Hypothesis on the Conroe benchmarks results [Update]

[Diasper]

Extra links:
Rahuul Sood's weblog of Voodoo PC expressing original concerns over the poor choice of the DFI board and more particularly the problems with the bios they choose.

Anand's follow up test using the latest bios - FEAR corrected but otherwise 20-25% faster in games!

--------------------------------------------

Edit:

With anand's follow up test I happily stand corrected. With the examination of the bios, extra benchmarks, re-testing of FEAR and uniformity of the results they look very believable.

Conroe seems 20-25% faster - certainly a good job and all the Intel engineers involved with Conroe and with the technologies in it should be proud of what they've achieved

The only remaining question is over the driver changes they made but it seems fair to trust what the Intel representative said.

 

[boeki]

intel picked crossfire mobos for a reason.

i wonder how the results would have been if the conroe was run on an nf4 sli intel edition mobo and the fx-60 on a dfi sli-dr expert or on an a8n32 sli mobo.

 

[Keysplayr]

Actually, I would expect Conroe to do even better than the benchies we saw. After all, it's not do for six months. A good amount of time to refine the manufacturing process for better yields, higher clocks. I don't know what you're going on about FEAR for. It's the only one that was not an Intel demo, yet the only one you are trashing? What about the media encoding?

Anyway, it's fun to speculate, but Intel has six months to perfect. Yes, the only reason they let out so very much information at IDF was to halt any further purchases of AMD processors. And it is working beautifully. Look how many AMD users are salivating over this Conroe. Except that one guy who says that even if Intel was 800X faster, he still wouldn't buy it for ethical reasons, LMAO. So, this would have been a good topic six months from now. And don't worry, AMD is saying OMFG right now and Hector will light the fires under everyone to do the best they can against the Intel monster. But remember man, nobody stays on top forever. I guess it's Intels turn once again. Somewhere down the road, it will be AMD's turn again. No biggie.

 

[Keysplayr]

Originally posted by: boeki
intel picked crossfire mobos for a reason.

i wonder how the results would have been if the conroe was run on an nf4 sli intel edition mobo and the fx-60 on a dfi sli-dr expert or on an a8n32 sli mobo.

Why wonder? Just ask around the forums with somebody using Crossfired X1900's and at least an X2 4800 o/c'd to 2.8GHz, and then run the FEAR demo and any other bench they can duplicate.

 

[Diasper]

Originally posted by: keysplayr2003
Actually, I would expect Conroe to do even better than the benchies we saw. After all, it's not do for six months. A good amount of time to refine the manufacturing process for better yields, higher clocks. I don't know what you're going on about FEAR for. It's the only one that was not an Intel demo, yet the only one you are trashing? What about the media encoding?

Anyway, it's fun to speculate, but Intel has six months to perfect. Yes, the only reason they let out so very much information at IDF was to halt any further purchases of AMD processors. And it is working beautifully. Look how many AMD users are salivating over this Conroe. Except that one guy who says that even if Intel was 800X faster, he still wouldn't buy it for ethical reasons, LMAO. So, this would have been a good topic six months from now. And don't worry, AMD is saying OMFG right now and Hector will light the fires under everyone to do the best they can against the Intel monster. But remember man, nobody stays on top forever. I guess it's Intels turn once again. Somewhere down the road, it will be AMD's turn again. No biggie.

I think the only improvement we'll see to Conroe is a move to DDR800 correct me if I'm wrong. I'm sure that'll net a couple more % increase.

I wasn't 'trashing' FEAR - rather as a game it's poorly coded compared to the other masses of games out there and if you wanted to be more representative you'd take a benchmark like Half Life 2 - also it is a game which is known to be CPU intensive as opposed to GPU intensive like FEAR. I'm saying they could have done better and been more open - the fact they didn't and went to some trouble I can only assume is to put more 'shine' on their product. What I'm saying is 40% is not believable especially as it doesn't fall within the margins of the other results ~20%

Media encoding has always been an Intel strong-point and Intel has always had lots of expertise there. Considering Prescotts general inefficiency it was pretty good in encoding. Now assuming a souped up and more efficient architecture from Intel the results are what I expected.

Sure, I agree everything is cyclical when it comes to performance leadership. History has certainly shown that. Heck within 18 months down the line after Conroe they'll be something even better I'm sure. Everything's perpetually becoming 'obsolete'.

Maybe I'm too cynical because of all the marketing around (and the BS spewed by all companies in the past in situations like this) but I'm aware this preview is first and foremost a marketing excise using custom Intel benchmarks (with the exception of FEAR) - given the nature of things you have to assume 'polish has been added'. As such I'm calling it out as that. I'm sure when Conroe comes out it will be the fastest but I think there isn't enough perspective yet and people aren't recognising this preview as marketing it is.

 

[DAPUNISHER]

They have HL2 and others right here and all I see is Conroe domination. I have already seen it posted that "well, they need to up res and AA/AF because it is X1900 XT crossfire, so in reality you would use more GPU limited res and settings" Which is nonsense because they are testing CPUs here. No matter what res and settings you apply Conroe is going to be faster judging by these numbers. Also, I don't see anyone using that argument when AMD is trouncing the P4 variants in similar benchies.

 

[Munky]

I dont care if Intel shows a 200% lead, until that lead can be established in a controlled environment by reputable review sites, it means nothing to me. I dont know why all these people are getting their panties wet over these "benchmarks."

 

[Diasper]

Next up is a Half Life 2 Lost Coast demo, once more an Intel supplied demo

I wish it had been a standard demo.

Outside of fanboys, AMD64 troucing P4 has been independently verified by countless websites is why.

As I maintain, significant performance increases with Conroe will be real over current AMD64 and certainly looks to be AM2 as well - but I'm still calling this as a marketting effort - the Intel representatives could have done a much more transparent test. Making custom benchmarks is going out of your way and thus deliberately making them un-transparent.

 

[DAPUNISHER]

No arguments there, they doubtless cherry picked. However, that is only going to make the lead a bit less brutal. It is evident IMHO, that it can outperform AMDs' current best@lower clockspeed. Should the pricing hold up, Intel has a serious winner on the way.

 

[Zstream]

Originally posted by: munky
I dont care if Intel shows a 200% lead, until that lead can be established in a controlled environment by reputable review sites, it means nothing to me. I dont know why all these people are getting their panties wet over these "benchmarks."

 

[Diasper]

Originally posted by: DAPUNISHER
No arguments there, they doubtless cherry picked. However, that is only going to make the lead a bit less brutal. It is evident IMHO, that it can outperform AMDs' current best@lower clockspeed. Should the pricing hold up, Intel has a serious winner on the way.

That's quite a nice way of putting it. I'm not going to excited at the benchmarks, certainly while they haven't yet been confirmed - as yet it is just marketing. BUT to get anything like those figures even if tweaked means you have a very solid processor behind it which does have some sort of real performance lead - probably at least 15% per same clock of what AM2 is likely to offer. HOWEVER, I'll still call Intel's benchmarking methods as not being marketting, not transparent and therefore likely to be tweaked to some greater or lesser degree. I want to know how much so there's less of this panic/excitement etc

BUT if there is anything to be excited about is that it should offer that excellent base price and performance and still be nicely overclockable after - such that the difference even if not as great as it currently stands should be very compelling.

 

[Fern]

Originally posted by: munky
I dont care if Intel shows a 200% lead, until that lead can be established in a controlled environment by reputable review sites, it means nothing to me. I dont know why all these people are getting their panties wet over these "benchmarks."

I hear ya man!

I'm seeing some threads like OMG! I gotta dump my X2 and get Pentium D to hold me over for 6 months and other silly over-the-top reactions.

If that little dog 'n pony show was supposed to generate a bunch of market hype it sure as h3ll worked.

Great fodder for AEG/Intel moles working the crowd here.

EDIT: Uh Oh, looks like this w/b locked. Every time I post in a Conroe thread it's locked

Fern

 

[Hacp]

Intel just Pwned AMD. This 2.66GHZ offering is not even high end. Get it guys? Its not even high end. So the high end should produce a substantial lead.

 

[dmens]

What's with the constant denial? You can draw two conclusions:

1. Intel engineers have no ethics and faked the numbers.
2. Merom is good.

Knowing IDC, I doubt they'd tolerate anyone tainting their product with fake marketing bullshit. So, to all those people insisting the numbers are rigged, or they mean nothing until independent benchmarks are released, have it your way. Continued insistence of fraud on intel's part without any proof is just evidence of desperate fanboism.

All those numbers say is that in 4 months time, there will be a mid-range desktop chip from intel that stacks up as demonstrated against the expected high-end X2 part from AMD. Digest that however you want.

 

[Jeff7181]

Originally posted by: dmens
What's with the constant denial? You can draw two conclusions:

1. Intel engineers have no ethics and faked the numbers.
2. Merom is good.

Knowing IDC, I doubt they'd tolerate anyone tainting their product with fake marketing bullshit. So, to all those people insisting the numbers are rigged, or they mean nothing until independent benchmarks are released, have it your way. Continued insistence of fraud on intel's part without any proof is just evidence of desperate fanboism.

All those numbers say is that in 4 months time, there will be a mid-range desktop chip from intel that stacks up as demonstrated against the expected high-end X2 part from AMD. Digest that however you want.

While all that may be true... it's suspicious because of the motherboard selection and outdated BIOS used. It's not like Intel can't spend $1500 and go to newegg.com and buy some popular hardware for a fair comparison. They could spend $150,000 on every possible motherboard/CPU combo and it wouldn't make a dent in their marketing budget. So the question is why was this uncommon motherboard used, and why was this old as dirt BIOS used? Does a BIOS that old even recognize a dual core CPU? Could it be that the second FX-60 core sat idle because the BIOS didn't know what to do with it? That would sure explain the performance differences since multi-threaded video card drivers have the potential to make a decent impact on performance.

 

[Fern]

Originally posted by: dmens
What's with the constant denial? You can draw two conclusions:

1. Intel engineers have no ethics and faked the numbers.
2. Merom is good.

Knowing IDC, I doubt they'd tolerate anyone tainting their product with fake marketing bullshit. So, to all those people insisting the numbers are rigged, or they mean nothing until independent benchmarks are released, have it your way. Continued insistence of fraud on intel's part without any proof is just evidence of desperate fanboism.

All those numbers say is that in 4 months time, there will be a mid-range desktop chip from intel that stacks up as demonstrated against the expected high-end X2 part from AMD. Digest that however you want.

I say pile on the hype!

I'm lookin to pick up some sweet used Opty's in the hysteria to stampede over to this new chip.

Fern

 

[dmens]

It seems like any hardware picked for this public pre-release demo would be criticized. Regardless, my point is, can the average enthusiast even make an educated guess about what's possible, or what's not posible with a machine? It takes an in-depth knowledge of architecture just to make generalized statements on a chip's performance characteristics. Statements like "these numbers are too ridiculous so they have to be fake" sound absurd coming from anyone except people intimately familiar with what makes an x86 machine tick.

On the side, if the demo team deliberately picked hardware to cripple the AMD machine, then that would fall into the fraud category. I can understand the flurry of allegations from the obvious distress certain people are undergoing (LOL), but in the end the real numbers will speak for themselves. If the numbers were faked, that'd be a really sleazy move. I might have to go work for AMD.

 

[robertk2012]

Originally posted by: Diasper


Any realistic gameplay will be GPU limited for a long time to come.

Discuss...

Well it was thought that fear was GPU limited at those resolutions with the fx-60. Apparently not.

 

[Jeff7181]

Originally posted by: robertk2012

Originally posted by: Diasper


Any realistic gameplay will be GPU limited for a long time to come.

Discuss...

Well it was thought that fear was GPU limited at those resolutions with the fx-60. Apparently not.

Unless the old BIOS prevented the second FX-60 core from being used, and the big performance difference is due to the multi-threaded display drivers offloading some of the work the GPU normally does to the second CPU core.

 

[Avalon]

Originally posted by: Jeff7181

Originally posted by: robertk2012

Originally posted by: Diasper


Any realistic gameplay will be GPU limited for a long time to come.

Discuss...

Well it was thought that fear was GPU limited at those resolutions with the fx-60. Apparently not.

Unless the old BIOS prevented the second FX-60 core from being used, and the big performance difference is due to the multi-threaded display drivers offloading some of the work the GPU normally does to the second CPU core.

The multi threaded display drivers only resulted in roughly a 5% performance boost. Other than that, FEAR doesn't benefit from dual core, so this would not account for the 30-40% advantage Conroe has at a resolution that is not considered CPU limited in FEAR.

I'm with Anand's gut instinct, in that these results are plausible.

 

[Jeff7181]

Originally posted by: Avalon

Originally posted by: Jeff7181

Originally posted by: robertk2012

Originally posted by: Diasper


Any realistic gameplay will be GPU limited for a long time to come.

Discuss...

Well it was thought that fear was GPU limited at those resolutions with the fx-60. Apparently not.

Unless the old BIOS prevented the second FX-60 core from being used, and the big performance difference is due to the multi-threaded display drivers offloading some of the work the GPU normally does to the second CPU core.

The multi threaded display drivers only resulted in roughly a 5% performance boost. Other than that, FEAR doesn't benefit from dual core, so this would not account for the 30-40% advantage Conroe has at a resolution that is not considered CPU limited in FEAR.

I'm with Anand's gut instinct, in that these results are plausible.

IIRC nVidia's multi-threaded drivers produced up to a 20% increase in performance depending on the application and the rest of the hardware. I could be mistaken, though.

 

[HurleyBird]

Originally posted by: DAPUNISHER
No arguments there, they doubtless cherry picked. However, that is only going to make the lead a bit less brutal. It is evident IMHO, that it can outperform AMDs' current best@lower clockspeed. Should the pricing hold up, Intel has a serious winner on the way.

Remember back in the days of Quake 3 benchmarking? Even though all the timedemos were created by ID AMD or Intel could have a pretty good lead depending on what time demo was used. Now remember that these timedemos were created by *Intel*

 

[Diasper]

This is taken from the other thread but is worth evaluating

Originally posted by: Quinton McLeod

Originally posted by: HurleyBird

A disclaimer: Conroe is going to be good, it's going to be a good chip and will make Intel competitive again. However, you guys have all slipped into some kind of reality distortion feild, and I'm going to try to pull you all out of it.

Now lets get down to it. The fact that all the game tests but FEAR were set up by Intel is no coincidence, as FEAR is the only test that Intel didn't have to rig.

You see, you guys think you are seeing some sort of marvelous new architecture. You aren't. You are seeing a good architecture that is around par with AMD64 that has lots, and lots of cache. The thing you guys are forgetting here is that Conroe has a 4MB *resizable* cache. In single threaded apps, you are seeing a core with 4MB of cache all to itself. Look at Quake 4, and the huge drop in leadership over AMD when both cores need to share that 4MB of cache.

Now why don't you guys also take a trip in the 'way back machine' to the days of Quake 3. ID made an assortment of time demos, and the performance difference between these time demos could be huge. AMD or Intel could be winning by a significant amount depending on which timedemo was used. Now, ye foolish, remember that these are *Intel created* timedemos. Who knows how much they could have swayed performance in their favour? Anywhere from 10-40% could be possible, but I suppose no one knows besides the Intel guys who made the timedemos in the first place. So in single threaded game performance the two chips will likely be even, with perhaps a slight edge to Intel, but in multithreaded tests AMD should win by quite a bit still. Of course that excludes FEAR...

FEAR is *extremely* cache sensitive. Despite a 200MHz clock disadvantage, the 165 beats the 3800+ by a good margin, and don't forget that AMD64 *is not* very cache sensitive. Without an IMF, conroe is a lot more cache sensitive than the A64, and in this case has *4X the L2 cache* as the FX-60 does. FEAR gets no performance advantage from going to 1 to 2GB's of memory, so who knows, a very good portion of the game code might fit in Conroe's L2.

Also The BIOS, which dates from 2003, does not even recognize the FX-60... come on.

EDIT: Even though the bios says copyright 2003 its actually from 2005, and as Rahul Sood points out:

-That Bios *enables Cool 'n Quite by default
-According to DFI the FX-60 is not recognized *and does not function poperly* with this Bios
-2-1-1-1-1 and 4-1-1 Mode Wrong & Fix Read Preamble Table Error (!)
-Fix Fill 3114 SVID&SSID under Cross fire mode (! Crossfire bug !)

Last but not least, the systems use a *modified* video driver. Why on earth would Intel need this? The excuse that it's needed to recognize Conroe doesen't hold water (When have you ever had to update your video card drivers to recognize a new processor?!) and almost cries out optimization. And with ATI's drivers generally sucking (that latest OpenGL fix should have been fixed *years* ago), the recent skype 'optimizations', the fact that ATI is now making a lot of Intel chipsets and this, how can you possibly accept these seemingly extraordinary numbers at face value?

Something to think about.

You're right, now that I think about it.
I have never heard of a single video driver that needed to recognize a processor.

We should all know to never trust benchmarks from a company making the product they're benchmarking. It's almost entirely skewed in their favor. Intel has been guilty of this before.

 

[F1shF4t]

A cpu should not be able to improove gpu limited games performance, its not logical, cause even the best X2 fx 60 will be killed by a lowly 6200 when it comes to games. cpus are useless when it comes to graphics, thats why u have gpu's.

Now saying that to using those setting different cpus should be used, if u benchmark a 3800+ or opteron 170 and fx60 at those fear setting and get slight to no change in performance, those conroe numbers are a bunch of bull. If u get a huge difference, then its possible tha the conroe numbers are true.

Also that they modified the video drivers to detect the cpu smells fishy.

 

[misanthropy]

What sophists these AMD fan boys are--ATI doesn't even have drivers that utilize dual-cores--that's Nvidia. Considering from I have seen, a jump from a 3200 to the latest FX series would not give you the 40% improvement we are talking about in games like FEAR this is quite impressive, and provides insentive for gamers to wait. AMD just got spanked. PWNED