HEXUS.reviews :: WORLD EXCLUSIVE! INTEL CONROE BENCHMARKED

[JumpingJack]

Benchnmark/System Intel Conroe E6700 system Intel Conroe E6600 system HEXUS FX-62 AM2 system HEXUS FX-60 S939 system HEXUS Intel Presler 965 system
ScienceMark 2.0 memory bandwidth (MB/s) 5048.13 5058.14 8144.90 5721.22 5678.11

Just one thought. With the 1/3x more bandwidth. Would this come into play when AMD gets into the 65nm - 45nm technology?

Years past AMD has always been 1 step behind. Since they started development later than Intel "ie.DDR-DDR2". When they reach the same point in development, though. They amply improve upon the process that beats Intel.
Courtesy of Intel 65nm web page:
Our leading-edge transistor technologies include:

Second generation strained silicon with 10-15 percent improved drive current for improved performance

1.2nm gate oxide and 35nm gates for improved performance

NiSi for low resistance cap on gates and source-drains

Lower interconnect capacitance through low-k carbon doped oxide dielectric and 0.7x line length scaling providing increased performance and lower power


Any Thoughts?

There has been no AMD process data reported that exceeds Intel's at anytime -- Intel's 90 nm process significantly out performs at the transistor level AMD's process based on the IEDM data published to date. This is consistent through all the device nodes.

Extra bandwidth from memory only helps if the CPU is capable of receiving that extra width, they only way this will happen is if clocks ramp significantly --- which, again, based on their 65 nm preliminary data looks to be only 10 or 15% above the 90 nm process. If 3.2-3.4 GHz is enough to warrant more data into cache, then yes it will be beneficial.

Intel's Ion/Ioff curves for the last 3 technologies data: http://www.realworldtech.com/page.cf...005001504&p=14

Note the table that compares technology to technology. FYI - higher Ion is better, lower Ioff is better. It's not even close, it's a butt spanking -- this is why, in general Intel chips OC to higher raw clocks than AMD chips. In a K8 vs Netburst comparsion at like technology nodes DO NOT confuse computation performance with transistor performance.

[acrobat]

Thank you Keta for the reply! That was pretty much exactly what i was hoping to hear. I think i will just buy 2 gig of the cheapest DDR2-800 ram and i'll be happy with that. I wont bother OC'ing it because i dont like doing that. But for a few FPS, i really dont mind

July 23rd is longer than i wanted to wait really But i am going to hold out. I have waited so long, another couple of months wont kill me. So i will wait until then, and buy a Conroe, 2 gig of DDR2-800, and a 7900GT, and it will be a HUGE improvement on what i have now And then maybe sometime next year i will upgrade to a DX10 graphics card.

Thank you all for the help. This thread has been really useful to me.

[mxyztplk]

Oxbow,

The greatly lower Conroe bandwidth benchmark result (i.e., vs. that of the FX-62) could be attributable to the dramatically different test memory configurations. Until the systems are tested with equivalent memory configurations, this aspect will remain unclear.

The FX-62 test system was configured with FOUR (4) 800MHz DDR2 DIMMs. By contrast, the Conroe test system was configured with TWO (2) 667MHz DDR2 DIMMs.

That means the aggregate potential DIMM transfer rate of the FX-62 test system was as much as 2.4 times that of the Conroe test system (i.e., ignoring limitations/considerations attributable to the processor designs themselves, or questions regarding how the software allocated/accessed the memory, or other such factors).

The comparative memory bandwidth benchmark test results, as well as potentially other results where memory bandwidth and/or contention is a large factor, could therefore easily have derived from the drastically disadvantaged Conroe test memory configuration.

As has been mentioned earlier in this thread, a HEXUS apples-to-apples comparison will have to await later testing, with comparable memory configurations. Until then, it is simply an interesting fact that Conroe was able to do as well as it did compared to FX-62, with Conroe's potential memory bandwidth so dramatically comparatively crippled.

Alternatively, you can review some of the other benchmarks that have been done elsewhere, e.g. see list at:
http://forumz.tomshardware.com/hardw...ict183765.html

[Oxbow]

Thanks for the feedback fella's. I appreciate it. May you all have a fun and safe weekend.

Take Care All;

Ox

[andyisc00l]

I KNOW HOW TO ACT LIKE A MORON!!

OK here I go...

In order to figure out % better or worse you need a constant( you can't just say "highest number / lowest number" since that messes with "lower score = better")...
(Best score/Worst score) = # ... IF #>1 then (# - 1 = percentage), IF #<1 then (1 - # = percentage)...that yields percent
Just because In "lower score = better" cases the higher score isn't the best, doesn't mean you do highest / lowest... understand?

A good way of thinking about it is... Which score lost? ... how much did it lose by... or in other words... how much does this number need to increase or decrease in order to beat the winning score?

example:

When its "higher is better" you do this...
memory bandwidth (MB/s) (higher is better) AMD 8144.90 , Intel 5048.13

8144.90 / 5048.13 = 1.6134 - 1 = 61.34% better
mathamatically checking:
1.6134 * 5048.13 = 8144


When its "lower is better" you do this...
DIVX (seconds lower is better) - Intel 125 , AMD 164

125 / 164 = 1 - .7622 = 23.78% better
mathamatically checking:
It's different since your taking away numbers in order to match the best score, so you have to do an inverse (i think thats the right word) of the percentage...

164 * (1 - .2378) =
164 * (.7622) = 125


applied throughout the whole set of numbers you get this...

whoever messed up sciencemark bandwith must have some troubles with math lol...

-38.02% ScienceMark 2.0 memory bandwidth actual: -61.34%
- 2.12% ScienceMark 2.0 memory latency -2.12%
+26.41% HEXUS Pifast +20.97%
-12.57% HEXUS Cryptography -14.38%
+19.09% Realstorm Raytracing 2004 correct
+31.20% DivX encode - multithreaded +23.78%
+32.46% WAV conversion multithreaded +24.5%
+15.77% CINEBENCH multi-CPU render correct
+51.32% KribiBench v1.1 - Jetshadow model correct
+39.47% Far Cry - 1024x768 - correct
+ 3.67% Quake 4 - 1600x1200 - 4x AA 16x AF correct
+ 0.88% Splinter Cell: CT - 1600x1200 - 4x AA 8x AF correct

The only thing this really ends up changing that matters is Wav and Divx...

don't forget Intel EE is messed up too.


Possible argument:

DIVX...(always put the larger number on top)
164/125 = 1.3120, or 31.20 percent...but thats saying Intel needs to increase 31.20% in order to match AMD
we want AMD to match intel...

thats why its 125/164 = (1 - #(7622)) = 23.78% percentage... AMD needs to increase (in this case it's decreasing but the term is still the same) 23.78% better in performance... (technically it's decrease the number, but in order to match 125 seconds it has to improve 23.78%)

[Steve]

Hi andyisc00l,

I may do engineering mathematics at degree level, but that doesn't mean I'm a statistician. Still, I'm going to put a statistician's hat on for a moment.

The thing with stuff like percentages is you can use them as comparisons in too many different ways to be definitively useful. That aside, they can still be used to paint a picture.

To save me some time, let's take but one example of yours...

-38.02% ScienceMark 2.0 memory bandwidth actual: -61.34%

Conroe needs to increase its score 61.34% to reach that of the FX-62's, yes... 5048.13 * 1.6134 = 8144.65. However, our figure there is esentially saying that Conroe gets 62% (100-38) the score of the AMD... 8144 * 0.62 = 5049.

So you can argue the points of both figures. What's important here, however, is consistency. I haven't checked other numbers yet, but can probably do so in a little while, to see if they all paint the picture in the same way. As long as the do that, then really there's nothing wrong and I leave hating percentages. If there is something inconsistent, then it probably needs rectifying, and I still hate percentages.

[andyisc00l]

[/B]

Hi andyisc00l,

I may do engineering mathematics at degree level, but that doesn't mean I'm a statistician. Still, I'm going to put a statistician's hat on for a moment.

The thing with stuff like percentages is you can use them as comparisons in too many different ways to be definitively useful. That aside, they can still be used to paint a picture.

To save me some time, let's take but one example of yours...

-38.02% ScienceMark 2.0 memory bandwidth actual: -61.34%

Conroe needs to increase its score 61.34% to reach that of the FX-62's, yes... 5048.13 * 1.6134 = 8144.65. However, our figure there is esentially saying that Conroe gets 62% (100-38) the score of the AMD... 8144 * 0.62 = 5049.

So you can argue the points of both figures. What's important here, however, is consistency. I haven't checked other numbers yet, but can probably do so in a little while, to see if they all paint the picture in the same way. As long as the do that, then really there's nothing wrong and I leave hating percentages. If there is something inconsistent, then it probably needs rectifying, and I still hate percentages.

Yea, well I thought we were dealing with percent/better worse. You lost me there. But saying conroe is 38.02% worse then AMD is really wrong. AMD's score is nearly DOUBLE Intels... the correct forumula...

(best score) / (worst score) = #
If # > 1 then .... (# - 1) = %
If # < 1 then .... (1 - #) = %

Its something like an inverse relationchip. I can't remember what its called.


31.20% for DIVX is representing Intel needing to increase its numbers by 31.20% in order to match AMDS score...

we want to know how much better Intel is better then AMD...
AMD 164 seconds Intel 125 seconds (lower is better)

your calcuation 164/125 = 1.312 or 31.2% (125 seconds times 1.32 = 164 seconds)
Intel has to increase 31.2% to beat amd

actual cacluation 125/164 = .7622 ... 1 - .7622 = 23.78% better
AMD needs to increase (actually decrese the number) in performance by 23.78%

not to mention this...

FEAR
174.48 / 125.1 = 39.47%
Notice how they said AMD needs to increase performance by 40% to beat conroe?

However, our figure there is esentially saying that Conroe gets 62% (100-38) the score of the AMD... 8144 * 0.62 = 5049.

Your score is stating that Intel is performing 38% worse when in reality it would have to increase %62 percent (like 5 to 10, thats DOUBLE (100%)) to beat AMD...

AMD needs to increase " " performance in order to match conroe...

(best score) / (worst score) = #
If # > 1 then .... (# - 1) = %
If # < 1 then .... (1 - #) = %

you need to do that with EVERY test, that formula...

[kempez]

"Lies, damn lies and statistics"

This thread has gone so far off course its now navigating round the Cape of Good Hope

Whatever people say its obvious that Conroe will be an excellent chip and Intel are onto a winner at least until the end of the year or until AMD makes another very sweet chip

[DR]

Cape of Good Hope

[Holly]

"Lies, damn lies and statistics"

hmm, strange we think the same thing... In fact it is favourite saying of my statistics professor - "There are three levels of lies : lies, damn lies and statistics"

to be honest, the % expression is pretty puzzling to pretty much people... it would imho be the best if you took the best score in row and claim it 100% and other score in line make proportional to this best score... ie with ScienceMark mem bandwidth you would get:

Conroe 6700 | Conroe 6600 | FX-62 | FX-60 | Presler 965
61,97 % | 62,10% | 100% | 70,24% | 69,71%

as for the lower is better, you can drop whole maths wizardry and make following...
the slowest gimp in line takes 100% time to finish the task... with DivX test you get:

Conroe 6700 | Conroe 6600 | FX-62 | FX-60 | Presler 965
71,02% | 78,41% | 93,18% | 100% | 86,93 %

then just make the best result bold and you'll get nice well-arranged table.
It would be also interesting to see similar table with performance per watt. I am quite convinced Conroe's victory would be much more significant there, but otoh we don't know how much power 975X takes etc...

Anyway, thanks for nice review :-)

[DR]

I am thinking we are all debating too much - I am suprised with some of the crazy suggestions which have been made that someone hasn't blamed numerecy errors on the Conroe CPU.

I think the funniest thing which I have seen is the comment about "Conroes cache is optimized for XP, it will be rubbish in Vista" The irony - I don't think any company who signs to sell CPUs and Core-logic to a vendor who doesn't ship with Windows would do this. I am talking about Apple....

... anyway - competition is real important for the industry and this just spices things up!

[keta]

yea i agree david , its just people finding it hard to belive that intel is back in teh game , it wsa gnana happen sooner or later and im glad its happening now its at the perfect time for intel cuse this is a huge upgrade year for every one , most people r gnnaa be buyin new pcs this year and i no i will b

neways stop talking about this percentage stuff and get over it , its a ruff idea and the percentages seem fine to me , who cares if there wrong by 1 or 2 percent , also if u dont like the percentages just look at the numbers and see the differnces ur self.

[HunterBruno]

Well sorry but I had a discussion with Doc on the Motherboard.org and I'm asked him about the FX 62 review of him. This is what he give me as answer.
The answer is below my question. www.motherboard.org
Dear,

I have read your review about the FX 62 and find this amazing. The following link will show how the Conroe beats the AMD in many benchmarkings.

I have not even tested the Conroe yet Bruno,,,, the CPU does not even exsist yet so the reviews you are reading are BULL****!!!!

We write what is real and available, not stories.

And my name is Doc,,,,

So who can I beliiev today?

[DR]

I am more than happy to drop Doc an email

Of course you can't buy the Conroe yet - we did an early preview...

... its not my fault he can't get a Conroe CPU.

[kempez]

There's a load of Conroe chips available - most are engineering samples though. I would think Hexus get first dibs on Intels new batches

[Steve]

Hi HunterBruno, welcome to HEXUS.

the CPU does not even exsist yet so the reviews you are reading are BULL****!!!!

First up they're not reviews, they're previews. Secondly, if that's the case we've been lieing since March, and so have Anandtech... not really either site's style. It also doesn't seem like Doc to be so... angry.