INTEL MULTICORE APPROACH NEEDS PSYCHOANALYSIS

Page 3 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
the link shows a sossman system for servers not for desktop cpus, and btx i think for now is more like a oem (eg dell) only idea to make it easier for em?
Click to expand...


Well, perhaps Intel will start with the BTX server market, first; but, if you care to have a look at their [website] white papers on the subject, BTX form factor IS intended to replace ATX in all fronts, asap.

The overall market (and OEMs) have been dragging ATX as far as possible, due to its well established versatility, cost & pervasiveness. However, there's yet no consensus among the relevant OEMs although the original proponent, Intel, should have the last & decisive word.

One thing's for certain: We've seen this turn before and we'll see it again. ATX is going to be replaced, sooner than later. I just don't think BTX is (will be) the consensual ultimate form factor.

In my humble opinion, any form factor platform to be, must rely upon the basic principles of thermodynamics; all else must adapt to comply with these principles and, I don't think Intel BTX is (will be) that platform...


Cheers!

Geez man stop overanalising everything, BTX IS TO COOL HOT PRESCOTTS IN DELL MACHINES, IT WONT MAKE A DIFFRENCE FOR AMDS THERE ALREADY COOL ENOUGH (AND THE REST OF THE COMPONENTS)
 
I looked at those benchmarks, and I noticed something odd. 2 Dual-Core 2GHz Sossomans (4 Processors) gets 884 CineBench 2003. the Dual-Core 3800+ is 2.0GHz and it gets 536. Let's see....a 2.0GHz Dual Dual-Core (4 Processors) is only 64% faster than the 3800+....Hmm...Looks like "Sossoman" isn't faster on a clock for clock basis than an Athlon X2, which everybody kept saying "It's faster clock for clock". And Cinebench is threaded for up to 16 Processors, so it's fully using all 4 of those. I guess AMD still has the edge against those CPU's, unless I'm missing something...which I doubt.

~~Mad Mod Mike, pimpin' the world 1 rig at a time
 
The reason why Sossaman's performance didn't double in those benchmarks when the 2nd CPU was added was because they were using a E7520 Lindenhurst chipset. That only contains a single 800MHz FSB to feed both processors. Lindenhurst was never designed to support Sossaman anyways so this is probably only a modded test setup. Sossaman is supposed to use the Bensley platform and the Blackford server chipset when it ships with it's dual FSBs. The chipset will run in dual 1066MHz mode for Dempsey and dual 667MHz mode for Sossaman.

What I found most interesting is the fact that performance increased about 70% when moving from 1 2GHz dual core Sossaman to 2 2GHz dual core Sossamans. You already mentioned that the dual dual core setup is 64% faster than the 3800+. All despite the fact that only a single 800MHz FSB was used. This flies in the face of everything people have been saying about the FSB making Conroe and family extremely bandwidth limited. If a dual processor Sossaman system can still scale 70% on a single 800MHz FSB while it's single processor variant Yonah has a 667MHz FSB, I seriously doubt a single dual core Conroe will be bandwidth limited on a 1066MHz. Woodcrest will be even less bandwidth limited on its independent 1333MHz FSBs.

It's also nice to see that dual dual core Sossaman setup producing 64% more performance than the 3800+ with only a 95.9W system load.

Edit:
I just noticed that those dual 2GHz Sossaman numbers are linearly scaled computer projections from the dual 1.5GHz Sossaman sample they had. This of course means that they should be taken with a grain of salt. However, my comments on the non-issue of FSB limitations still appear valid. The actual 1.5GHz Sossaman sample they had actually scaled better than their 2GHz projections. The dual dual core 1.5GHz Sossaman performs 73% better than a single processor dual core setup despite the single 800MHz FSB. It is the dual processor 1.5GHz Sossman setup that has a 95.9W system load.
 
The reason why Sossaman's performance didn't double in those benchmarks when the 2nd CPU was added was because they were using a E7520 Lindenhurst chipset. That only contains a single 800MHz FSB to feed both processors. Lindenhurst was never designed to support Sossaman anyways so this is probably only a modded test setup. Sossaman is supposed to use the Bensley platform and the Blackford server chipset when it ships with it's dual FSBs. The chipset will run in dual 1066MHz mode for Dempsey and dual 667MHz mode for Sossaman.

What I found most interesting is the fact that performance increased about 70% when moving from 1 2GHz dual core Sossaman to 2 2GHz dual core Sossamans. You already mentioned that the dual dual core setup is 64% faster than the 3800+. All despite the fact that only a single 800MHz FSB was used. This flies in the face of everything people have been saying about the FSB making Conroe and family extremely bandwidth limited. If a dual processor Sossaman system can still scale 70% on a single 800MHz FSB while it's single processor variant Yonah has a 667MHz FSB, I seriously doubt a single dual core Conroe will be bandwidth limited on a 1066MHz. Woodcrest will be even less bandwidth limited on its independent 1333MHz FSBs.

It's also nice to see that dual dual core Sossaman setup producing 64% more performance than the 3800+ with only a 95.9W system load.
Click to expand...

From my personal testing, it doesn't seem CineBench is a very RAM Hungry program, consuming only 67MB to 195MB in it's most intense benchmark, and I tested multiple overclocks on RAM (DDR400 to DDR500). That would lead me to think that the FSB isn't a factor, but rather RAW CPU power, which it seems the Sossaman lacks in, as far as my personal testing goes on the workstations I have used (these were high-end systems, Single CPU Athlon 64's and Dual CPU Opteron 64's). That website also stated it was running the FSB @ 600MHz x4, and there's no way 95w = 2.4GHz FSB.

~~Mad Mod Mike, pimpin' the world 1 rig at a time
 
I just missed your last post with my edit.

In any case, I think you're right about that CPU-Z image. I noticed it when I read the article but I mainly discounted it as CPU-Z misrecognization. I don't think version 1.30 has correct support for Yonah (which should read Core Duo) much less Sossaman. Part of the problem is the modded Lindenhurst chipset that I mentioned. A 600MHz FSB mode doesn't exist much less a 2400MHz bus speed. The 2.5x multiplier is obviously wrong as I thought the lowest multiplier with EIST is 8x.

In regards to power numbers, each Sossaman only has a TDP of 31W and given that they are only using integrated graphics and two sticks of RAM, 95.9W for a system load might be possible.
 
Humm?! Could you translate that, please?

Thanks.
Click to expand...

Translate to what? I only speak English. :roll:

If you mean example I can think of several...

Servers, Rendering farms, Encoding, Cad worskstations.

Disk drives are plenty fast for all of these, but they all would benefit from faster Proc's and memory.
 
pip_seeker said:
joset said:
I am not an AMD Fanboy, I simply promote the technology that is better right now.
Click to expand...

Excuse me to barge in but... you really got this funny, fan...ny way of promoting it.
Aren't you being a little biased by any chance... on your own right, of course?! :)


Cheers!

Well it looks like consumers are now talking with their dollars. So the only ones holding intel up right now are none other than the fan boys.....[/quote]

THIS.
Click to expand...
Click to expand...
 
I just missed your last post with my edit.

In any case, I think you're right about that CPU-Z image. I noticed it when I read the article but I mainly discounted it as CPU-Z misrecognization. I don't think version 1.30 has correct support for Yonah (which should read Core Duo) much less Sossaman. Part of the problem is the modded Lindenhurst chipset that I mentioned. A 600MHz FSB mode doesn't exist much less a 2400MHz bus speed. The 2.5x multiplier is obviously wrong as I thought the lowest multiplier with EIST is 8x.

In regards to power numbers, each Sossaman only has a TDP of 31W and given that they are only using integrated graphics and two sticks of RAM, 95.9W for a system load might be possible.
Click to expand...

I in no way doubt the 95w, I only doubt the 600MHz FSB which if that was somehow true, that would null the 95w.

EDIT: Maybe you can shine some light on this LT-Data, I thought Yonah had 64KB/64KB Level 1 Cache, that CPU-Z also said it was 32KB/32KB, doesn't it have 64KB?

~~Mad Mod Mike, pimpin' the world 1 rig at a time
 
wow, does anyone still do real comparisons of systems or do they just read benchmark scores? Neither of the next generation processors are released, and when they are, one site will say something, and the other will say something completly different. As it stands, HT is Great... so is HT tho... (hypertransport and hyper threading respectivly). My favorite thing is my MSI p4n board, which used Intel chips with Nforce 4 chipset, so, it has intel processor with AMD stuff includeing Hyper transport on every device, save the cpu (and it is kinda faked on the ram aswell... since AMD didnt see DDR2 as a big thing).

Which brings me to my next point, WHO CARES about the chip wars. All we need to do is get some accurite testing done (as in multiple boards, multiple chips, multiple ram configs etc...) and see which PLATFORM does best.

BTW AMD AND INTEL SUCK!!!!! THEY ARENT RISC!!!!
Click to expand...

Actually, if you knew anything about the Athlon 64, you'd know it's architecture is more closely related to that of a RISC CPU than a CISC, you must be a MAC-Fiend.

EDIT: From what I can tell, the only thing HyperTransport connects is the NB to SB in the MSI P4N, and it's only 1.6GB/s which makes it about 200MHz.

~~Mad Mod Mike, pimpin' the world 1 rig at a time
 
Yonah has a 32KB/32KB L1 cache which is the same per core as Dothan. That was doubled from the 16KB/16KB of the Pentium III along with the associativity doubling to 8-way and the line size doubling to 64-bytes. K8 of course has 64KB/64KB L1 caches, but that's because their exclusive architecture mandates as big an L1 cache as possible. The exclusive architecture that Intel uses puts more emphasis on L2 cache size instead, as well as the ratio between between L1 and L2 cache sizes. With the L2 cache size in Merom and family increasing to 4MB it's quite possible that the L1 cache will be increased to 64KB/64KB also. Intel didn't increase the L1 cache when Dothan doubled the L2 to 2MB so they've been letting the ratio slip up to now. Might be why the benefits of doubling the L2 cache hasn't shown up as well in Intel processors.
 
I would have to say that a lot of people who like the "bang for your buck" shopping style love the chip wars because it forces the manufacturers/designers to produce better products at a lower price. I personally can't say Im not biased because Im an AMD freak (I prefer AMD but run an Intel system [odd, ins't it?] ) :? .
 
Well, we're not disputing that the dual-core Pentium Ds are faster than the single-core P4s at the same clock speed at multithreaded apps. That is pretty much a given. What we are arguing about is whether an AMD Athlon 64 X2 will encode faster than a Pentium D. Dual (and more, up to a cerain point) cores are certainly the way of the future.

There are some pretty big differences between the AMD and Intel architectures, but you know what? You probably can't tell the difference between an AMD or Intel at equivalent performance levels. It is just whose chips have what performance levels at what price point that determines who "wins" in real life. I bought an AMD X2 4200+ as it was the best bang-for-the-buck for compiling, which is what my computer does frequently. A Pentium D 940 or 950 would have done about as well, but it was significantly more expensive. If I was doing a bunch of video work or encoding, there would be a Pentium D sitting in my case instead of the X2 for the same reason. The same logic held when I bought my laptop in 2002. I got a Pentium 4-M as it ran a lot cooler and faster than the K7 Athlon XPs of the day even though its NetBurst architecture was less efficient than the K7.

I buy the most speed for my dollar for what apps I run. You'd be a fool not to as well. This whole debate reminds me of the Cope vs. Skoal argument in high school...
 
I think Intel is in a temper tantrum and the reason they won't resort to On-Chip Mem Controller, is because AMD already uses it. It's kinda like the reason Intel won't touch SOI, because IBM has so many patents on SOI and Intel isn't too fond of IBM.

~~Mad Mod Mike, pimpin' the world 1 rig at a time
Click to expand...

Fer sure. I've sat in meetings with Intel folks brainstorming about new products. They can be a bit arrogant and proud, to the point of not really seeing the forest for the trees. On one hand, they are highly skilled at production - and particularly at transferring processes across sites and maintaining productivity. But they have not mastered the art of listening to the consultants they hire. That is, at least not until later when those ideas magically became theirs! Viola, SOS.
 
Sossman is as quick clock for clock as yonah - equal to an X2, its just well hidden in the fact that its with 4 cores at lower speeds and stressing the FSB which i now doubt is 600/2400 since we havnt seen 1333mhz fsb and dual fsb1066 or anything yet so it must be a misread...

Nothing really beats one or two high speed threads, not even 4 lower clocked cpus.
 
Sossman is as quick clock for clock as yonah - equal to an X2, its just well hidden in the fact that its with 4 cores at lower speeds and stressing the FSB which i now doubt is 600/2400 since we havnt seen 1333mhz fsb and dual fsb1066 or anything yet so it must be a misread...

Nothing really beats one or two high speed threads, not even 4 lower clocked cpus.
Click to expand...

No it's not equal to an X2, and those benchmarks show it. Stop promoting your Intel Fanboyism without knowing the facts.

~~Mad Mod Mike, pimpin' the world 1 rig at a time
 
you obviously havnt read or ignored the yonah benchmarks - THG compared it directly to an X2 clock for clock (2ghz vs 2ghz, both 2mb total cache) and yonah was right with AMD (and far better then the P4) - go do some reading, and take note yonah is just a slightly revised dothan with fsb667, not a (almost) whole new core like conroe will be.
 
you obviously havnt read or ignored the yonah benchmarks - THG compared it directly to an X2 clock for clock (2ghz vs 2ghz, both 2mb total cache) and yonah was right with AMD (and far better then the P4) - go do some reading, and take note yonah is just a slightly revised dothan with fsb667, not a (almost) whole new core like conroe will be.
Click to expand...

You obviously don't know that Conroe is based off of the Pentium M architecture, and clock for clock, that wouldn't make Conroe 20% faster w/o a SEVERE change in the CPU. Adding DIB is not going to be the saving grace, because the RAM is the limiting factor. I would like to see a link to these benchmarks, as I am too lazy to find it myself.

BTW: Only way they tested a 2Ghz X2 w/ 2MB Cache is if they underclocked a 4400+ or a 4800+.

~~Mad Mod Mike, pimpin' the world 1 rig at a time
 
you obviously havnt read or ignored the yonah benchmarks - THG compared it directly to an X2 clock for clock (2ghz vs 2ghz, both 2mb total cache) and yonah was right with AMD (and far better then the P4) - go do some reading, and take note yonah is just a slightly revised dothan with fsb667, not a (almost) whole new core like conroe will be.
Click to expand...

You obviously don't know that Conroe is based off of the Pentium M architecture, and clock for clock, that wouldn't make Conroe 20% faster w/o a SEVERE change in the CPU. Adding DIB is not going to be the saving grace, because the RAM is the limiting factor. I would like to see a link to these benchmarks, as I am too lazy to find it myself.

BTW: Only way they tested a 2Ghz X2 w/ 2MB Cache is if they underclocked a 4400+ or a 4800+.

~~Mad Mod Mike, pimpin' the world 1 rig at a time

I know perfectly well Conroe is based mostly from pentium M, and Yonah - minor tweaks pulled it up to an X2 (multimedia tweaks), conroe will have twice the FSB, twice the cache, 4 ipc (then the standard 3 for amd and intel etc) aswell as SSE4 and other unlisted tweaks- is that enought changes to the core?

Its rather funny intels releasing a 2.66ghz dual core model to take on AMDs 2.8ghz dual core released at the same time - thats about a 20% raw speed diffrence - makes sense doesnt it.

DAM FCUKIN AMD FANBOY

Im going to buy the quickest thing just after conroe is out (month or two) - i dont care wether it be intel or amd - get the idea?
 
you obviously havnt read or ignored the yonah benchmarks - THG compared it directly to an X2 clock for clock (2ghz vs 2ghz, both 2mb total cache) and yonah was right with AMD (and far better then the P4) - go do some reading, and take note yonah is just a slightly revised dothan with fsb667, not a (almost) whole new core like conroe will be.
Click to expand...

You obviously don't know that Conroe is based off of the Pentium M architecture, and clock for clock, that wouldn't make Conroe 20% faster w/o a SEVERE change in the CPU. Adding DIB is not going to be the saving grace, because the RAM is the limiting factor. I would like to see a link to these benchmarks, as I am too lazy to find it myself.

BTW: Only way they tested a 2Ghz X2 w/ 2MB Cache is if they underclocked a 4400+ or a 4800+.

~~Mad Mod Mike, pimpin' the world 1 rig at a time

I know perfectly well Conroe is based mostly from pentium M, and Yonah - minor tweaks pulled it up to an X2 (multimedia tweaks), conroe will have twice the FSB, twice the cache, 4 ipc (then the standard 3 for amd and intel etc) aswell as SSE4 and other unlisted tweaks- is that enought changes to the core?

Its rather funny intels releasing a 2.66ghz dual core model to take on AMDs 2.8ghz dual core released at the same time - thats about a 20% raw speed diffrence - makes sense doesnt it.

DAM FCUKIN AMD FANBOY

Im going to buy the quickest thing just after conroe is out (month or two) - i dont care wether it be intel or amd - get the idea?

I said show me the link where Tom's compared the X2's to the Yonah's, I am waiting to see this, I want to look at it.

~~Mad Mod Mike, pimpin' the world 1 rig at a time
 
http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2627
http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2648

anandtech's site seems to be down but im sure its one of em, and its anandtech - more amd bias then thg.
Click to expand...

Those benchmarks shown an X2 beating the Core Duo and Pentium M in almost all the benches. I want the benchmarks you said Tom's did comparing X2 to Core Duo where the CD beat the X2.

~~Mad Mod Mike, pimpin' the world 1 rig at a time
 
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom