Yes indeed. That's what I was suggesting in the event the specs aren't that competitive or promising. It would be wiser to remain silent for the time being to keep everyone guessing and not upsetting market forces against their favor prior to the official launch date.
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
New R600 info 320-stream processors
- Thread starter trek554
- Start date
matrix563
Fully [H]
- Joined
- Sep 1, 2006
- Messages
- 19,136
how can i get them to run on my x1950pro? i want to run seti on my gpu
chinesepiratefood
2[H]4U
- Joined
- Oct 8, 2005
- Messages
- 2,629
since unified gpus
Well, I haven't a suitable card for the GPU FAH client and therefore no personal experience, but DL the latest version of the software from Stanford and install it if you're interested. Apparently, the client will utilize one processor core in addition to the GPU on the card, so you will lose the processing of the CPU, but the net results will be greater WU production and increased PPD.
The latest ATI cards do very well with the FAH GPU client. The R600 is slated to improve these results to a great degree if there's any truth to the spec rumors. There's much anticipation in the folding community for the R600. We'll see.
Here's some forum links for additional info on GPU folding:
http://www.hardforum.com/showthread.php?t=1157364&highlight=gpu
http://www.hardforum.com/showthread.php?t=1136155&highlight=gpu
http://www.hardforum.com/showthread.php?t=1130264&highlight=gpu
http://www.hardforum.com/showthread.php?t=1126722&highlight=gpu
I don't think there's any clarity here.
Efficiency is not what you make it out to be. An efficient device is one that performs a task with a minimum of resources. Does having half (or one third, or two thirds) the number of SPs more efficient? NO.
An SP is a executional device or an arrangement of executional devices tied to some sort of scheduler or blocks of schedulers. A single SP may consist of a number of executional devices (one SP may consist of one, two, four, five, eight, or ninety-six arithmetic logic units -- or really any number of ALUs per SP). These ALUs have operations that they can perform -- adding/subtracting, dividing, multiplying -- and can execute these tasks in a specific manner. Some can do more than one of these ops per clock cycle, and have the capability of doing any of these ops, while others may be only to do one MUL per clock (for instance), and cannot perform other functions. We use terms such as vector and scalar to describe the capabilities of these ALUs.
All SPs are not created equally. G80 has 128 SPs consisting of scalar processors, and has a scheduling granularity that fine (it can schedule to ALUs, save for the "missing MUL", independently). R600 has "320 SPs", consisting of who-knows-what, and having a scheduling granularity of who-knows-how-fine.
"320 SPs" doesn't mean a fucking thing. Don't pretend like it does, and don't pretend like we know the SP efficiency yet. Only when we know the types of processors, the scheduling setup, clock speeds and the number of transistors will we be able to define R600's shading efficiency.
Effeciency in shaders you also have to take into consideration the clocks, the energy it uses at those clocks and real world results. Its not that simple to just say its got less units and so its more effecient. Actually the r600 can do more calculations thoretically then the g80 if shaders are made optimal for it, even when you consider the clock frequencies of the g80 shader domain, thats the nice thing about the g80 scalar architecture it really doesn't need to worry about the shader construction to get high utlization of it's units.
with minimum of resources? wtf are u talking about? the r600 has half the physical stream processors the g80 does and if it has the same performance or better, then there is really not much to discuss about the physical shader unit efficiency (i'm talking about performance, santa clause doesn't even know what the hell you are talking about) ..let me guess, power usage? typical nv f@nboy argument,
when 64 stream processors perform better than 128 stream processors given the same task, then yes , that makes the 64 SPs more efficient
you are damn right about that
what is your point?
r600 does not have 320 physical SPs, it has 64
go back and find out what the 320 number represents,
we dont' exactly know but have enough info to make a educated guess,
sure we can take the clocks into consideration, but it will not benefit the nv side since their shader clock is like 1300+mhz?
it's a good thing to compare the power usage but it really doesn't have much implication when evaluating the performance of top video cards,
it's like saying, "hey this drag car is 10 seconds faster than the other, but uses 10 more gallons of fuel,(nonsense talk)
majority doesn't care which car is going to burn more fuel, the majority cares which car is going to cross the finish line first
Obviously, you did not try to understand what I attempted to explain. You do not understand the concept of the SP and the concept of the ALU. Ergo, you will not understand my post. I tried to give a basic rundown, but somewhere along the line, I must have failed.
To put it very bluntly: one SP is not specifically one ALU. The transistor count of one SP in R600 is NOT the same as the transistor count in one SP in G80. One SP in R600 is NOT the same as one SP in G80.
As I said, efficiency is the degree to which resources are utilized to perform a given task. This means transistors and power consumption, NOT a number applied to the components that accomplish the task of shading. That specific number is irrelevant, and anyone familiar with the finer aspects of a modern GPU would understand that.
And, to address one of your points, I'm not a !!!!!! if the shading efficiency of G80 is higher than that of the shading efficiency of R600. That's a number one can apply, and we can't apply that number yet. We don't know the shading or scheduling efficiency of R600 yet. Again, please don't pretend as if we do.
If you could link me to where it has been stated by an AMD representative that R600 has 64 physical SPs, I'd love to see it.
Well you have to look at the clock of the shader units x number of shader execution of each unit to get the thoerectical shader performance, that can be compared to real world results, to get you effeciency with shaders per clock or over all shader performance.
Energy or power used per frame rate would be the other way to calculate effeciency.
Or transistor per watt used is another way as well.
Just saying can't really look at effeciency in this manner: if a 64 vec 5 shader units performs the same as 128 scalar shader units so the 64 shader unit piece is more effecient. Well these two ALU's are competely different. Vec 5 unit consists of 5 scalars so in reality we do have 320 scalar shader executions per clock vs 128 scalar instructions per clock. But this is theoretical. In real world it will not end up the full 320 instructions nor the 128 instructions. If the Vec 5 units are only 50% effecient then we are down to 160 intructions per clock. Then you have 90% efficiency for 128 which is 115 shader instructions per clock. Now factor in the clock differences 1.35 ghz vs 800 mhz.
115 shader instructions per clock x 1.35 ghz = 155
160 shader instructions per clock x 800 mhz = 128
now I'm pulling numbers out of my ass for the r600 shader usage in real world tests so don't put too much stock into it. I'm just trying to show you can't really talk about only the # units. Doing these comparisions you have to look at like units or items, otherwise the results have no meaning.
He's trying to explain to you that an "SP" isn't some universal concept that's equal in all architectures. You should listen.
If an R600 "SP" is a vec5 unit and a G80 "SP" is a scalar unit your simple comparisons are simply put...garbage.
0.5 TFlops and not just with MAD instructions... well thats easy with G80 and "future driver release" (hopefully coming soon)
G80@575MHz + 97.92 series drivers
-------------------------------------------------
gpudip 18: cel shading - 23 70 1 512 768 10000 - 682,3311 msec - 403,3983 GFlop/sec
G80@575MHz + "future driver release"
-------------------------------------------------
gpudip 18: cel shading - 23 70 1 512 768 10000 - 584,9589 msec - 470,5480 GFlop/sec
G80@702MHz + "future driver release"
-------------------------------------------------
gpudip 18: cel shading - 23 70 1 512 768 10000 - 496,1952 msec - 554,7236 GFlop/sec
I hope you gave Fudo these numbers to use as his own. Cause it looks like he copied them wholesale and claimed they were his.
http://www.fudzilla.com/index.php?option=com_content&task=view&id=554&Itemid=1
you and phide sound like you are trying to make excuses that somehow the sps can not be compared by the quantity of them just because they are not the same,
of course they are not the same and probably never will be unless nv decides to copy amd in a next gen or 2
,instead of breaking it down (which is what you 2 are insisting on) , sum them up by whatever they can do and compare them to each other as they are as whole. one sp to one sp
phide, i can see you often come out with tech mambo jumbo that not many outside of graphics industry understand or care to understand, and thats fine , but when you do it , why don't you lay it out so everyone can understand what you mean,
if you feel the need to discuss such things as you mentioned in you last post, go ahead but find a capable engineer to have a such discussion with.
look, i said that from the info thats available right now, the r600 is going to have 64 sps, and thats all i said, 64*5=320, what it really is we will find out in a few short weeks,
i never said anything about scheduling efficiency, i don't know where you got that one from,
my simple comparison was 64 sps against 128sps =same performance = one amd sp is more efficient then one nv sp ,
Yeah so if an amd sp takes up 4x the transistors and burns 6x the power of an nv sp is it still more efficient? You obviously don't understand this stuff (and doesn't seem that you want to either) so you really should just sit back and try to learn something. Your naive comparison of unit "counts" just doesn't make sense in any context no matter how you spin it.
i understand it just fine, and im willing to learn but i don't see you teaching anything,
i see a lot of ifs, sure if 1 amd sp takes up 4x transistors and uses 6x the power then i don't have a problem agreeing that its not more efficient, but there is no info on that so its out of the picture right now,
razor1 did an excellent job of summarizing some of the math needed to make a real comparison.
As I said, a stream processor, in GPU land, is a processor or arrangement of processors and/or schedulers/issuers that complete a simple task. They can complete this task because they feature one (or more) arithmetic logic units. An SP could technically be simple enough to be an ALU, or it can consist of multiple ALUs and scheduling-related devices. ALUs perform very basic math operations on a per pixel basis at a defined degree of accuracy.
If R600 uses 64 VEC5 processors, you have 320 ops per clock if scheduling efficiency is 100%. While you may not care about scheduling, it's critical for the operation of a unified GPU. The scheduler must ideally issue ops such that all ALUs are always chewing on data. With VEC5, this will rarely be possible. G80 has an advantage in that it can schedule ops with one op granularity. That is, it can issue any kind of shader (pixel, vertex, geometry) to any ALU independently. The split could be 84/13/31 or 126/2/0 or any arrangement. G80 makes up for a fairly low number of ops per clock by having efficient and scalable issuing and high clock speeds. R600 is a more brute force method: coarse granularity, less scheduling efficiency and lower clock speeds (and probably more transistors).
There's no way of simplifying this discussion. If you want to keep up, you'll simply have to read up. None of this is that complex.
And it's an absurd comparison. That simply isn't what efficiency is about.
If a V6 produces the same 300hp as a V12, while weighing the exact same (transistor count) and burning the exact same amount of fuel (power consumption) as the V12, is that V6 engine more efficient? No!
- Joined
- May 18, 1997
- Messages
- 58,668
Fuad's true sources come through. He simply reposts things found on public anonymous forums?????
You dont know much about cars do you
- Joined
- Dec 9, 2000
- Messages
- 16,336
|CR|Constantine
Limp Gawd
- Joined
- Jan 20, 2006
- Messages
- 218
Because there is no way on God's green Earth a V6 will weigh the same as a V12
All other aspects of the engines being equal, there is no way for a 6-cylinder to produce as much power as a 12-cylinder, and use the same amount of fuel, unless the displacement is the same; you would simply have a larger bore across 6 cylinders to produce the same displacement as the smaller bore in the 12 cylinders.
- Joined
- Dec 9, 2000
- Messages
- 16,336
nah there are alot of v12s that weigh less then current v6 engines, different materials are used to produce the engine,
its a horrible analogy because you won't ever find a v12 and a V6 engine that can produce the same powerband even if they did weigh the same and use the same amount of gas
even assuming the displacement is the same, the V6 will produce more Torque at a lower RPM, and the V12 will produce more HP at the higher end of the RPM range
the bore and stroke on the V6 would probably be different,
its a horrible analogy because you won't ever find a v12 and a V6 engine that can produce the same powerband
even if they did weigh the same and use the same amount of gasGuy above me said:
even assuming the displacement is the same, the V6 will produce more Torque at a lower RPM, and the V12 will produce more HP at the higher end of the RPM range
the bore and stroke on the V6 would probably be different,
- Joined
- Dec 9, 2000
- Messages
- 16,336