Exuberant Optimism?

[ltcommander_data]

Well, Intel is introducing Quad-Channel RAM into their architecture and the early Dempsey reviews I read didn't indicate decrease performance. In fact, a dual processor 3.46GHz Dempsey system beats a dual processor 2.8GHz Opteron 280.

The other "quad" type architecture that Intel is planning on introducing is a quad 1333MHz FSB. Having 4 FSBs will remove the bandwidth limitations for 4-way systems, and will help in 8-way situations. There is speculation about the cost of such an implementation, but it should be noted that Intel is transitioning chipset production from 130nm 200mm wafers to 90nm 300mm wafers. That should provide plenty of transistor room for 4 FSB controllers, and will also cut production costs in half. PCB costs will probably be a separate issue of course, but I don't use 8-way systems so it's not my concern.

AM2 will be introducing a 1333MHz HT which I presume is the HT2.0 standard you are mentioning. However, it will only be available on the top end FX-62 model. All the other processors at launch will still be using a 1GHz HT bus.

I read that AMD plans to be able to run DDR2-667 at 3-3-3-8 timings, which is a standard value-RAM PC3200 speed, and much lower than current DDR2.

Current DDR2-667 is available at 3-3-2-8 for any system, it's just a matter going out and getting it.

One thing I didn't like of what I heard, I believe I heard right, was that increased clock speed of AMD HyperTransport doesn't increase bandwidth, but I find different.

It's not that it doesn't increase bandwidth, it's whether that bandwidth is really needed. With an onboard memory controller memory traffic no longer clogs the connection. The remaining traffic is mainly to the GPU. In most cases a 1GHz HT is sufficient. It's only when your working with ultra-high end graphics cards in a SLI or Crossfire configuration that the additional bandwidth is really necessary.

Endyen has mentioned that faster HT will reduce latency. This may be true but real world results don't reflect a noticable performance increase. Going from a 800MHz HT to a 1GHz HT only gained 0.1 fps in Quake 3, 0.02 fps in Unreal Tournament 2004, and 0.09fps in Far Cry. The synthetics were similarly unimpressive.

http://www.xbitlabs.com/articles/cpu/display/athlon64-3800_3.html

Granted these games and results are old, but their should have still been some relative increase if bandwidth or latency was such a concern.

I don't think Intel developing an OMC would outright infringe on AMD's patent. It would be how Intel's is designed and what technologies it uses. Intel remains reluctant on developing an OMC for desktop consumer and mobile chips since it reduces flexibility for differentiation in their various product lines and is difficult to update and modify. What Intel plans to do for desktop is introduce a HT equivalent. OMC will have to appear in Intel's server chips though in order to reach 4-way and higher configurations cost effectively. CSI and OMCs won't be appearing in Intel designs for a while though. Earliest would be the 45nm transition in H2 2007 or more likely in 2008.

In terms of AMD's brand new architecture, that would be K10 which has been either cancelled or delayed. The earliest K10 or its replacement will appear is 2008. In the mean time, AMD has K8L which is a modification of the current K8 architecture. K8L is scheduled for H1 2007.

 

[endyen]

So basicly what you are saying is that Intel has somehow managed to sever the corelation between speed/pipes/IPC.
Somehow, they will get a chip to run ~ as fast as thier fastest 20something stage netburst, with 14 odd stages, and higher IPC than a 12 stage chip. OOOOKay. If you, and the Inquirer bothe sayu so.
Maybe we should ask them to end world poverty and disease while they are at it.

 

[Ruby]

ltcommander_data you seem to be chock full of information, mind sharing where all of this information just happen to come from? I'm a bit curious.

 

[pat]

Yup.. bring on more bus speed and bandwidth.. we always need more to waste more.

SATA3.0GB/.. AGP8X.. PCIe16X, memory bus, HT link.. do I forget one or 26

They put high spped bus for one reason. And it is not speed. It is marketing. big number sell better.

The AMD memory controller can run at 2.8? So? memory still run at 200MHz(non OCed) Hypertransport is one of a big marketing joke. Seriously. Intel at the time has 800FSB, so HT was set a 4X(4x200=800), then intel talked about faster fsb.. HTlink went to 1000MHz. Now, intel is talking about 1066 or 1333? Stupid marketters added the directional speed of the HT bust o make it 1600MHz or 2000MHz.. And there are stupid buyer that get fooled by that!

Seriously, the speed of a CPU is not how fast does it run or its buses are, but how fast does the job get done. I don't care about number. Just bring me facts.

 

[rettihSlluB]

OK, let's go by parts:

1)

A point I found interesting from Part II, was the fact that the HT and onboard memory controller design is a disadvantage to expansion cards. This is because any main memory access requires a jump from GPU to northbridge and then to the processor memory controller, 2 hops

Still, AMD has the lead in gaming if you're talking about PCIe graphics and latency is still lower than going with a traditional northbridge solution.

2)

Since you brought up a quote you liked, I'll bring up some too.

Quote:
If Conroe appears on schedule, and it has a clock speed of at least 3.0GHz, and I believe the clock speed will be at least 10% above that for top-end parts, we think it will have quite a performance advantage over the current Athlon 64 scaled up to above 3.0GHz, if it is still available at that time.

From Part III.

It seems that they feel that clock for clock, Conroe will have "quite a performance advantage". AMD scaling above 3GHz will require their 65nm process

The FX62 will debut at a speed of 3.0GHz using AMD's new 90nm SOI process which will give it a boost 0f 25% over current 90nm processors.
Once going to 45nm with the same process, performance will increase to an astounding 40% over 90nm processors. (I think I posted a link about that a few weeks ago about AMD and IBM astounding achievement with the SOI process).

3)

According to The Inquirer's calculations, a 2.66GHz Conroe will be faster than a 3 GHz AM2 X2. Now we can call the calculation speculation, so we'll attach a large margin of error. We already know that Intel has a 3.33GHz Conroe read which is clocked 25% faster than the 2.66GHz and 11% faster than the 3GHz AM2 X2, which should be a large enough margin of error. Speculation is that AMD will be able to release the X2 at around 3.2GHz by the end of the year assuming the 65nm transition goes smoothly. Even then Intel will still maintain a clock speed advantage compared to an efficiency advantage. Since we are concerned about propaganda, it should be noted that those calculations were done by The Inquirer and were not quoted from Intel PR figures. Granted The Inquirer itself may have bias, but at least its not straight from Intel's mouth.

The sad thing about all this excitement is that Conroe will never see the light at 2.66GHz.

According to the motherboard makers, Smithfield and Presler offer clock speeds between 2.80GHz and 3.60GHz, while samples of the Conroe start from 2GHz. Intel's desktop strategy is expected to differ from its notebook strategy, in which there is a clear differentiation in the clock speeds of its low- and high-end solutions: the Yonah from 1.06GHz to 2.33GHz and the Merom starting at 2.33GHz.

http://www.digit-life.com/news.html?05/37/25

4)

AMD has plans to use Z-RAM which may be integrated into their 65nm processors later. It's still debatable whether it'll be fast enough to be used in the L2 cache however. An L3 cache implementation looks more likely

In deed, AMD is optimizing this new technology to use it as a replacement for its current L1 and L2 implementation. Also, they'll use it for a shared L3 cache in Opterons and it makes a lot of sense since AMD's design is scalable to up to 32.000 processors. This will enhance the abilities of SMP, something that Intel would kill to have.

5)

The target size is currently 5MB. This really isn't that impressive though considering Intel desktop chips already have 4MB of L2 cache, and Woodcrest will have 16MB

I really don't know what you're trying to say with this staement, but once again AMD doesn't needs huge amounts of cache like intel's offerings does because Athlons are not bandwidth starved.
Anyhow, read again because in deed it's 5MB of cache per core if AMD is going the Z-ram route.

 

[ltcommander_data]

The FX62 will debut at a speed of 3.0GHz using AMD's new 90nm SOI process which will give it a boost 0f 25% over current 90nm processors.

Well, if I am optimistic than you are too. The FX-62 is not 3GHz. It is a 2.67GHz processor. AMD wasn't planning on hitting 3GHz or higher until closer to the end of the year.

http://www.theinquirer.net/?article=28367

Secondly, maybe it's my optimism coming through, but I read that digital-life article differently. They mention Presler starting at 2.8GHz and going up to 3.6GHz for the product line. Conroe's 2GHz start will replace Presler's 2.8GHz and move up from there. It is Merom that will likely be limited at 2.33GHz at launch to keep power levels in line.

Now even if Conroe launches at 2.33GHz it won't be competing against a 3GHz X2. The worst it'll compete against is the 2.67GHz FX-62. Originally, the FX-62 would have launched 3 months ahead of Conroe, but now it seems they will launch right on top of each other. The entire AM2 platform has been delayed 3-6 months.

http://www.theinquirer.net/?article=29320

I don't think I need to reiterate the fact that Intel already has 3.33GHz Conroes available which is a full GHz faster than your 2.33GHz scenario and this is 6 months before launch.

I also wasn't aware of an astounding new 90nm SOI process from AMD. I thought AM2 was just a mature stepping of the current 90nm SOI. Unless of course you are referring to the 65nm process which isn't included with AM2 at launch.

Yeah, the Z-RAM would only really be beneficial in server requirements since desktop chips don't need that much. In any case, it's not that K8 isn't bandwidth starved that it doesn't have large L2 caches. It's inheirent in the exclusive design of the caches that K8 won't benefit as much from large L2 caches. The inclusive cache of Intel's architecture is more sensitive to cache increases. I'm not going to judge which is better since they are both valid approaches.

And for my statement that you were quoting, I was just mentioning that the cache increases offered by Z-RAM don't currently seem very large. Woodcrest can have up to 8MB per core based on the current 65nm process and will likely increase as the process matures. Z-RAM is to be integrated into AMD's 65nm process, but it's unlikely to be in the launch processors to avoid complications. The 65nm launch processors were to be near the end of 2006, but with AM2 delayed AMD may adjust their plans. In any case, this would mean Z-RAM won't show up until early 2007. By then Intel's 45nm processors would already be in production for the H2 2007 introduction, and they would have caches larger than 8MB per core. That is why I was saying that Z-RAM's 5MB per core doesn't appear as impressive in comparison. However, if compared to AMD's current 1MB per core processors then Z-RAM would be a major improvement.

This is confirmation that the size will be 5MB per core:
http://www.theinquirer.net/?article=29106

And this is a statement from AMDZone about it's viability:
http://www.amdzone.com/modules.php?op=modload&name=News&file=article&sid=4767&mode=thread&order=0&thold=0

It appears now that current Z-RAM technology is fast enough for an L2 cache, but not for high end processors. It is possible that AMD could utilize Z-RAM for a L3 cache, or that the memory technology could improve enough so that L2 cache usage is possible.

It's safe to say that if AMD supporters confirm it can't currently be used for L2 cache that Z-RAM's use in that role is questionable. It can be used for L3 cache and for low-end L2 cache. Why a Sempron would need 5MB of L2 is questionable though.

 

[Ruby]

Though some of this may be true, I find it interesting where most of this so-called "Truth" is coming from...The Inquirer... The website with the most inaccurate and false information on the web... Yet everybody here believes everything they say.

I still am waiting for an answer ltcommander_data, as to where your "What I say is going to happen" knowledge is coming from, sounds to me like you spend too much time on Wikipedia and Google to sound smart, and you're so sure "I don't think I need to reiterate the fact that Intel already has 3.33GHz Conroes available which is a full GHz faster than your 2.33GHz scenario and this is 6 months before launch." Have you been to Intel's FAB's and seen it? Have you asked Paul Otellini about it? I think not, please refrain, everybody, from the "I know blahblah" about computers when all you've done is read a website that claims this, I'm tired of hearing the hype BS of google-goers.

Everybody just talks about how much they know about CPU architecture, about how fast next-gen CPU's and architecture's will be, and I know all you've done is looked at diagrams on a website and claim to be an expert, talk like one, and it's really getting old. Don't talk about how Z-RAM and L2/L3 Caches are going to be, you don't know, stop talking like you do. I'm not saying I know, because I don't, but I'm tired of going to a THG Forum article about a question, and end up with a bunch of people blabbing on about archtectures, because you don't know wtf you're talking about, please, everybody, just STFU.

(Waits for the flames from everybody...)

 

[ltcommander_data]

Well what information are you referring to. Most of my comments included links like the articles from The Inquirer, HT comparisons from Xbit, and there were many that Bullshitter posted that I quoted from.

If you want Dempsey reviews then I'll post two.

This one isn't a fair competition vs the AMD, but I'm not comparing to AMD in this case. It just happens to have a nice diagram showing the 2 FSBs and Quad RAM channels.

http://www.2cpu.com/review.php?id=110&page=1

And this is their conclusion that addresses your concerns about quad channel performance decreases.

http://www.2cpu.com/review.php?id=110&page=8

While we're on the subject of memory, the addition of the FB-DIMM to the new Xeon line-up is an interesting one. Initially I was leary of the latency penalty of adding the buffer, but once I actually used the machine and ran the workloads my fears were gone. The bandwidth is there and the added features that FB-DIMMs bring to the table are certainly welcome in the server space. Some wondered if changing memory formats again so soon after the introduction of DDR2 was a good idea, but I think it looks like they did the right thing.

And this is about the bandwidth restrictions being improved in general.

As for Bensley as a platform: Intel has done a lot of work on the Blackford (and Greencreek) MCH. The memory bandwidth is up quite a bit compared to Irwindale, and we can see some definite advantages in real-world applications. Although still not on par with the on-die controller sported by AMD's Opteron, the additional memory bandwidth is quite an improvement over current Xeon platforms, and a much needed one.

Since its there I'll just comment on the power consumption figures at the top of that page. It's important to keep in mind that the Opteron system only has 2 cores to Dempseys 4 and 4 HTT cores. Of course, the current generation Xeons aren't noted to being power savy anyways so higher power consumption is expected.

Argh, I just noticed. I mentioned the 2.8GHz Opteron 280 in my previous post. Obviously that's in error as the 280 is a 2.4GHz dual core 1MB per core processor. Whether you condemn me or not, it wasn't deliberate. In any case the following review was the comparison between the dual processor 280 and the dual processor Dempsey. It's in German, so you'll need a translator to read the text or you can just look at their charts and graphs.

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

They look at integer and FP calculations, various applications and analyse the cache efficiency.

It should be noted that both comparisons were done in November of last year on validation platforms which meant they had to scrounge around for drivers. The platform to be released in Q1 of this year will be more polished. On the other side, AMD will likely introduce a 2.6GHz FX-60 Opteron equivalent so as always the games a foot. I wonder if the Socket F has also been delayed along with AM2.

And if you are looking for DDR2-667 RAM at 3-3-2-8 it's available from Corsair. Originally they actually sold it at 3-2-2-8 but they cut back the timings to increase yields.

 

[DuxSyagrius]

I must say you are very learned on this subject. But even one as so smart such as you woud know that AM2 is delayed because the 939 parts are still so competitive to current P4s and what intel has forcasted in the near future. The only thing holding AMD back is Dell. Such a shame.

Intel also had prescotts running at 4ghz in the lab. They didnt release them though. We can speculate as much as we want, but none of know for sure.

For all these stats everyone is pushing here the old maxim holds true: LIARS FIGURE AND FIGURES LIE! (no, I am not calling you a liar)

The bottom line is that for all of these improvements, the chips comming out this summer are not going to be sme kind of revolutionary improvement (a la K6-2 to athlon or willamette to northwood). Its just another small step in the evolutionary path. AMD is paving the way by moving to DDR-2 (I think it will be at least 1 year before we really see what DDR-2 can do-lord knows it didnt help prescott) and Intel has bought itself a new lease on life by cooling its chips down. You wanna have something to talk about? How about this: I bet the average joe schmuk wont be able to tell the difference in performance bewteen these chips come September.

I will say that if all the talk is true, then Intel wins this round. Both chips are PROJECTED to have a TDP of 95W.

 

[ltcommander_data]

After such a long winded debate I'm tired of writing too.

I couldn't help but make one last parting quip though. The X2 may have a TDP of 95W, but Conroe's projected, as you say, TDP is actually 65W. Now we all know AMD and Intel measure TDP differently. While AMD measures at full load, Intel measures at 75%. AMD's 95W full load at 75% load becomes 71.25W, so Conroe will still be cooler. Now we can count the memory controller, but it's hard to do since we don't know how much it produces on an Intel system. To be fair, maximum power ratings look to be a wash between AMD and Intel. I have no idea about idle power readings because I have no idea how agressively Intel will be adopting Yonah's power saving features onto Conroe or how AMD will be improving theirs. I hope Intel isn't too aggressive with sleep by default for instance since those introduce latency and aren't really critical to desktops.

Since, The Inquirer has been questioned I'll just post from Anandtech with the original 65W slide from IDF. As I've said, this is projected, just like the rating for AM2, so it may change.

http://www.anandtech.com/tradeshows/showdoc.aspx?i=2503&p=2

 

[DuxSyagrius]

I stand corrected. But excuse me if I fing that hard to believe. Granted conroe is a hybrid pentium M/P4 (am I right?) but basically cutting TDP in half is hard to believe! Nonetheless, I think things are getting a little even between the two. Competition is good.

 

[rettihSlluB]

Well, if I am optimistic than you are too. The FX-62 is not 3GHz. It is a 2.67GHz processor

Sorry about that since I was referring to overclocking with standard HSF.
I've read about the FX62 coming out with a speed of 2.66GHz and some other sources says that it will run at 2.8GHz. Time will tell, but I really hope AMD to release this CPU at 2.8GHz to still keep its dominance over Intel offerings.

The entire AM2 platform has been delayed 3-6 months

I don't see that as an issue since AMD's current lineup can give Intel a real pain in the @ss.
What I see here is a confident AMD that doesn't worry about Intel's "so-called" next-gen processors.

Once more, all we see here is "exuberant Optimism"