![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Can this 16x SLI bencmark be right?
- Thread starter Ingonuts13
- Start date
Lazn_Work said:
Look closely, that's only 2x AA in the 2nd graph. And the 1st graph doesn't reference AA at all. Neither refer to SLI-AA (and SLI-AA must either be 8x or 16x I believe). The text accompanying really doesn't do much to clear it up. And things are further confused by them referring to "8x" when referring to the old bus speed rather than AA settings.
I don't think Anandtech was aware of the hothardware review or the question being debated here.
H
Because it's relatively safe to assume that the bus bandwidth increase would have similar effects across platforms, especially given that NVIDIA build the chipset "platform" for both.J-Mag said:
So, smarty pants, do you expect AGP 8X and AGP 4x to behave differently across AMD and Intel platforms?
Hurin said:
If you take an Intel SLI board and an AMD SLI board and then compare the differences between Single GPU performance and SLI performance you will notice a greater improvement on the AMD platform than on the Intel platform. Therefore, there might be a NEED for the intel side to have more bandwidth.
Anyway AGP is a standardized spec across platforms. Is SLI?
J-Mag said:
No. There is a greater improvement because the AMD platform is better at games. All SLi benchmarks will naturally be better on the AMD processors as a result of their superior gaming performance. Not because they have different bandwidth needs when it comes to video cards. This notion is absurd as PCI-E is PCI-E. Both platforms impliment it the same way. The reason for the difference in chipsets is due to AMD having an integrated memory contoller into the CPU, and Intel doesn't.
Intel and AMD based rigs are very hard to compare for something like this as the technology behind them is like night and day.
Hurin said:
Exactly correct.
This article shows that when SLi AA modes are used, there is a bennefit to having 8 more lanes worth of bandwidth per slot. Nothing more.
SirFragalot has already addressed most of what youv'e said (thanks!). You should also be made aware that you aren't even asking the right question here. AGP is a bus standard just like PCI-E. SLI is a technology for running two video cards. By equating SLI and AGP, you're comparing apples and oranges. The proper question (which you wouldn't ask because it's obvious) is: "AGP is a standardized spec across platforms. Is PCI-E?"J-Mag said:
That answer: Of course it is. Which is why guaging an increase between 8X PCI-E performance and 16X PCI-E performance on an Intel platform is useful in determining the validity of hothardware's report of a dramatic performance increase between 8X PCI-E vs 16X PCI-E SLI-AA on AMD platforms.
Sir-Fragalot said:
Then why does the Anand Article show an "Once we changed the standard benchmark settings to include 2x AA and 16X AF, the benchmarks ended up favoring the Asus P5N32-SLI Deluxe by 25%. "
Then the Hothardware article on the AMD platform you don't see improvement at all? (yes they are tesing different games, but at least they tried 4xAA on Farcry)
Sir-Fragalot said:
Which is what I said in the first place, it is pointless to compare.
The drastic performance increase we are discussing here manifests itself only when SLI-AA is used. Not just AA. Not AA on an SLI system. But a discreet AA rendering mode called "SLI-AA" that must be specifically enabled and selected. This mode shares the AA duties between the cards. When that is selected, as hothardware reports, there is a dramatic jump in performance between the Asus motherboard using 8x PCI-E and the Asus motherboard using 16x PCI-E.J-Mag said:
Now, if you have an Intel system with 8x PCI-E and another Intel system with 16x PCI-E, and a similar jump is reported, that is good information that can be used to confirm what hothardware reported. It is not silly or stupid to compare Intel and AMD platforms in this instance because the components in question are really just the video cards and the PCI-E bus. And a jump on both systems would be indicative.
Sadly, Anandtech didn't really seem to test SLI-AA in a way that would be helpful in confirming hothardware's findings. But, had they tested an Intel system in that manner, it would indeed have been useful.
Clear enough for you?
H
Hurin said:
I am not worried about the bus performance. I am worried about SLI performance.
AGP is a technology which is standardized across platforms and so is PCI-E, I know this.
SLI is a technology as well and it IS NOT standardized across platforms. There is different core logic on the chipsets themselves.
Hurin said:
OK well maybe it is useful to some people, but not me. Assumptions are worthless from a scientific standpoint, and are technically a logical fallacy.
Either way we cannot disprove each other unless another review comes out and uses FEAR with a 32x PCI-E AMD motherboard. I will bet you anything that you wont see a 25% increase like you did on the Intel side.
Hurin said:
I know what SLI AA is, you don't have to be condescending... And yes I agree there is only a performance increase on the AMD platform in SLI AA modes.
What I am talking about is there is a 25% increase in performance in standard AA modes on the INTEL platform but no increase on the AMD side.
Hurin said:
It is silly to compare because there are totally different results showing at standard AA modes. INTEL gets an increase (according to Anand) and AMD doesn't (according to Hot).
Hurin said:
I agree
Clear enough for YOU?Hurin said:Clear enough for you?
Sigh. . .
However, your statements are quickly becoming more and more contradictory and untenable. For example:
Your sole contribution to this thread is to claim that we are somehow silly or lame for looking at results from an Intel system while trying to determine whether hothardware's findings are valid where an increase in SLI-AA performance is concerned. Yet, you do now admit that comparing the Intel and AMD results could be useful in determing whether hothardware goofed:
H
I suggest you go back and look at your first post in this thread and reconsider who is being "condescending" here.
However, your statements are quickly becoming more and more contradictory and untenable. For example:
I'm not sure how you can say this when you then go on to agree with me that Anandtech did not properly test for an increase in SLI-AA performance on the Intel system.
But, again, as you concede, Anandtech did not test properly to verify or discredit hothardware's findings. Are you saying that, had Anandtech actually tested 8x SLI-AA and 16x SLI-AA between the 8x PCI-E and 16x PCI-E platforms, that those results would have been "silly" or "meaningless" in determining if hothardware goofed or not?
Your sole contribution to this thread is to claim that we are somehow silly or lame for looking at results from an Intel system while trying to determine whether hothardware's findings are valid where an increase in SLI-AA performance is concerned. Yet, you do now admit that comparing the Intel and AMD results could be useful in determing whether hothardware goofed:
Might I therefore respectfully suggest that you contribute to another thread more in-line with your goals? The results that you malign from an Intel board would go a long way in satisfying the question raised by the original poster and the purpose of this thread . Up to this point, your posts have amounted to little more than subtle thread-crapping and/or an attempt to hijack the topic and redirect it away from its intended topic (whether the SLI-AA performance improvement is real).
H
kirbyrj
Fully [H]
- Joined
- Feb 1, 2005
- Messages
- 30,693
Matrices said:
No, you have a couple different facts mixed up
. First, the original issue was an "ultra" board running SLI with certain drivers...which was locked down by Nvidia pretty quick. Second, a modded Ultra chip will actually function like a 8x+8x SLI board because essentially it becomes the SLI chipset (just flash the board to the SLI bios and go I guess...never tried it myself).Hurin said:
Because i mean the only time one sees a performance increase on the AMD platform is when in an SLI AA mode. I did not mention Intel in this statement. Although I can see how my failure to use proper punctuation could lead to your confusion.
Hurin said:
You sure do infer well my friend.
Hurin said:
I never said anyone was lame or silly. I only said you were the delusional one. Anyway it is the concept of inferring results across platforms because they use the same bus, I called silly.
Hurin said:
Feel free to suggest all you want. Oh and yes my primary goal was to hijack this thread from it's inteded topic. Projection sure is fun, isn't it?
I don't believe this stands up to scrutiny:
"No. There is a greater improvement because the AMD platform is better at games. All SLi benchmarks will naturally be better on the AMD processors as a result of their superior gaming performance."
It's utterly irrelevant what processor you are running when you are talking about real-world gaming performance between a P4 2.8+ and an AMD643000+ at 16x12 4xAA. That is entirely GPU-bound. We're not talking about running Q3 at 8x6 and that is meaningless.
The bottom line here is the Anandtech article shows 25% performance increase on 16xPCI-E versus PCI-E in SLI, in FEAR, at 16x12, 2x REGULAR AA, on an Intel board, compared to another Intel board.
What is so hard to understand about this? Unless you can claim that for some reason such results cannot be reproduced in an AMD situation involving 8x PCI-E vs. 16x PCI-E, there is no basis for dispute.
"No. There is a greater improvement because the AMD platform is better at games. All SLi benchmarks will naturally be better on the AMD processors as a result of their superior gaming performance."
It's utterly irrelevant what processor you are running when you are talking about real-world gaming performance between a P4 2.8+ and an AMD643000+ at 16x12 4xAA. That is entirely GPU-bound. We're not talking about running Q3 at 8x6 and that is meaningless.
The bottom line here is the Anandtech article shows 25% performance increase on 16xPCI-E versus PCI-E in SLI, in FEAR, at 16x12, 2x REGULAR AA, on an Intel board, compared to another Intel board.
What is so hard to understand about this? Unless you can claim that for some reason such results cannot be reproduced in an AMD situation involving 8x PCI-E vs. 16x PCI-E, there is no basis for dispute.
Matrices said:
Well, I actually doubt the performance difference is all that good. I have had full x16 PCI-E lanes for each of my graphics cards for quite some time, while I haven't benchmarked my rig with the newest drivers, I seriously doubt I'd get a 25% performance increase.
What we doubt, is the extra bandwidth on the PCI-E lanes is actually having any real world impact in performance.
Perhaps you are correct. On the AMD platform the HH benches show no difference at normal AA. Perhaps this means there is something exclusive to the Intel chipset going on here. I think more benchmarks on more configurations is needed, and above all what's needed is an explanation of why there is such a drastic increase on the Intel platform.
Uh, okay. . . you first came here and pointed out how ridiculous you thought looking at an Intel board was. So, no, you didn't say it was "silly" you merely attempted to convey how ridiculous you thought we were for analyzing Intel board performance. Then, oddly, you admitted that analyzing Intel results would be useful. Now you're back to calling doing so silly. So, really, I don't know what to do with you at this point. But I think everyone reading this thread can see that you were confused, gradually grasped what we were talking about, and are now just sorta flailing aroundJ-Mag said:
RESTfulADI
2[H]4U
- Joined
- Feb 20, 2005
- Messages
- 2,218
we cannon compare intel SLI to amd SLI, especially to see the difference between 8x SLI and 16x SLI. Intel cpu's are much less powerful and many benchmarks will be cpu limited, not showing the full performance potential of 16x SLI. the fact that there was up to a 25% difference in the anandtech benchmarks is actually very encouraging and amd performance might be even better when benchmarked properly. also, since intel cpu's lack an integrated memory controller (they suck), nvidia had to build one into the northbridge on the sli chipset and that makes direct comparison to amd sli invalid as well.
Uh, the new review uses 81.85 drivers.Lazn_Work said:
Hothardware's used 81.84 drivers.
Anyways, when I went back to check on this, I noticed that Hothardware had indeed fixed their review. The increase in SLI-AA performance has now been debunked and corrected by Hothardware themselves. The difference between the two boards was caused by a BIOS problem on the older board that made it perform much slower than it should have.
H