This would be good for AMD if this is true.
One hundred years from now It won't matter
What kind of car I drove What kind of house I lived in
How much money I had in the bank Nor what my cloths looked like.... But The world may be a little better Because, I was important In the life of a child.
-- from "Within My Power" by Forest Witcraft
Jack, I think you should take your own advice. I'm taking it now...
Motherboard: ASUS P5Q
CPU: Intel Core 2 Quad Q9450 @ 3.20GHz (1.07v vCore! )
RAM: 2GB Kingston HyperX 800MHz
GPU: MSI Radeon HD 4870 @ 780/1000 (default)
Explain ... sure.
You see within a few FPS two CPUs performing perfectly the same, you drive straight to the bandwidth and nothing in the data suggests that.... what it suggests is that the GPU is the limiter and within the noise of the measurement running high res max AA and AAFS the GPU will yield X FPS... that is all it says.
So, let's think about it for a second ... how would you prove it is actually a BW problem? Experimentally.
One hundred years from now It won't matter
What kind of car I drove What kind of house I lived in
How much money I had in the bank Nor what my cloths looked like.... But The world may be a little better Because, I was important In the life of a child.
-- from "Within My Power" by Forest Witcraft
One hundred years from now It won't matter
What kind of car I drove What kind of house I lived in
How much money I had in the bank Nor what my cloths looked like.... But The world may be a little better Because, I was important In the life of a child.
-- from "Within My Power" by Forest Witcraft
This has been proven before as well, and i think Jack's results are perfectly legit
high resolutions + AA/AF = phenom beating core2 quad at simplar speeds, even if its by 1-2 fps. if the c2q is overclocked things might be different tho
Jack can you run tests with the Q9450 overclocked to see at what speeds its more or less equal to the phenom?
i7 920@4.34 | Rampage II GENE | 6GB OCZ Reaper 1866 | 8800GT (zzz) | Corsair AX750 | Xonar Essence ST w/ 3x LME49720 | HiFiMAN EF2 Amplifier | Shure SRH840 | EK Supreme HF | Thermochill PA 120.3 | MCP355 | XSPC Reservoir | 3/8" ID Tubing
Phenom 9950BE @ 3400/2000 (CPU/NB) | Gigabyte MA790GP-DS4H | HD4850 | 4GB Corsair DHX @850 | Corsair TX650W | T.R.U.E Push-Pull
E2160 @3.06 | ASUS P5K-Pro | BFG 8800GT | 4GB G.Skill @ 1040 | 600W Tt PP
A64 3000+ @2.87 | DFI-NF4 | 7800 GTX | Patriot 1GB DDR @610 | 550W FSP
Surely can, and will. I have done the low res/low quality regime already, I will scale that up to high res regime (have monitor capable of as high as 1900x1200).
In the GPU limited regime, the switch over to favoring Phenom occurs around 1024x768 in my Lost Planet tests, but the CPU limited regime remains fixed.
Jack
EDIT: Here is the article I wrote with respect to Lost Planet, http://www.xcpus.com/GetDoc.aspx?doc=12&page=1 ... here is the thread opened up to discuss it, http://www.xtremesystems.org/forums/...d.php?t=189310
Last edited by JumpingJack; 08-09-2008 at 12:30 AM.
One hundred years from now It won't matter
What kind of car I drove What kind of house I lived in
How much money I had in the bank Nor what my cloths looked like.... But The world may be a little better Because, I was important In the life of a child.
-- from "Within My Power" by Forest Witcraft
JumpingJack: Do you know the latency when Intel needs to send data through the FSB?
If the game has two threads, one is writing data and one is sending data to the GPU. Some of the requests are done at the same time. Will the latency be the same as when the FSB isn’t "taken" by another request?
Compare this with the situation on AMD. Hypertransport for I/O (GPU traffic) and IMC for memory.
Do you know the latency for AMD when it needs to write or read data to external hardware?
You also know that if one game has two rendering threads, then this game needs to synchronize commands to the GPU. On C2D this would work well but on C2Q, what happens if one thread is located on one C2D and the other is located on the other C2D? (You know that C2Q is two C2D that communicates through the FSB)
Phenom has L3 cache that can synchronize four cores.
Yes I have measured it.
Irrelevant, the bottleneck in a high res/high setting situation resides on the GPU/Card, not the CPU.If the game has two threads, one is writing data and one is sending data to the GPU. Some of the requests are done at the same time. Will the latency be the same as when the FSB isn’t "taken" by another request?
So here is the ticket explain an experiment we can run that will prove that it is FSB driven in the CPU under your conditions.
Again, irrelevant, the workhorse is the GPU ... it is not this issue. What you are doing is looking at a case where at 2 million pixels being rendered by the GPU results in two different platforms giving the same FPS within a few FPS. You then conclude based on some marketing PR from AMD that their IMC is the ticket.Compare this with the situation on AMD. Hypertransport for I/O (GPU traffic) and IMC for memory.
Do you know the latency for AMD when it needs to write or read data to external hardware?
You are also confused on how Intel's bus architecture works with respect to handling IO from the periphrials....
The CPU will be waiting on the GPU, the GPU is processing pixels to finish the frame. The fact is at gaming code, multithreaded, C2Q is much faster at accomplishing the task.You also know that if one game has two rendering threads, then this game needs to synchronize commands to the GPU. On C2D this would work well but on C2Q, what happens if one thread is located on one C2D and the other is located on the other C2D? (You know that C2Q is two C2D that communicates through the FSB)
Phenom has L3 cache that can synchronize four cores.
Just answer the question ... how would you prove that it is a BW or latency argument in the FSB?
Last edited by JumpingJack; 08-09-2008 at 03:14 AM.
One hundred years from now It won't matter
What kind of car I drove What kind of house I lived in
How much money I had in the bank Nor what my cloths looked like.... But The world may be a little better Because, I was important In the life of a child.
-- from "Within My Power" by Forest Witcraft
Ok, I will answer you question I need to dig up the file.... average read latency is 65 nsec, with a min block size of 1K and a max bock size of 4K. It depends on the board as well, not all board produce the same latency due to physical difference in the traces.
However, let's take this slowly .... this is the third time I have asked you to answer a simple question.
(I will edit the answer to your question in a moment)
So I edited answer in, could you now answer mine.
What experiment should we run to prove your hypothesis?
Better yet.... let's start even more simple.
If I compared a Phenom 9850 @ 2.5 Ghz running a game at 640x480 against a QX9650 @ 2.5 GHz (running the same clock speed), who would win?
Jack
Last edited by JumpingJack; 08-09-2008 at 04:16 AM.
One hundred years from now It won't matter
What kind of car I drove What kind of house I lived in
How much money I had in the bank Nor what my cloths looked like.... But The world may be a little better Because, I was important In the life of a child.
-- from "Within My Power" by Forest Witcraft
Pick any test that is sending a lot of data through the FSB. When programmers optimize code this is in fact something that they are working hard with. Using the cache as much as possible and avoiding going to memory.
Your mama!
It depends on the application (the code). You can analyze things until you die... But if you check the code then you have it!
Single threaded games then QX9650 will win big, the big L2 cache is a huge advantage on singel threaded games at low res.
If the game is using two threads then C2D will win, C2Q is going to be a bit slower and last is Phenom. If the game is using three threads and they are doing some synchronization. Then Phenom will probably win, it depends on how the threads are using memory and talks to each other. C2D is going to runt two threads on one core and that will slow that processor. C2Q isn’t good at synchronizing and if two threads are writing to the same data then it also is going to decrease its speed. More threads and Phenom will increase its advantage.
If the game is using MUCH memory (almost as databases) then Phenom will probably win if the game is using two threads.
The Core 2 architecture is very good for single threaded applications. This isn’t any arguing about. It is built to perform well for that.
The assumptions above is of course dependent on how the implementation has been done, if you create code that isn’t using what is fast on the processor, like reading one byte each read when BIG blocks are read etc then the result could be anything.
Here you can read article that explains some techniques creating threaded games so even a non programmer will understand.
http://www.gamedev.net/columns/event...le.asp?id=1345
Programming is very complex, understanding what takes time, how to optimize and how to maximize the processor and hardware is VERY complex.
But if you are blind to information or having some trouble with feelings on how it works. It will not be possible to understand.
In order to understand why some applications behave in certain ways, you need to have the code. Fix the code first and then it will be possible to explain.
Don't know how "supreme commander" performs.
In the article supreme commander used "render split", this works well on Intel.
Last edited by gosh; 08-09-2008 at 06:49 AM.
as enduser i dont need to understand how something is coded, nor i really care, all that counts is how it performs in the end.
I could care less if a app is codec perfectly but still runs slower then a other app that is counted as "hackjob".
If the majority off apps run faster on cpu a then on cpu b, i buy cpu a. I dont care if cpu b is faster under very special circumstances that only are achived artificially...
If cpu b fits your need, fine, buy it noone stops you. But trying to prove at all costs that cpu b is superior over cpu a, for something that isn't relevant for 99% of the endusers, will only eran you some
@ supreme commander, a intel quad is the first choice, for grid i dont know cause there are afaik no benches out there that compare how grid performes if it is cpu limited on the different cpus.
Look, I will make this simple for you ....
If your hypothesis is correct, then if I change the FSB speed and only the FSB speed, then the frame rate should change accordingly.
So if I run a game at high resolution with at a clock of 333 Mhz (1333 MHz), 2.5 GHz CPU clock then run exactly the same benchmark but at 200 MHz (800 MHz) and 2.5 GHz clock speed, will the frame rates go up or down?
You do not need to see the code in order to make an empirical observation.
Last edited by JumpingJack; 08-09-2008 at 11:47 AM.
One hundred years from now It won't matter
What kind of car I drove What kind of house I lived in
How much money I had in the bank Nor what my cloths looked like.... But The world may be a little better Because, I was important In the life of a child.
-- from "Within My Power" by Forest Witcraft
Motherboard: ASUS P5Q
CPU: Intel Core 2 Quad Q9450 @ 3.20GHz (1.07v vCore! )
RAM: 2GB Kingston HyperX 800MHz
GPU: MSI Radeon HD 4870 @ 780/1000 (default)
One hundred years from now It won't matter
What kind of car I drove What kind of house I lived in
How much money I had in the bank Nor what my cloths looked like.... But The world may be a little better Because, I was important In the life of a child.
-- from "Within My Power" by Forest Witcraft
Okay... I guess you have more tools than me anyway. Or maybe if you need data from a different setup, I'd help.
Motherboard: ASUS P5Q
CPU: Intel Core 2 Quad Q9450 @ 3.20GHz (1.07v vCore! )
RAM: 2GB Kingston HyperX 800MHz
GPU: MSI Radeon HD 4870 @ 780/1000 (default)
One hundred years from now It won't matter
What kind of car I drove What kind of house I lived in
How much money I had in the bank Nor what my cloths looked like.... But The world may be a little better Because, I was important In the life of a child.
-- from "Within My Power" by Forest Witcraft
One hundred years from now It won't matter
What kind of car I drove What kind of house I lived in
How much money I had in the bank Nor what my cloths looked like.... But The world may be a little better Because, I was important In the life of a child.
-- from "Within My Power" by Forest Witcraft
One hundred years from now It won't matter
What kind of car I drove What kind of house I lived in
How much money I had in the bank Nor what my cloths looked like.... But The world may be a little better Because, I was important In the life of a child.
-- from "Within My Power" by Forest Witcraft
The problem is this.... at high res ( > say 1600x1200), depending on the game, the GPU becomes a major factor. The bus mastering in pushing textures around becomes a factor. At typical game play settings, the C2Q really holds no major advantages over the Phenom ...
His reason for it is wrong.
One hundred years from now It won't matter
What kind of car I drove What kind of house I lived in
How much money I had in the bank Nor what my cloths looked like.... But The world may be a little better Because, I was important In the life of a child.
-- from "Within My Power" by Forest Witcraft
Bookmarks