Kentsfield is compatible with same chipset Core 2 runs on. It means Kentsfield will be 1066MHz FSB. Q2 2007 is supposed to be Bearlake chipset with 1333MHz FSB. We probably won't see it until then. The real answer to this problem won't be solved until the next architecture, the Nehalem.
Kentsfield is compatible with same chipset Core 2 runs on. It means Kentsfield will be 1066MHz FSB. Q2 2007 is supposed to be Bearlake chipset with 1333MHz FSB. We probably won't see it until then. The real answer to this problem won't be solved until the next architecture, the Nehalem.
ltcommander_data
Distinguished
- Dec 16, 2004
- 997
- 0
- 18,980
I'm still of the opinion a 1067MHz FSB is more than enough to satisfy a dual core Conroe. Merom performs just fine on a 667MHz FSB and it does show definite performance increases over Yonah. I'd like to see a real comparison, but I'm willing to bet there would be very little benefit between a 1067MHz FSB and a 1333MHz FSB at the same clock speeds for a dual core.
I don't remember the exact setups for whether they were done at the same clock speeds and FSB speeds. I would imagine they would have used a 2.16GHz Yonah against a 2.33GHz Merom and the different multiplier would have made a direct comparison difficult. Regardless, the difference in scaling with overclocking may not be as much extra memory bandwidth through the FSB as the prefetchers responding better to the raw clock speed increase.
In any case, if memory bandwidth is the problem, the fault isn't the FSB. The memory controller still has a lot of work to do to reach full efficiency. THis is especially the case on the Bensley platform.
http://www.gamepc.com/labs/view_content.asp?id=xeon5000&page=7
Dempsey is only getting 5GB/s out of the quad channel DDR2 667 memory, which is far below even Dempsey's 1067MHz FSBs. Hopefully, northbridge revisions or BIOS updates will clear up the matter.
On a side note, Dempsey does pretty well against the Opterons which is a nice change. At least the last Netbursts aren't leaving as a total disappointment.
Yes, that is why I'm pushing so hard for a 1333Mhz FSB quad core. It's not just the 1066MHz FSB that will lead to disappointing results. It's also that Intel will have to decrease clock speeds in order to fit quad cores into their thermal profile. That means instead of say a 3.2GHz dual core Extreme Edition on a 75W TDP, we'll likely be looking at a 2.67GHz quad core on a 95W TDP (I seem to remember something like that for Kentsfield). This will mean quad core performance is constrained by FSB, while dual core performance is less than previous models due to the lower clock speeds. Certainly with a 1333MHz FSB, Cloverton could clock up 3GHz on the "performance" 120W TDP.
Have you read those literatures in detail though?? They think 2% is significant. Which is because those 2% adds up.
Charlie thinks Clovertown will be so limited that it won't be worth getting over Woodcrest. I think he's view has a value aside from the fact he's close to major industry players.
Merom does show definite advantages over Yonah, but that's because Merom is an already tremendously superior core over Yonah.
It is really your imagination, because both CPU uses the same clock speed and FSB which means same multiplier: http://www.iamxtreme.com/cpu/t7400/cpu060501_03.htm
Most noticeable differences:
SuperPI 32M:
Merom 2.16GHz vs. Yonah 2.16GHz: 2.6%
Merom 2.9GHz vs. Yonah 3.0GHz: 19.6%
3Dmark01:
Merom 2.16GHz vs. Yonah 2.16GHz:
Merom 2.9GHz vs. Yonah 3.0GHz:
3Dmark03
Merom 2.16GHz vs. Yonah 2.16GHz: 6.6%
Merom 2.9GHz vs. Yonah 3.0GHz: 8.9%
Cinebench
Merom 2.16GHz vs. Yonah 2.16GHz: 2.6% in 1CPU and 1.7% in 2CPU
Merom 2.9GHz vs. Yonah 3.0GHz: 8.7% in 1CPU and 8.0% in 2CPU
It shows greater advantages for Merom when they are both overclocked, even though Merom is running at lower clock speed than Yonah for the OC results.
Verdict?? Merom likes bandwidth more than Yonah, and benefits more. In fact it seems it likes it too much that it will be worrying for Clovertown/Kentsfield.
Now the FSB speeds aren't even 1066 yet. At 1066 the differences will be even greater.
Pure speculation, but its a good speculation. All the high performance CPUs like, Sun's Niagara, AMD's future CPUs, are going shared cache, so they are facing "cache thrashing", which will be minimized in various ways.
Think the fact that Pentium 4 Extreme Editions that gained 1066FSB gained so little over the 800FSB Extreme Edition parts is coincidence?? The reason the newer cores, which was true with Pentium 4's too benefit better from bandwidth is because they have performance potential that has yet to be gained without enough bandwidth. Pentium 4 gained significantly with more memory bandwidth: http://www.anandtech.com/showdoc.aspx?i=1834&p=3
In such a long time, we haven't seen a CPU which gains such an advantage as Core 2 does. As such, the performance will be memory bandwidth limited. Pentium 4, on that chart gained 8%(!) average from 400 to 533FSB. While Athlon XP gained only 3.2% from 333 to 400FSB(same % FSB increase would have resulted in 5.3% for Pentium 4).
http://www.tomshardware.com/2004/11/01/intel/page7.html
From here Pentium 4 Extreme Edition gained average of 4% for same 33% increase in FSB with 1.8% higher clock speed.
I meant literatures about comp sci. in general, I was reading quite a few of those cause I was looking into Itanium performance.
Sigh... I am saying there is enough evidence to say rather than cache thrashing, it will be memory bandwidth requirements that limits Core's scaling.
AlexanderTheGreat
Distinguished
- May 25, 2006
- 110
- 0
- 18,680
yes, one thing I hate to see is major cache trashing and another is being limited by the memory bandwidth. Do you not think that intel hasn't kept this in mind when they said that they will hit the market sooner more often or does 2% still doesn't snowbal thick enough? If this here is for stability, forget the other companys a second, let 'em be stable and tey'll build on it. other companys architecture isn't well thought throguh enough. It's not what some rich VP yupy said. And this has got noting to do with the OS either. Think back and think what a BIOS update can do.
*for one a fater bus will make things run smoother, and intel will go past 1333mhz... the day will come and shared L2 will be there... well if not L2, then L3 but it won't be as bad as AMD's "major cache trashing"
*in adition to that they best make that memory all chill and give it some good clocks cause the way things looking now I am not buying it...(1066) and what, what r u gona do when 1333 memory hits the market like a litle Bi that it is..? mh? those timings aren' t going to look any better till more cores come out with less nm's and the memory can then be made of cooler stacks = better clocks
*there is a way to piece this puzzle and it's through good 'ol process of reconstruction. by the time the quad cores hit the market the redesigned 965x will be beter tuned and more redefined for the channel to both open up and drop back down to go... they best know what I mean, they gota start bottom low if they have to...
for now we take all the gain they give us.... 2%
________________________________
The truth is out there
*for one a fater bus will make things run smoother, and intel will go past 1333mhz... the day will come and shared L2 will be there... well if not L2, then L3 but it won't be as bad as AMD's "major cache trashing"
*in adition to that they best make that memory all chill and give it some good clocks cause the way things looking now I am not buying it...(1066) and what, what r u gona do when 1333 memory hits the market like a litle Bi that it is..? mh? those timings aren' t going to look any better till more cores come out with less nm's and the memory can then be made of cooler stacks = better clocks
*there is a way to piece this puzzle and it's through good 'ol process of reconstruction. by the time the quad cores hit the market the redesigned 965x will be beter tuned and more redefined for the channel to both open up and drop back down to go... they best know what I mean, they gota start bottom low if they have to...
for now we take all the gain they give us.... 2%
________________________________
The truth is out there
I read a bit of the argument that was going on for the PCMark05. It seems PCMark05 is the worst benchmark to show real life performance out of all synthetic apps.
If there is one thing I notice, see how Core performs not so great on the synthetic benchmarks, but it performs really good on the real world ones like encoding, gaming, business apps and such.
ltcommander_data
Distinguished
- Dec 16, 2004
- 997
- 0
- 18,980
I actually haven't seen those benchmarks.
It's fairly obvious that Merom would benefit from a an FSB bump more than Yonah simply, because it's more powerful design needs more bandwidth to feed it. That said there is a limit. Yonah's core is only so efficient so as you increase FSB you reach a point where you have more bandwidth than the core can process. Merom would also have a point, and it's going to be at a higher FSB than Yonah since Merom's more powerful core can make use of more bandwidth. While Merom may make use of a higher FSB than Yonah, I still believe that that FSB threshold is below 1067MHz.
Those benchmarks don't really dissuade me away from the belief that Conroe is bandwidth starved with a 1067MHz FSB. In the case of SuperPI, I thought that program was used as an indicator of raw processor performance, because it's designed to be isolated from memory bandwidth. All it's doing is crunching numbers and outputing so I doubt it's reading much data over the FSB. Similarly, Cinebench spends more time crunching data than loading it.
What Cinebench is good for is looking at processor scaling. Now looking at Yonah at 2.16GHz, performance increases by 89% when going from single core to dual core. Now looking at Merom at 2.16GHz, we see performance scaling of 87% going from single core to dual core. If Merom was so bandwidth starved on a 667MHz FSB, then even in Cinebench we should notice poor performance scaling by going dual core. Even when you look at the overclocked numbers, Yonah at 3GHz increases 86% going dual core while a 2.9GHz Merom increases 85%.
I'd really like to see someone take an Extreme Edition and figure this bandwidth requirement out. Set it at 167Mhz*16, 200*13, 267*10, and 333*8 to test at 667MHz, 800Mhz, 1067MHz, and 1333Mhz FSBs at clock speeds around 2.67GHz. A test at 133*200 for a 533Mhz FSB might also be useful since ULV models will be using this configuration. This should definitively determine where the bandwidth threshold is.
It's fairly obvious that Merom would benefit from a an FSB bump more than Yonah simply, because it's more powerful design needs more bandwidth to feed it. That said there is a limit. Yonah's core is only so efficient so as you increase FSB you reach a point where you have more bandwidth than the core can process. Merom would also have a point, and it's going to be at a higher FSB than Yonah since Merom's more powerful core can make use of more bandwidth. While Merom may make use of a higher FSB than Yonah, I still believe that that FSB threshold is below 1067MHz.
Those benchmarks don't really dissuade me away from the belief that Conroe is bandwidth starved with a 1067MHz FSB. In the case of SuperPI, I thought that program was used as an indicator of raw processor performance, because it's designed to be isolated from memory bandwidth. All it's doing is crunching numbers and outputing so I doubt it's reading much data over the FSB. Similarly, Cinebench spends more time crunching data than loading it.
What Cinebench is good for is looking at processor scaling. Now looking at Yonah at 2.16GHz, performance increases by 89% when going from single core to dual core. Now looking at Merom at 2.16GHz, we see performance scaling of 87% going from single core to dual core. If Merom was so bandwidth starved on a 667MHz FSB, then even in Cinebench we should notice poor performance scaling by going dual core. Even when you look at the overclocked numbers, Yonah at 3GHz increases 86% going dual core while a 2.9GHz Merom increases 85%.
I'd really like to see someone take an Extreme Edition and figure this bandwidth requirement out. Set it at 167Mhz*16, 200*13, 267*10, and 333*8 to test at 667MHz, 800Mhz, 1067MHz, and 1333Mhz FSBs at clock speeds around 2.67GHz. A test at 133*200 for a 533Mhz FSB might also be useful since ULV models will be using this configuration. This should definitively determine where the bandwidth threshold is.
THG's own "preview" article on Conroe.... Benchmarks run with a 2.67ghz Conroe using only DDR2-667 memory against a stock and 3ghz AMD FX62 that was running DDR2-800, 933, and 1016mhz memory. http://www.tomshardware.com/2006/06/05/first_benchmarks_conroe_vs_fx-62/ Revision model is not shown.
Artmic
Distinguished
- May 27, 2002
- 311
- 0
- 18,780
Wow looking at those numbers, if AMD gets a 3.2Ghz part out and competes against a 2.93Ghz Conroe part, then the difference between the 2 is probably 1-2 percent in favor of Intel (At least until Intel gets the 3.2Ghz part out)
I wonder how the Conroe stacks up in Multi-theaded situations, will it be a hog?
I wonder what the Xeon line will look like for Intel.
maximus0206
Distinguished
- Jun 4, 2006
- 2
- 0
- 18,510
False. As this thread has been discussing, Core 2 is very limited by available memory bandwidth, and in Tom's review, was running DDR2-667, with the FX-62 at 800mhz, so to say that shows that AMD is close and only needs another 200mhz to compete is incorrect.
ethernalite
Distinguished
- May 24, 2006
- 215
- 1
- 18,680
Don't forget that that the Conroe used only DDR-667 memory, while the FX-62 was loaded with the top of the top of the line memory.
ethernalite
Distinguished
- May 24, 2006
- 215
- 1
- 18,680
While I agree at ignoring the synthetics, one has to wonder why Conroe - which has such a clear performance advantage in low resolutions - fails to impress, and even falls behind, in the high quality benchmarks. It doesn't seem like they are GPU bound, based upon the fact that the FX-62 manages to gain quite a bit from the overclocking.
I don't recall seeing such a dramatic change in relative performance in an apparantly non-GPU bound test. I suppose it could have something to do with the difference in setups that made the Intel box become GPU bound sooner, or that the treshold was just at the maximum of the FX-62 overclocked could produce while the Conroe missed out on some of its performance capacity.
Any other ideas?
I don't recall seeing such a dramatic change in relative performance in an apparantly non-GPU bound test. I suppose it could have something to do with the difference in setups that made the Intel box become GPU bound sooner, or that the treshold was just at the maximum of the FX-62 overclocked could produce while the Conroe missed out on some of its performance capacity.
Any other ideas?
AlexanderTheGreat
Distinguished
- May 25, 2006
- 110
- 0
- 18,680
ok. Take conroe. yes that same conroe that was in that test some time long back with those same germans. Place it in a beter board with beter capabilities (AMD had an advantage why not us?). Get the best memory, 1066 for all I care. A beter GPU. c'mon, give us an advantage, an ATI 3200 will do. Overclock everything. the FSB, the cache and the Memory and the Video card to the MAX. Conroe will win all of the benchmarks that Fx62 had beaten. For all I care it isn't vs. Core 2 Extreme.
Results:
CONROE > AMD FX62
yes! Conroe is a beter choice.
__________________________________________
The truth is out there
Results:
CONROE > AMD FX62
yes! Conroe is a beter choice.
__________________________________________
The truth is out there
Action_Man
Splendid
- Jan 7, 2004
- 3,857
- 0
- 22,780
Action_Man
Splendid
- Jan 7, 2004
- 3,857
- 0
- 22,780
Action_Man
Splendid
- Jan 7, 2004
- 3,857
- 0
- 22,780
AlexanderTheGreat
Distinguished
- May 25, 2006
- 110
- 0
- 18,680
Artmic
Distinguished
- May 27, 2002
- 311
- 0
- 18,780
Wow looking at those numbers, if AMD gets a 3.2Ghz part out and competes against a 2.93Ghz Conroe part, then the difference between the 2 is probably 1-2 percent in favor of Intel (At least until Intel gets the 3.2Ghz part out)
I wonder how the Conroe stacks up in Multi-theaded situations, will it be a hog?
I wonder what the Xeon line will look like for Intel.
Nope, won't happen at 3.2 GHz... Let's summarize the data a little differently.... Take only the CPU bound, low res gaming benchmarks, the audio and video (a little bit pressed for time so not competely all of them), throw out the synthetics, they are usually do not give a concise picture and are overly weighted in my opinion to Intel.
The average advantage of Conroe over FX-62 with decked out memory is 20%, over the over clocked it is about 10%, so for each 200 MHz (assuming linear scaling) FX-62 closes by 10%.
Based on this data a 3.2 GHz will roughly equal the performance of a 2.67 GHz Core 2 Duo on average, the 2.93 will be another roughly 10% on top of that Core 2 result, so to achieve parity they will need to release a stock 3.4 GHz part by July -- not gonna happen -- maybe with 65 nm, but Intel will have 3.2 GHz by then. The FEAR bench is likely more GPU throttled than anything else so these numbers are likely slightly skewed in AMD's favor in regard to raw CPU performance.
Multithreaded tests are mixed, ExtremeTechs multithreaded report shows good numbers, Bit-Tech preview data shows mediocre number. The jury is still out on this one... I believe AMD will have more of a chance in multi-threaded/tasking environments.
Great rundown, so it seems Conroe will hold its own for the foreseeable future.
I haven't purchased a new system in a long time, and I’m aching to do so, but i want to make my next purchase last, i was hoping to get my hands on a 3.3Ghz Conroe, but i guess that won't be, until they release the 3.2Ghz part in Q4.
I do a lot of multitasking on my current computer, and if benchmarks show that AMD’s offering is 10 percent better at multithreading I’d get an AMD processor though, so I’m crossing my fingers Conroe is not some botched up hack that lacks dual-core kick.
I think that some of those gaming benchmarks are gpu limited due to the fact the FX has sli and conroe is only 1 ati card. If we kick out the 3 outliers which have lik 14% decrease from the average you get that the amd needs a 3.3 or more to match conroe..
Artmic
Distinguished
- May 27, 2002
- 311
- 0
- 18,780
Hmmm, someone should start a woodcrest thread, i figure if i can get a faster woodcrest than the desktop version of it, then i migth as well go for that... I just want to get a 3.2 - 3.3Ghz cpu, my current crap is a P4 2.2Ghz, served me well, but i got so many things running on this PC, that it is becoming a freaking punishment to work on it.
- Apr 10, 2006
- 428
- 0
- 18,780
New June 6, 2006: The independent reviews are starting to hit the web. Here Anandtech.com got their hands on a Core 2 Extreme and put it up against their FX-62 in their hotel room. To keep the tests as fair as possible, they used the same RAM, graphics card, harddrive, etc... They also ran them both at their stock shipping speeds. If any of the evidence below in this thread is valid, when overclocked, the dominance of Core 2 will only extend further.
They also ran their own benchmarks and were not guided by Intel in any way.
Link: http://anandtech.com/cpuchipsets/showdoc.aspx?i=2771&p=1
And here is a quick analysis of their data with additional data for the older Intel flagship processors thrown in for comparison. Needless to say, Core 2 Extreme wins across the board:
They also ran their own benchmarks and were not guided by Intel in any way.
Link: http://anandtech.com/cpuchipsets/showdoc.aspx?i=2771&p=1
And here is a quick analysis of their data with additional data for the older Intel flagship processors thrown in for comparison. Needless to say, Core 2 Extreme wins across the board:
306maxi
Distinguished
- Feb 7, 2006
- 679
- 0
- 18,980
Make sure you highlight that INDPENDENT part of the data
Fireing squad ran some of their own INDEPENDENT bench marks:
Being across the world in Taipei at Computex presents many interesting and unique opportunities. One of these was our chance to sit down with an Intel Core2 Duo system - with no Intel representatives around. We had several unsupervised hours to fiddle with the system, running artificial and real-world benchmarks. To top it off, we also tested against an AMD Athlon 64 FX-62. Keep in mind that the Core 2 Duo Conroe system is running at 2.4Ghz vs. the FX-62 at 2.8Ghz. Both are equipped with ATI's X1900XTX graphics card, both running Catalyst 6.5 drivers. Finally, we managed to overclock the Core2, though we ran out of time before we could do more benchmarks.
http://www.firingsquad.com/hardware/intel_core_conroe_benchmarks/
Shakira would have you think that Anand is on Intel's payroll. What rubbish considering Anand has been reccomending AMD products for aaaaages.
- Apr 10, 2006
- 428
- 0
- 18,780
Updated the chart data slightly to fix two of my errors, doesn't really change much of anything, but I wanted to fix it.
Updated the initial post to only contain the corrected image.
Updated the initial post to only contain the corrected image.
TRENDING THREADS
-
-
-
-
-
-
Question Origin web server does not have a valid SSL certificate
- Started by cool09
- Replies: 1
Facebook
Twitter
Instagram