Conroe (intel) vs AMD SERIOUS ONLY

Page 4 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Lovely, I wonder what those ??? next to the prices mean?
Click to expand...
The brackets next to the prices give time of availability and we don't have a firm launch date yet. All that is known is that it will be sometime in Q3, probably as early as possible since Intel is ahead of schedule on them.

As RichPLS already said, AM2 will be ahead in launch.
 
I can't even imagine what AMD will do to fight Conroe. I mean, Conroe has 2 major advantages: being based on something more efficent than K8, and 65nm. AMD processors really top out at 3ghz, how much more can AMD really scale up with 90nm? As much as I'd like AMD to win the performance battles, I'm not so sure it will since Intel seems to have everything planned out and ready this time, and not with a Prescott'ish release.
 
Interesting tests ... BUT they were comparing unreleased Intel quad core CPU's against the "ancient" AMD Opteron 246 SledgeHammer 800MHz HT Socket 940 Processor!! Apples and Oranges here? 😀 BTW, there WERE a few benchies where the ancient Opteron "stuck it to" Intel's latest greatest 😀
 
Ya know what AMD64s are still 25-70% faster in 64bit benchmarks so even if the Conroe really is 20% faster in 32bit mode BIG DEAL -- AMD64s will still be 5-50% faster in 64bit mode than Conroe in 32bit mode.
Click to expand...
It really wouldn't be fair to compare K8 running on 64-bit mode to a Conroe on 32-bit mode. I have no idea how Conroe will do in 64-bit mode so I'm not really going to comment on it. We can only hope that Intel decides to do it right this time since they have an opportunity to design and integrate it from the ground up.

2 Xeons = 1/2 bandwidth

4 Xeons = 1/4 bandwidth

8 Xeons = 1/8 bandwidth ( theoretical )

16 Xeons = 1/16 bandwidth ( theoretical )

32 Xeons = 1/32 bandwidth ( theoretical )
Click to expand...
Now that isn't exactly correct. Intel is introducing dual and quad FSBs so both 2 and 4-way systems will operate with full bandwidth with 1333MHz per chip. Going up to 8-way or when quad cores are introduced, Intel will probably be able to squeeze by since the Yonah architecture that the Core family is based on doesn't seem as memory bandwidth hungry as Netburst, and the larger caches will help too. It's not the most elegent solution by any means, but if it works out in the end thats really all the matters. Now for 16 and 32-way systems, Intel is hopeless.

Opteron 940 vs. P4 Xeons

Currently 2 x 940 CPUs with 1 OMC each and 4 stix of PC3200 have 2 TIMES the memory bandwidth = 12.8GB/s

4 x 940 CPUs with 1 OMC each and 8 stix of PC3200 have 4 TIMES the memory bandwidth = 25.6GB/s

8 x 940 CPUs with 1 OMC each and 16 stix of PC3200 have 8 TIMES the memory bandwidth = 51.2GB/s

In the P4 Xeon line memory bandwidth is

1/2 with 2 CPUs
1/4 with 4 CPUs
It doesn't scale at all!

With socket 1207 AMD will be able to easily add Quad Channel RAM or Dual Memory Controllers for Dual-Core CPU's so:

You have 1 PHY CPU
2 cores
2 on board memory controllers
2xDual Channel RAM -- 2 stix / Core -- 4 stix total for TWICE the memory bandwidth!

In a 2way Dual Core SMP Opteron 1207 you would QUADRUPLE your memory bandwidth = 51.2GB/s MAX Theoretical Memory Bandwidth

In a 4way Dual Core SMP Opteron 1207 you would have 8 TIMES the bandwidth = 102.4GB/s MAX Theoretical Memory Bandwidth

in an 8way Dual Core SMP Opteron 1207 you would have 16 TIMES the bandwidth = 204.8GB/s MAX Theoretical Memory Bandwidth

Suppose Intel does release the NEW Xeons with a 1333MHz bus, how long before they actually hit the market???? 1Q 2007? 1Q 2008? Will they have anything close to 204.8GB/s Memory Bandwidth?

Will they do 64bit??????? Why didn't Intel demonstrate 64bit performance @ IDF???????

The Conroe is soooooooooooooooooo much better it must be able to do 64bit very well right???

Hmmmmmm maybe not.
 
Interesting tests ... BUT they were comparing unreleased Intel quad core CPU's against the "ancient" AMD Opteron 246 SledgeHammer 800MHz HT Socket 940 Processor!! Apples and Oranges here? 😀 BTW, there WERE a few benchies where the ancient Opteron "stuck it to" Intel's latest greatest 😀
Click to expand...
Those weren't quad core CPUs, just hyperthreaded dual cores. However, it was just BS to include a dual CPU 246 in there since thats 2 cores versus 4, and in server benchmarks it's obvious who should win.
 
In case anyone cares, here is a direct comparison between the 3.46GHz Dempsey and the Opteron 280.

http://www.tecchannel.de/server/hardware/432957/

The results are quite good for Dempsey, and show that it is ahead most of the time. The benchmarks themselves are pretty comprehensive and include comparisons in 64-bit Linux as well as charts on cache efficiency. In many tests were the 3.46GHz Dempsey was ahead, it was also downclocked to 3.2GHz to see how much that changes the results. Tthe performance comes as a price though as it's already been pointed out that the power consumption is appalling.

Of course, AMD has already released a faster 2.6GHz Opteron 285 while Dempsey hasn't even been released yet. Intel blaims this on the lack of availability of FD-DIMMs and that the platform is perfectly ready, although it's really their fault for the RAM choice. Intel will also be releasing a 3.73GHz Dempsey after the 3.73GHz 965EE launches to better compete with the 285 and Socket F.

http://www.theinquirer.net/?article=29576
 
They don't have the 930 listed on there for Q3 this year. You think they will phase it out or perhaps, they just weren't aware of the price at time of print?
 
In case anyone cares, here is a direct comparison between the 3.46GHz Dempsey and the Opteron 280.

http://www.tecchannel.de/server/hardware/432957/

The results are quite good for Dempsey, and show that it is ahead most of the time. The benchmarks themselves are pretty comprehensive and include comparisons in 64-bit Linux as well as charts on cache efficiency. In many tests were the 3.46GHz Dempsey was ahead, it was also downclocked to 3.2GHz to see how much that changes the results. Tthe performance comes as a price though as it's already been pointed out that the power consumption is appalling.

Of course, AMD has already released a faster 2.6GHz Opteron 285 while Dempsey hasn't even been released yet. Intel blaims this on the lack of availability of FD-DIMMs and that the platform is perfectly ready, although it's really their fault for the RAM choice. Intel will also be releasing a 3.73GHz Dempsey after the 3.73GHz 965EE launches to better compete with the 285 and Socket F.

http://www.theinquirer.net/?article=29576
Click to expand...


ah ja ein netter Artikel in der Tat jedoch schlägt das Opteron noch das Intel zur Hälfte ungefähr Test

:-D
 
Whatever it was, there has been a change in plans. It's already been reported that the 920D won't be receiving the C-1 stepping update, and now Intel has announced to developers that the 920D will be phased out beginning April 24th, which is one day after those price cuts goes into effect. That's probably why the 930D and 920D are listed at the same price on the site for April 23rd.

http://developer.intel.com/design/pcn/Processors/D0105972.pdf

As to what will happen when Conroe ships, the 930D may well be phased out too. We'll probably see a 905D type processor to replace the 805D and clear out the final Presler inventory. It could just be the 920D being reintroduced as Digitimes suggests or it could be a 2.8GHz dual core with a 533MHz FSB in order to encompass defective chips.
 
Whatever it was, there has been a change in plans. It's already been reported that the 920D won't be receiving the C-1 stepping update, and now Intel has announced to developers that the 920D will be phased out beginning April 24th, which is one day after those price cuts goes into effect. That's probably why the 930D and 920D are listed at the same price on the site for April 23rd.

http://developer.intel.com/design/pcn/Processors/D0105972.pdf

As to what will happen when Conroe ships, the 930D may well be phased out too. We'll probably see a 905D type processor to replace the 805D and clear out the final Presler inventory. It could just be the 920D being reintroduced as Digitimes suggests or it could be a 2.8GHz dual core with a 533MHz FSB in order to encompass defective chips.
Click to expand...
Pentium D is the new Celeron ! =D
Overclocking with a $300 budget !
 
Like I said in my last paragraph, I did note that the unreleased 3.46GHz Dempsey was compared to the Opteron 280 which has already been replaced with the 285. As to when Dempsey will be released, I really can't say. It was previewed again at IDF and it would have to be released before Socket F to avoid being extremely outclassed. The motherboards and everything is all ready, but they're just waiting on sufficient RAM availability.

As for memory bandwidth, I'm aware that Intel's architecture doesn't scale. I'm just saying that they seem to have enough of it to satisfy their needs at least in the 2 and 4-way segments. The memory bandwidth they have will be enough to satisfy the FSBs. Having more memory bandwidth is obviously good, but it really depends on how much you really need. You can of course question the dual core Conroe benchmarks, but it seems to do quite well on it's 1066MHz FSB. Woodcrest will be even wider at 1333MHz with independant FSBs, so it really shouldn't be a problem. Now when quad cores are released, you're right that the architecture doesn't scale and so memory bandwidth will be cut in half. I can't really tell how much that will affect performance, but Yonah already runs on a 667MHz FSB and the quad cores have at least twice the L2 cache of Yonah so the reduction is buffered somewhat. Again, it isn't elegent, but we'll have to see the final results to know if it works out in the end.

As to when Woodcrest ships, it will be in H2 2006. Originally, Merom and Conroe would ship Q3 and Woodcrest Q4. Woodcrest doesn't appear behind Merom or Conroe, and the reason for the delay is to mainly avoid overlapping and canabolizing Dempsey sales and to avoid mixed messages and conflicting advertisement space with too many Core platform launches. Now with Merom switched to Q4, it's likely Woodcrest will be pulled forward to Q3. This would be better for Intel in any case, since the server market is their weakest sector. A Q3 launch is also given more credance with Intel now saying that they are not concerned with the proximity of Woodcrest and Dempsey and that they will be marketing them differently.

The 4-way Core replacement won't come until H1 2007 though, which is to be expected since Intel always is slow to update it.
 
Ok that's great so the new intel Xeon replacement runs a bit faster and beats the current Opteron 280 in a FEW tests but gets beaten by the 280 and the 254 in other tests.

Let's see it go up against the 285, 290, 295 and the new Opteron Socket F 😀

I could question their benchmarking procedures but I won't bother -- I have better things to do.

And what is the ETA on this new beast? 1 year? 2 years?
 
Funny the Single Core 254 beats the new Dual Core Intel in more tests than the Dual Core 280.
Click to expand...
It's hard to tell whether that should be something to gloat about. In any case, you have to take into consideration the circumstances of those results. For example when the 254 beats everything in the first charts of SPECfp_base2000 and SPECint_base2000 thats because those were run in a single-threaded environment. In such a circumstance, obviously higher clocked speeds would be better which is why the 254 and the 3.6GHz single core Xeon can outperform their lower clocked dual core cousins. Similarly, when SPEC_CPU2000 is run in multithread mode then the dual cores should obviously see benefits.

The 254 also wins in the first 64-bit Linux because it's single-threaded. When 8 threads are used, Dempsey's advantage becomes more clear.
 
Funny the Single Core 254 beats the new Dual Core Intel in more tests than the Dual Core 280.
Click to expand...
It's hard to tell whether that should be something to gloat about. In any case, you have to take into consideration the circumstances of those results. For example when the 254 beats everything in the first charts of SPECfp_base2000 and SPECint_base2000 thats because those were run in a single-threaded environment. In such a circumstance, obviously higher clocked speeds would be better which is why the 254 and the 3.6GHz single core Xeon can outperform their lower clocked dual core cousins. Similarly, when SPEC_CPU2000 is run in multithread mode then the dual cores should obviously see benefits.

The 254 also wins in the first 64-bit Linux because it's single-threaded. When 8 threads are used, Dempsey's advantage becomes more clear.

This means that AMD can simply increase the core clock on the Dual core CPUs and easily match or beat the new Intel CPUs.

I can't wait to see the 290 and 295 and Socket F :-D
 
And, for those who liked the Conroe preview, we have a preview of two Intel chips destroying an opteron
Click to expand...


Oh my gosh..... in the name of all that is holly...... what kind of benchmarks are these????????


Where are the dual-core Opterons?!?

I'm sure I will catch a ton of heat about including a pair of Opteron 246s in this comparison. The truth of the matter is... They are the only Opterons I have to test with. I have been in touch with AMD recently to plead my case and they have promised to rectify this situation very soon. I will definitely rerun all the numbers (and add a few more, I'm sure) if/when that happens.
Click to expand...


The author of the article admits this is not a fair test??? What are you trying to say here?

How would Intel like it if I compared an Opteron 285 against a Pentium I @ 60MHz without any cache??????

PLEASE!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

thats two P4s with DIB not woodcrest so power consumption is bad
 
Funny the Single Core 254 beats the new Dual Core Intel in more tests than the Dual Core 280.
Click to expand...
It's hard to tell whether that should be something to gloat about. In any case, you have to take into consideration the circumstances of those results. For example when the 254 beats everything in the first charts of SPECfp_base2000 and SPECint_base2000 thats because those were run in a single-threaded environment. In such a circumstance, obviously higher clocked speeds would be better which is why the 254 and the 3.6GHz single core Xeon can outperform their lower clocked dual core cousins. Similarly, when SPEC_CPU2000 is run in multithread mode then the dual cores should obviously see benefits.

The 254 also wins in the first 64-bit Linux because it's single-threaded. When 8 threads are used, Dempsey's advantage becomes more clear.

Let's analyze this Mr. TDP:

Those Xeon's use FB-DIMM DDR2 533 w/ 1066MHz FSB. The FSB don't matter, that is irrelevant w/ DDR2 533, but that 533 is the main point, that's 8,528 MB/s Dual Channel compared to 6.4GB/s on the Opteron 64 using DDR400. And FB-DIMM's But wait...FB-DIMM DDR2 can be scaled perfectly to use Quad Channel and Six- Channel which is 17,056 MB/s and 23,456MB/s, using FB-DIMM DDR2533. Intel plans to use Quad Channel, let's look at some numbers:

800MHz FSB: 6.4GB/s
1066MHz FSB: 8.5GB/s
1333MHz FSB: 10.4GB/s <=Approximation

Well...as we can see, even a 1333MHz FSB isn't quite to par w/ their Quad Channel ability and Six Channel, well, let's not go there. Let's see the Dual Independent Bus (DIB) plan for Intel:

Dual 1333MHz: 20.8GB/s

That is if they would implement Quad Channel and have 2 Channels per FSB...and that would give 2 seperate CPU's 10GB/s Bandwidth each, but each Dempsey is Dual Core and now each Core gets 5GB/s Bandwidth w/ each CPU having a 1333MHz FSB. But what about just Dual Channel? 8.5GB/s total, would they give each FSB a Single Channel? But think about the logistics, is it possible to seperate the RAM channels going to each FSB w/ 1 Memory Controller? Maybe, Maybe not, I don't know, but we'll find out. What if you can't? That means only 1 CPU can use their FSB at a time for Memory...but doesn't that create another bottleneck? Hmm...interesting...something to ponder...and what about DDR2-800 FB-DIMM's? :O! 1333MHz just aint enough...

Let's look at the Opteron 64 Socket F using FB-DIMM DDR2:

According to AMD, all A64's and O64's access memory @ CPU Frequency and have a direct link back @ RAM speed (hence Direct Connect Architecture), so let's see:

Opteron 64 FB-DIMM DDR2-533 Bandwidth:

Dual Channel: Full 8.5GB/s per CPU w/ independent DIMM's
Quad Channel: Full 17.56GB/s per CPU w/ independent DIMM's
Six Channel: Full 23.4GB/s per CPU w/ independent DIMM's

WoW, from that, it would seem that the Opteron 64 Socket F is better suited for the future, let's take a peek at Six-Channel FB-DIMM DDR2 but this time, let's throw in DDR2-800:

Six Channel:: Full 76,800MB/s per CPU w/ independent DIMM's

Holy smokes, w/ just 2 way systems, you're looking at over 150GB/s Memory Bandwidth for Socket F Opteron 64's...man that's pretty good. Let's look hypothetical here...for an extremely High-End Server w/ FB-DIMM's and 1207 pins, AMD could easily put on Multiple Memory Controllers per CPU since FB-DIMM's only require 69 pins from the 240 on DDR2 to actually go to the CPU..uh oh..can you say hundreds of GB/s....:O TB/s! o darn, things don't look good for Xeon's. But what about the Xeon's? I think you're the nice person to say "Intel plans Quad Independent Bus's", but how come Intel isn't touting that like a monkey waiting for a banana? What about the problems I noted w/ using the DIB alone? Hmm...anyway you look at it, the Opteron 64 CPU's are looking like the better option for the future, but since everybody is saying it, so will I: Only time will tell.

If you think something I said is wrong, make sure to reply and let me know, but let's keep it civilized, I enjoyed my 24 hour vacation but that doesn't mean I want another one.

~~Mad Mod Mike, pimpin' the world 1 rig at a time
 
I never doubted that Opteron could scale better, I'm just wondering how much memory bandwidth does a chip really need. Afterall, a CPU can only process so many things at a time no matter how much information you thrust down it's throat. As it stands, with dual 1333MHz FSBs and quad channel DDR2 667 memory each of Woodcrest's cores will have 5.33GB/s of memory bandwidth as you mentioned. That's already 1.06GB/s more than Conroe, whose performance good, bad or invalid has been debated endlessly, and is also twice as much as what Yonah gets per core. It's probably safe to say that Woodcrest has sufficient memory bandwidth.

The situation would be similar in the 4-way sector where quad 1333MHz FSBs would be used. I'm not sure what type of memory Intel would choose, but a 6-channel DDR2 800 implementation would supply up to 38.4GB/s or 4.8GB/s which is slightly less than Woodcrest but still 0.53GB/s per core more than Conroe. In any case, Intel won't be updating the 4-way sector until 2007, which gives the advantage to AMD of course. Intel is also scheduled to introduce DDR3 in 2007, so it's possible that Intel will go straight to DDR3 and higher frequencies to ensure that the FSBs are filled. This is obviously speculation on my part.

The concern is of course when quad cores are released. Bandwidth would then be cut in half with 2.67GB/s per core. Performance per core would obviously be reduced, but it would be interesting to see how much. As I've pointed out, that's the same bandwidth as Yonah recieves per core right now. Cloverton will probably have the advantage of more cache though. Not very elegent or preferable by any means, but we'll just have to wait and see if performance is truly crippled.

Six Channel:: Full 76,800MB/s per CPU w/ independent DIMM's

Holy smokes, w/ just 2 way systems, you're looking at over 150GB/s Memory Bandwidth for Socket F Opteron 64's...man that's pretty good. Let's look hypothetical here...for an extremely High-End Server w/ FB-DIMM's and 1207 pins, AMD could easily put on Multiple Memory Controllers per CPU since FB-DIMM's only require 69 pins from the 240 on DDR2 to actually go to the CPU..uh oh..can you say hundreds of GB/s
Click to expand...
The 76.8GB/s number you quote with DDR2 800 (6.4GB/s) is probably reached using 2 memory controllers (1 per core) each with 6 channelsfor 38.4GB/s per core. You mention multiple memory controllers per CPU, but I'm not sure if they could really fit more than 2 controllers each with 6-channels on the 1207-pin Socket F. I may be wrong, but with 12-channels split through 2 controllers, that means 69x12 or 828 pins out of 1207. Adding another 6-channel controller would put them over. Even a 4 channel quad controller setup (1 per core for a quad core chip) would yield 69x4x4 or 1104 pins. A seriously doubt the rest of a quad core chip would only need 103 pins. Of course, the 76.8GB/s of the the 2 memory controller each 6 channel setup is already a lot more than Intel would have. Again, this goes back to how much you really need? Personally, I don't know.

I know most of what I've said is just a rehash of what you've already said, but I've kind of just been mulling things over around my head. Really, I don't see Intel's FSB approach being much of a limitation right now in the 2 and 4-way segments since the dual cores should have sufficient bandwidth, a lot more than what they are used to now in any case. Things will get complicated when quad cores arrive, but Intel may still be able to keep things together by throwing cache at the problem like they usually do. However, I've admitted it before, and I'll admit it again, the 8-way segment looks extremely shady and higher segments are downright impossible. Certainly Intel can't compete with a 32-way Opteron with their current FSB implementation.

The only additional comment I can add is that the initial Socket F Opterons appear to be using regular DDR2 800. So whether FB-DIMMs will introduced with K8L next year or DDR3 will or both, I guess like you said we'll have to see.

BTW, its nice to see that you weren't permanently banned and it's nice to have you back.
 
The 76GB/s I got was from DDR2-800 using 1 Memory Controller per CPU in a Dual CPU Environment (You have to remember, in Dual Opteron 64's, each CPU has it's own memory bank on some boards, and that means 2 mem controllers total). When it comes to extreme systems, the more bandwidth, the better. Yes, it is possible to flood the crap out of a CPU, but there wouldn't be 76GB/s going straight to 1 CPU, but rather 2 PHY chips and 2 COR per PHY chip.

I see what you were saying about Multiple Memory Controllers, but check out this theory:

AMD, with 1207 pins, has the oppurtunity to do a few things,

1: Give Dual-Core Opterons 2 Memory Controllers and each Memory Controllers has Dual-Channel Memory, giving it an effective Quad Channel Memory to 2 PHY chips, thus, automatically having 8-Channel Memory w/ only 4 Dual-Channel Memory Controllers.

2) Give Quad-Channel Memory to each CPU in Dual-Core PHY chips and thus, again having 8-Channel but w/ only 2 Memory Controllers total, 1 per PHY chip.

The Opteron 64 1207 Socket F is an extremely scalable and modifiable platform, and that is the main point I am trying to get across.

BTW, thanks for welcoming me back, I enjoyed my 24-hour vacation, but I hope I never have to have one again.

~~Mad Mod Mike, pimpin' the world 1 rig at a time
 
A tualatin P3 (1266/133/512k) took out a AMD Thunderbird 1400 (1400/266/256k) and a P4 Wilamette (1700/400/256k) - ram bandwidth and fsb speeds arnt everything, we just dont know HOW efficent things can get and conroe has shown that with the old P4 fsb, and AMD with DDR400.

Intel chips in terms of dual core and quad core (with the help of DIB) might JUST get away without any huge performance hit and take on the Opterons but ram and fsb will be at its knees, we are yet to see how efficent the FSB really is yet.
 
😀

This was an Intel controlled and administered test and just like any AMD controlled and administered test should BE IGNORED without a fair and properly administered test by an independent and objective 3rd party!!!

That's right folks I do not trust AMD's or Intel's own benchmarks!


The Fact is Intel chose the motherboard, CPU and drivers for a REASON.

It was an Intel controlled and administered test so it was automatically RIGGED.

I reject their results until the product actually launches and we can test actual PRODUCTION hardware.

I am just as skeptical about AMD benchmarks - so please do not try to challenge my statement.

Intel's marketing dept. routinely selectively publishes benchmarks in which their products beat AMD by a wide margin.

These benchmarks should be taken with a mountain of SALT -- most of the time they are RIGGED, FAKE, unreliable and highly questionable.

I want to see REAL, fair and properly administered tests by multiple independent and objective 3rd parties!!

Thank YOU

Live long and prosper 😀
Click to expand...

Hey wait

Anandtech did a test and Conroe did in fact beat the FX processor

i'm neither an intel fanbpy nor an amd one, i'm just a performance fanboy

so if intel tops amd then this is what we all have to accept
 
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom