Go Back   Hardware Canucks > HARDWARE > Video Cards

    
Reply
 
LinkBack Thread Tools Display Modes
  #21 (permalink)  
Old July 21, 2008, 10:56 AM
Rookie
 
Join Date: Jul 2008
Posts: 8
Default

Quote:
Originally Posted by SKYMTL View Post
Even in that review all tests between Crossfire solutions on PCI-E 2.0 boards are WELL within the margin of error. Run any of those benchmarks used 5 times and you will probably get 5 different results. That is why many sites out there (this one included) use AVERAGES of 3 or more benchmark runs.

I am not sure about LH's methodology but it doesn't mention anywhere how they tested each game.

As I mentioned, these differences can even be chalked up to the variances between one motherboard and another as well. Check reviews which pit one X38 against another and you will see benchmark scores which are quite a bit different from one game / synthetic benchmark to the next. This has NOTHING to do with PCI-E bottlenecking but rather how one board may perform slightly differently than another.
I've never seen a x38 board that was pitted against another x38 board showing any significant performance scores that wasn't about on par with each other regardless how many bench runs you gone through. Perhaps you would like to post a site showing this since alot of the major sites disagree with this. It's all about the chipset and if each board does have everything basically the same then at stock speeds, there shouldn't be any significant difference at all providing each board is using the same chipset driver.

Perhaps you are referring to overclocking since that's the only main significant inconsistency between alot of boards since there are different caps used just like on the p45 vs x48's. If you did any extreme overclocking on both boards then you would know that the P45 is a more solid chip than the X48 in itself allowing you to push beyond the 450fsb barrier on a quad in more cases than the x48.

Also your statement of margin of error is totally subjective since you are only posting comments without providing objective facts. IF tweaktown is only doing 1 benchmark run, whereever it is that they say they are or aren't, then going off of that, you can provide with evidence that their 1 benchrun testing is a hoax and should be ran multiple times displaying the avg results. If you do not do this then how are we suppose to intake your words other than a grain of salt?
Reply With Quote
  #22 (permalink)  
Old July 21, 2008, 11:57 AM
SKYMTL's Avatar
HardwareCanuck Review Editor
 
Join Date: Feb 2007
Location: Montreal
Posts: 11,839
Default

Quote:
I've never seen a x38 board that was pitted against another x38 board showing any significant performance scores that wasn't about on par with each other regardless how many bench runs you gone through. Perhaps you would like to post a site showing this since alot of the major sites disagree with this. It's all about the chipset and if each board does have everything basically the same then at stock speeds, there shouldn't be any significant difference at all providing each board is using the same chipset driver.
Looks like 5-10% difference in some cases to me:

bit-tech.net | Review - DFI LANParty LT X48-T2R

X48 heavyweights go head to head - The Tech Report - Page 7

Or how about we go a little farther off the beaten track with some AMD:

bit-tech.net | Review - AMD 770X Motherboard Duel

I could easily pull out more but I think these three illustrate my point quite well. It has very little to do with the chipset in many cases but rather the BIOS and general board design. In between the last official and the new Beta BIOS for my own DFI X38 board I have seen an increase in PCI-E card performance due to God knows what. It is a simple matter of the communication between chipset and the PCI-E lanes which can vary between one BIOS and the next.

You chould have the best chipset in the world with a crappy BIOS and it will still turn out to be a shitty board.

Granted, many times there will be little to no difference between boards with the same chipset but there are other times that there WILL be a difference which cannot be discounted.

How about we move on to single card performance from one chipset to the next. Let's take for example X48 vs P45. Both PCI-E 2.0 boards having a 16x lane when running in single card mode:

bit-tech.net | Review - MSI P45 Platinum

Yup, that's right. Up to a 15% difference when going from one PCI-E 2.0 chipset to the next when taking minimum FPS into account.

So, tell me again how this would NOT translate into someone (ie: Tweaktown, LH or any other site) ASSUMING there is a difference in CROSSFIRE performance when going from one chipset to the next. Heck, we just saw that there are difference in single card performance on two PCI-E 2.0 mobos with different chipsets.

Discounting these differences and making a leap of judgment by saying "P45 will bottleneck xxxxx cards" is to me ridiculous. Granted, it makes a great article and gives your site lots of hits but in the end it proves nothing other than the fact that there may be differences in BIOS or board issues which the reviewer is not taking into account.



Quote:
Also your statement of margin of error is totally subjective since you are only posting comments without providing objective facts.
Explain to me how a 5-10% difference when measuring 60+FPS in some games is subjective. I won't even go into posting the number of sites that test games more than once for this very reason which includes MYSELF since I see it on a daily basis when testing GPUs.

Last edited by SKYMTL; July 21, 2008 at 12:03 PM.
Reply With Quote
  #23 (permalink)  
Old July 21, 2008, 12:04 PM
CanadaRox's Avatar
Allstar
F@H
 
Join Date: Feb 2008
Location: Scarborough (Toronto)
Posts: 614
Default

I think this article is also worth looking into: http://www.tomshardware.com/reviews/...2-0,1915..html
Its a bunch of test with different cards running at PCIe 2.0, 1x 4x 8x and 16x. Even with a 9800GX2, most games barely change at all, even at PCIe 2.0 4x. If a 9800GX2 barely has a change in performance from PCIe 2.0 16x to 4x, how would a 4850 have that much of a drop from PCIe 2.0 16x to 8x?
Direct link to the 9800GX2 part: http://www.tomshardware.com/reviews/...0,1915-10.html
__________________
Project: Black and White
i7 920 D0 | 3 x 2GB DDR3 | EVGA X58 SLI LE
XFX 4890 | Corsair HX750 | Corsair Obsidian
Reply With Quote
  #24 (permalink)  
Old July 21, 2008, 01:19 PM
Rookie
 
Join Date: Jul 2008
Posts: 8
Default

Quote:
Originally Posted by SKYMTL View Post
I'm going to use your "margin of error" logic by saying that 5-10% difference is hardly a significant reason to say that 1 board with the same chipset is faster than the other. Mainly because neither site displayed if they did multiple test runs or not which could have in fact shown a different outcome. That doesn't really prove anything.

Quote:
Originally Posted by SKYMTL View Post
Granted, many times there will be little to no difference between boards with the same chipset but there are other times that there WILL be a difference which cannot be discounted.
If you can find an actual signifigant performance difference between boards with the same chipset used then I'll agree but, please don't say this when posting a few benches done on 1 test run when the margin is very miniscule.

Quote:
Originally Posted by SKYMTL View Post
How about we move on to single card performance from one chipset to the next. Let's take for example X48 vs P45. Both PCI-E 2.0 boards having a 16x lane when running in single card mode:

bit-tech.net | Review - MSI P45 Platinum

Yup, that's right. Up to a 15% difference when going from one PCI-E 2.0 chipset to the next when taking minimum FPS into account.
This clearly doesn't prove anything since I even mentioned before that these 2 chipsets obviously have different qualities which would allow 1 to have more overclocking success than the other. It would only be fair to say that 1 is faster than the other. Just the same as how X48 is nothing more than a fine picked X38 both being the same chip. Just the same as the differences between P35 and P45.

Quote:
Originally Posted by SKYMTL View Post
So, tell me again how this would NOT translate into someone (ie: Tweaktown, LH or any other site) ASSUMING there is a difference in CROSSFIRE performance when going from one chipset to the next. Heck, we just saw that there are difference in single card performance on two PCI-E 2.0 mobos with different chipsets.
The differences as you make it sound would appear to someone that it would be a visual difference which it is not. When I said signifigant, I meant as in visualy which is far beyond looking at the difference on paper at small percentages as you posted above.

I kind of like CanadaRox's responce to the issue by posting a good site with results that I have not seen yet. I don't have time to get deep into the review but, I'm sure if it was questioned then the reviewers there would retest since that is a good site to use.
Reply With Quote
  #25 (permalink)  
Old July 21, 2008, 01:51 PM
SKYMTL's Avatar
HardwareCanuck Review Editor
 
Join Date: Feb 2007
Location: Montreal
Posts: 11,839
Default

Quote:
The differences as you make it sound would appear to someone that it would be a visual difference which it is not. When I said signifigant, I meant as in visualy which is far beyond looking at the difference on paper at small percentages as you posted above.
I completely agree with you but looking at the LH results you posted, I don't see any "significant" differences either that can't be accredited to BIOS irregularities between the different boards.

Let me put this into different terms: I am not refuting that there may be some applications and graphics cards in the near future which will benefit from higher bandwidth PCI-E interfaces (I think the Toms test shows that Flight Simulator benefits quite a bit) BUT other than the method Toms used, there is no consistent way of testing different bandwidths. The main issue is that getting an even playing field from one motherboard to the next is impossible and at this point it seems like only Toms is testing via a method that effectively eliminates mobo irregularities. Unfortunately, there are still some exceedingly odd results in that article by Toms where the 8x link surpasses the performance of the 16x but it is only by a few percentage points.

My whole stance is centered around the fact that some sites have posted results from different mobos, with different BIOS revisions, with different chipsets and claimed that an 8x / 8x PCI-E 2.0 interface will bottleneck the cards. Personally, I think this is a leap of faith rather than a well though-out conclusion.

There is a reason we haven't tested this same thing here and that is because I have not come up with a testing methodology (other than copying Tom's) that would give consistently accurate results.
Reply With Quote
  #26 (permalink)  
Old July 21, 2008, 02:51 PM
DK2 DK2 is offline
Allstar
F@H
 
Join Date: Apr 2008
Location: GTA
Posts: 821
Default

I look at it this way, if the FSB is only 333Mhz, quad pumped thatís 1.333GHz.
No data is going anywhere faster than this rate.
The problem might be more size of the burst rate of data being transferred,
and the system memory required by a given application, since the NB is a shared resource.
Reply With Quote
  #27 (permalink)  
Old July 21, 2008, 04:36 PM
Rookie
 
Join Date: Jul 2008
Posts: 8
Default

When the results are soo close then it's really hard to even say that it was the result of the bios.

One thing to consider about the benchmark done on the Tom's site is that the 9800gx2 has been bested by the 4870 at times even though the 9800gx2 is a dual 8800GTS G92 on 2 PCB's. This is SLI vs 1 card solution and the difference from 8x to 16x is minor with the 9800gx2. WHat I would do if I were a reviwer at this point would be to do a 8x/8x vs 16x/16x using the 4870's. It's simply too far of a longshot to say it either way with the 4850 so I would like to see the 4870 tested.

That would seriously end this debate.

Last edited by OblivionLord; July 21, 2008 at 06:52 PM.
Reply With Quote
Reply


Thread Tools
Display Modes