness what was that crazy algebra all about... anyways hope this really scales good.

[Xoulz]

Just draw it out, you will see that a constant increase by % is not linear.

Exactly... it's lost in translation.

Lucidlogix claims near perfect linear scaling. That means as you add cards, they are near 100% of their potential. Thus adding one to your system increasing it's performance by 2. Or, a 100% gain over a single card. Don't confuse that with.. both cards are now only doing 50% of the workload. The performance gets calculated by 1 magnitude greater, for each additional card. 2x~4x.


It is the same for every additional card you add. Though each just adds to the collective performance. But can be translated as % or fraction of additional value (First order 1/2, second order 1/3, third order 1/4 total gain).

You do start @ 100%, not zero. 100% is the base and you chart off that!

[trinibwoy]

In an ideal scenario, you want 1+ GPUs. I want to include data that makes sense. 1, 2, 3, 4 GPU set-ups.

You are the one who wants to include 0 GPU data. Which makes no sense.

You're avoiding my question. You can't just ignore 2 GPU setups because they're inconvenient to your opinion. So I'll ask again, how would you determine whether you are getting linear scaling in a 2 GPU setup?

[Manicdan]

You're avoiding my question. You can't just ignore 2 GPU setups because they're inconvenient to your opinion. So I'll ask again, how would you determine whether you are getting linear scaling in a 2 GPU setup?

you kinda cant. all you can do is know that 1 gpu is 100%, and 2 GPU would have to be 200% to get linear scaling due to the assumption that 0 GPU is 0%.

but in reality, if a reviewer said you get 100fps with one card, 150fps with 2, 200fps with 3 and 250fps with 4. the reviewer could establish the second and third and fourth cards all offered the same increase in fps. even though the change is only 50% of the single card. it is linear, and its very important that its linear.

trying to just read data and determine the curve of the chart is not nearly as important as why the data resulted as it did. it the example i mentioned, it could be determined the extra cards were not working to full potential, due to possibly having a mobo thats 16xpcie for the first lane, then 4+4+4 for the rest. it would also show the game/benchmark has the poential to scale just as well with any number of cards, while normally we can see even smaller gains and more cards are thrown in. but with this example the gains were LINEAR.

stop arguing over how to read a chart, and think for a second how to read into the causes of the data.

[trinibwoy]

The causes of why GPUs don't scale perfectly have nothing to do with the rudimentary definition of linearity. You guys are trying to complicate something that is very simple. Nobody in their right mind would consider 150% performance after adding a second GPU as linear scaling. Shifting the baseline just to accomodate neo-math doesn't work either.