Conroe vs. X2 predictions

[BrownTown]

Meroms integer will be relatively better then its floating point, if you want to get a full explanation go to google, its somewhere out there, im to lazy to find it

Also, SSE4 wont affect performance for 1 or 2 years from now, so that wotn effect anything till 2007-2008 range in terms of real apps.

 

[Hulk]

SSE4 may have some additional routines to help currently released software.

The longer 14 stage pipe of Merom/Conroe will hurt integer performance.

I don't know why instead of arguing you guys just don't name the chips and which will be faster.

We will see who is right and who is wrong at launch.

 

[Cookie Monster]

Hmm, the conroe XE chip clocked at 3.3ghz on a 1333mhz FSB, 4mb L2 cache should be really competitve to AMDs fastest CPUs.

The new chips will deifitnately improve intel's gaming apps.

 

[Hard Ball]

Originally posted by: dmens

Your prediction is almost certainly wrong, it is likely to be the other way around; Merom will excel in ALU, while X2 at FP

Care to explain? By the way, I'm talking about real large workloads, not dinky little ones that fit in the cache.

First of all, Intel has been leading integers since the advent of the P4, and AMD has been better at FP calculations since the inception of the K7; unless there is some compelling evidence out there that shows otherwise, this trend is likely to continue.

Conroe specs are not out; and there is no concrete evidence that there is significant architectural difference between Yonah and Merom except for EM64T and 4 issue core; there are some rah rah stuff on intel website that could mean anything about anything, but provides no detailed blueprint of how merom actually works internally.

Intel is unlikely to abandon its double pumped ALUs, and AMD is actively working on newer FP instructions; and again, if you are going to claim that it will be different for the chips this fall than what's happened in the last 5 years, you better have some concrete and credible evidence.

 

[AkumaX]

We all know Conroe's gonna be a powerhouse when it's released. I hope AMD has something up their sleeves, to keep this processor war in dead heat. Competition is great!!!

 

[dmens]

Claims have been flying left and right, but for some reason my claims need solid evidence? Especially when your claims talk about meaningless trends (i.e. P4 int perf indicative of merom int perf) and dubious predictions (double-pumped ALU's in a non-replay uarch).

To justify my predictions, I believe merom will be good at FP because of its additional FU's, large cache, improved prefetch and wider issue that won't stall often. And X2 will be comparatively better at int because of lower memory latency, which is a huge factor in many int loads (but not all). There, everything I said is based on publicly available information. You don't need any detailed specs to arrive at those conclusions.

Oh FYI, I'm not privy to exact merom perf numbers because I work on a different project, so I have no foreknowledge whatsoever.

 

[PetNorth]

Some details about upcoming X2 AM2

http://www.xbitlabs.com/news/cpu/display/20060215195420.html

 

[Hard Ball]

Originally posted by: dmens
Claims have been flying left and right, but for some reason my claims need solid evidence? Especially when your claims talk about meaningless trends (i.e. P4 int perf indicative of merom int perf) and dubious predictions (double-pumped ALU's in a non-replay uarch).

To justify my predictions, I believe merom will be good at FP because of its additional FU's, large cache, improved prefetch and wider issue that won't stall often. And X2 will be comparatively better at int because of lower memory latency, which is a huge factor in many int loads (but not all). There, everything I said is based on publicly available information. You don't need any detailed specs to arrive at those conclusions.

Oh FYI, I'm not privy to exact merom perf numbers because I work on a different project, so I have no foreknowledge whatsoever.

I think many sources will disagree with most of these statements, such as here:

http://www.theinquirer.net/?article=25623

The family picks up a lot from the previous Yonah architecture, the dual core, shared L2 and low power architectures. It also picks up some of the baggage like the long in tooth FSB and stronger integer performance than floating point. Everything is new, even if it looks similar.

http://www.anandtech.com/tradeshows/showdoc.aspx?i=2504

Contrary to wild speculation, Intel's new architecture will continue to feature an Out of Order execution core; a direct descendant of the Pentium M and Pentium 4 predecessors.

There is no evidence that I can see which support your claim that Conroe will have more FP units than the X2; where did you find that info, we would be interested to know.

 

[dmens]

Where'd I find the info? By looking at the merom rtl.

stronger integer performance than floating point

OK, you cannot compare int with float on the same machine. It is pointless because the load characteristics are too different. You can only make float/float or int/int comparisons between different machines, and sometimes you can equalize the clock speed to see what happens. I said merom will do well in FP compared to X2 at the same clock.

a direct descendant of the Pentium M and Pentium 4 predecessors.

Yeah, but closer to which one? The simple fact the global pipeline length has been chopped in half indicates the former.

There is no evidence that I can see which support your claim that Conroe will have more FP units than the X2

Never said that. More units compared to yonah. Besides, comparing the number of units between K8 and merom is pointless, their scheduling mechanism is wildly different.

 

[BrownTown]

no, their scheduling is not widely differnet, but it dont really matter, functional units in CPUs accound for only a very small part of the total die area, adding mroe functional units isnt the problem, the problem is keeping the units you do have full.

 

[josh609]

It's funny how we are comparing Intels "new" technology to something that is like 2-3 years old. (AMD 64) I gotta say Conroe looks really good, but I don't think Intel will release a 3.3Ghz Conroe this year, next year is more likely. I also don't think Conroe will be better than Athlon X2 in games, not with it's on board memory controller, and direct HTT links. With DDR2 800 on the AM2, I think gaming, multimedia/encoding will be all AMD, and Intel and AMD will fight it out on normal desktop benchmarks. When AMD goes to 65nm at the end of the year, or early next we'll easily see 3-3.5Ghz. Hmm, 3.2Ghz Athlon 64 FX-64 with DDR2 800, and 2 8900XTX's in SLI would be fun huh?

 

[classy]

I don't see it for Intel. The integrated memory controller I think will continue to win out. I think you'll see a decent leap in performance with AMD going to DDR2 and they also will have a faster HTT, not to mention AMD is king now will a smaller cache, which I believe in the future will not be the case.

 

[dmens]

no, their scheduling is not widely differnet, but it dont really matter

Actually yes it is different, and it does matter, and they're not very small (take a look at the yonah die shot and look for two column like stacks). You're correct in the end goal of keeping units busy, but having a different scheduling scheme is a factor in the arrangement of units, as well as the quantity.

 

[carlosd]

I predict it will be a TIE situation in terms of performance, advantage of AMD in games, advantege of intel in video encoding, and TIE in normal apps, at least until AMD adopts 65mn, the best will be defined by the prices.

 

[Dribble]

If intel are going to release these chips running at 3.33 then the lower end ones should be great over-clockers.

 

[Maximilian]

I predict that conroe will fail and my newly purchased opteron 170 will still be a top of the line processor

 

[BrownTown]

Originally posted by: dmens

no, their scheduling is not widely differnet, but it dont really matter

Actually yes it is different, and it does matter, and they're not very small (take a look at the yonah die shot and look for two column like stacks). You're correct in the end goal of keeping units busy, but having a different scheduling scheme is a factor in the arrangement of units, as well as the quantity.

the execution units are not a considerable part of the die, maybe like 10%, i havent seen a breakdown of the Yonah die, but on Operton its less then 10%, maybe more in Yonah, but still adding another floating point units will not affect die size more then a few percent. Considering that conroe has 4MB of cache id say its functional units are at most 10% of the total die...

just looking at the conroe die shots, you can see that the L2 cache is more then 50% to begin with. Then the L1 caches will take another good chunk. Add in the control logic and FSB logic, and there really isnt much room left for execution units. Intel coulkd easily double the number of units of each type if it just droped half the cache.

 

[dmens]

If you insist on taking on looking at the last level cache, then everything else pales in comparison, not just the execution units. Regardless, adding a "few percent" to the die size will increase costs significantly.

 

[Gldm]

I think Intel is going to lose some ground in the encoding benchmarks. A lot of the P4's performance there comes from the high speed SSE units, and at lower frequencies it's not going to be quite as good. The only way I can see getting around this is with double SSE units on Conroe. I don't know if that's going to be economical given the die size. They used to have two on the PIII, so if they can pull off two per core on Conroe they'll keep the lead, but I'm not sure they will. The L1 crossbar will help negate a lot of the integrated memory controller boost during multithreaded apps. Cache coherency is the biggest issue for SMP so the closer to the registers you can sync data the more efficiency you can get. Also, going to 4 issue wide cores will likely boost things as well. I don't know if Conroe will take the lead in everything, but I suspect it will narrow the gap significantly. It depends on how much boost AMD gets from DDR2 (if any given the higher latencies). I do think gaming performance will improve quite a bit. If Conroe can get within 10% of A64 X2's clockspeed I think it will take the gaming lead, considering what overclocked Dothans are known to do.

 

[robertk2012]

LOL so how did everyone do. Looks like many of the AMD fans fell flato n there face.

 

[robertk2012]

Originally posted by: robertk2012
Conroe is expected to remove certain power constraints and probably widen thermal envelope. It wont be the same architecture but will definately not be netburst. Just think of how much less power a core duo uses than an A64 to get the same performance. Im thinking that Conroe will be a beast.

I just had to quote myself.

 

[hans007]

intel is releasing supposedly a 1333 bus 2.93 4mb cache woodcrest server chip at some point. and that would lead to a 2.93 or 3.06 ghz Extreme edition for desktop.

the one they benched at idf was a 2.66 / 1066 bus 2mb. so well there is even more peformance to be had. it'll be interesting how cheap these are, but i would assume they do not cost intel as much to build as a presler for the 2mb version (since presler has 4mb cache, the yonah i know is only a 90mm^2 die, which is smaller than a sempron)

oh well i'm hoping i can get a 2.13 ghz slowerish one, for $200 or so in 5 months.

 

[robertk2012]

2.66 conroe looks to be priced at $530 and it kills an fx-60 overclocked to 2.8.

 

[bunnyfubbles]

Looks like AMD is going to have to go back to selling us bargain CPUs to overclock like crazy. Although now that I've started spending around $150 per core ($150 for a Winch 3000+ and later $290 for a Opty 165), I'll have to see if Intel offers anything in that range - assuming it can overclock well. Looks promising, and I'm pretty excited about what is to come.

 

[Maximilian]

Funny how the tables can turn, personally i dont think AMD will ever go back to the bargain bin days of the athlon XP, theyve built up too much of a reputation over the last three years. If they have somthing up their sleeve then the CPU industry may take the shape of the graphics card industry and become more competitive and fast growing.