Aces High Bulletin Board
General Forums => Hardware and Software => Topic started by: bloom25 on February 03, 2004, 01:16:04 AM
-
I've been reading some of the reviews of this new P4 core, and unfortunately it looks to be overall a bit slower than the Northwood 'C' type P4s are. It also seems to be a very hot (temperature wise ;) ) CPU. Even though the Prescott doesn't seem to really offer much (if anything) over Northwood, I did notice some very interesting things about it. Take this with a grain of salt, as these are only my opinions. I don't have any hard evidence to back any of these predictions up, but I figured some of you would be interested. (Besides, if I make a prediction that turns out to be right in the future I can pull this page up and say "I told you so." :D )
I was able to find a die photo (which is a picture of the actual layout of the chip itself) of Prescott, and comparing it to Northwood I've come to the conclusion (and this is my opinion only, so I may be wrong) that Prescott actually contains 2 seperate sets of execution units! In otherwords, it looks like Prescott has, in essence, two CPUs sharing some components (Level 2 cache mainly) on one die. The trace cache, the P4s equiv. of the conventional Level 1 instruction cache, also seems bigger than it should be given what Intel has published for Prescott.
Why is this interesting? Mainly because if I'm right, Intel may have 64 bit capabilities that are disabled but present in the Prescott. It looks to me that, if I'm right, they have acheived this essentially by gluing 2 slightly modified P4s together. In the past other CPUs, mainly microcontrollers, did something similar in what was called a "bit slice" design. Essentially what that means is that you have 2 32 bit CPUs, one processing the least significant 32 bits, and the other processing the most significant 32 bits. (In other words, the lower 32 bits on one core, the upper 32 bits on the other.) This would be an interesting approach for Intel to add 64 bit (like AMDs X86-64) support to the P4 design. The fact that Intel has not publically confirmed the existance of 64 bit support in Prescott, even though it is certainly disabled for now, would seem to indicate to me that it is there mainly as a sort of insurance policy against the success of AMD's Athlon 64 CPUs. More likely Intel doesn't want to admit it, because IMO, that would probably severely impact the sales of their IA-64 Itanium line in the high end server market. (Itanium definately doesn't need to be selling any slower than it already is. I won't get into the whole "Itanic" joke... ;) ) I'm wondering if Intel is concerned with the success of AMD's Opteron in the server market and wants to have the ability to enable a sort of AMD x86-64 compatibility in their future Prescott based Xeons to avoid losing face to AMD in the server market. I doubt they enable 64 bit support for the standard Pentium line in the next 6 months or so, as isn't hasn't been too long since they brushed off AMD's 64 bit Athlons as unnecessary. Enabling 64 bit support now would seem to be to be an admission of an error on their part in that regard.
Some other evidence I see of possible 64 bit support in Prescott is that its transistor count is really about double what it should be, as compared to Northwood. The fact that the pipeline was extended from 20 to 31 stages, which is one of the reasons why it is slower than Northwood - even with the doubled L2 cache, may be that some of the extra stages are necessary for proper scheduling of instructions to make best use of both cores if or when 64 bit support is enabled. (The other reason why it is slower is that even though the L2 cache is larger, it is also much slower.)
Obviously this is all speculation on my part at this point until Intel actually publically admits something to either confirm or deny the existance of 64 bit support in Prescott, and at least for now that doesn't change the fact that Prescott in its current form is somewhat disappointing to those who were hoping for better performance than Northwood.
(Interestingly, I remember a discussion I had when I was still at Oregon State that the original design of went on to become the Pentium 4 was a lot different than what ended up being released as Willamette. Willamette, and even Northwood, seemed to have some serious compromises made to the original design plans for the P4, probably to save in manufacturing costs. Prescott seems a LOT closer to what the original P4 design. If I recall correctly, the original design for what became the Pentium 4 was started as a 64 bit CPU, but that was cancelled when Itanium was released.)
-
Good post Bloom ...
I was reading somewere in the huge land of internet that Pentium was hiding/ holding onto the anser to the 64bit AMD cpu .
I read AMD is gonna release a bigger, better,and more standardized 64bit later this year ... and thats when Intel will probely bring out what there hiding .
(they gotta make the $$$ befor they sell off the Nos to compac LOL )
-
On another note Bloom, I went to THG and was a little surprised to see them revise their test setup for top of the line CPUs, it's almost as if they read what you said about the 5900 being a possible bottleneck and the problem with the low RAM latencies. :aok
The test results still show Intel in the lead for the most part but not by as much as before.
Myself, I'm waiting for Summer, I hope not in vain, to get my hands on a socket 939 A64 (if there'll be any available in the $200 range) or if that's too expensive I'll go with a Northwood P4, maybe I can get a cheap(er) 2.8/3.0 800MHz by then. :rolleyes:
-
Tom's Hardware is still handicapping the Athlon 64 systems in key office and content creation type applications. Take a look at the hard disk setup for the test:
Hard Drives (RAID 0 Stripeset)
Maxtor 6Y080M0 Serial ATA, 80 GB
80 GB Per Platter, 7200 rpm, 8 MB Cache
Disk Subsystem (AMD Platforms) Promise FastTrak S150 TX2plus (Bios: 1.00.0.30)
32 Bit PCI Add-On SATA RAID Controller
Disk Subsystem (Intel Platform) Intel FW82801ER ICH5R / South Bridge Controller
On-Board SATA RAID Controller
Notice that the Intel systems are using the onboard Serial ATA controller built into the Intel Southbridge. This controller is not limited to 133 MB/sec max theoretical transfer rate imposed by the PCI bus. That 133 MB/sec is shared with ALL PCI cards and devices on the PCI bus, so a PCI add on serial ATA controller is limited to significantly less than 133 MB/sec. Realistically, it probably has about 80MB/sec available to it. A good Raid 0 array can easily exceed this. So basically what you have is while the Intel system is using its onboard serial ATA controller, which is not limited by the PCI bus, the AMD system is using an add on PCI serial ATA controller! This doesn't make any sense, as the board they supposedly are using for the Athlon 64 already has onboard serial ATA ports capable of HARDWARE Raid 0 mode NOT LIMITED by the PCI bus. They are built into the VIA southbridge, just like the Intel system has Serial ATA ports built into the Intel southbridge. It would be one thing if the AMD system did not have onboard serial ATA ports, but since it does there is no reason to impose a bottleneck on disk I/O performance by using a software driver controlled PCI add-on Serial ATA card. Without question, this is going to impose at least a 10 - 20% drop in Disk I/O performance on the AMD Athlon 64 systems compared to the Intel platform. You can expect this to result in a corresponding drop in relative performance in disk I/O intensive applications, like video encoding, audio encoding, and file archiving. To a lesser extent it will also hurt office application performance and "content creation" applications. About the only thing that won't see a drop in performance would be gaming tests, and sure enough the Athlon 64 performs strongly in that area.
I just don't get it, surely Tom's reviewers are smart enough to realize the error they are making, so the only conclusion I can make is that they are doing it on purpose...
-
One other thing I should mention that may be VERY important to AH users is to note the huge 20% performance drop when comparing Prescott to Northwood (at the same clockspeed) in Commanche 4. This is probably the closest gaming test to AH run by much reviewers, and the new Prescott 3.2E falls behind even the "old" 2.8 'C' Northwood. If this holds true for AH, that is very significant, as it means the Prescott suffers a lot from its extended pipeline in flight sim type games. I did a quick calcuation and came to the conclusion that Prescott will need to run at 4 GHz to even match the 3.2 'C' Northwood in Commanche 4. This may hold true for AH 2 as well... In general though, Prescott is about 5% slower in most games than Northwood, which means that the 3.4 Prescott will probably perform a lot like a 3.2 'C' Northwood in most games.
To be honest, I have some questions concerning Intel's motives for Prescott. It seems to me that they are only concerned with allowing for higher clockspeeds to keep the somewhat hobbled P4 design viable until they can bring something better to market. (Tejas... Pentium 5) The whole point of Prescott seems to be that megahertz sells, and true performance is secondary. I think, however, that they were forced to delay Prescott once they realized just how much its performance suffered by extending the pipeline to 31 stages (from 20). That really required improving portions of the design to be closer to the orignal design for the Pentium 4 by implementing a real barrel shifter (which I myself questioned the motive for leaving out over 3 years ago at the release of Willamette in a post here on this BBS), implementing an integer multiplier instead of running it through the anemic x87 floating point unit, and greatly increasing the size of the cache. About they only thing that was a "new" addition that wasn't cut from the original design for the P4, is branch prediction improvements, and that came courtesy of Intel's Israel design center when they modified the Pentium 3 (P6) design to create the Pentium M. (The processor that powers their "Centrino" notebooks.) Prescott has the look and feel of a short term patch to keep in the game. If I'm right when I suspect that the Prescott actually contains a second, currently disabled, execution core illustrates the importance of marketing, and not performance, at Intel. The only reason not to enable 64 bit functionality, if it is indeed present, is to avoid the potential media backlash of appearing to simply be following AMD's lead and to avoid destroying the already fragile market for Itanium. The fact that Microsoft seems to indicate that x86-64 will be the only 64 bit x86 extension that Windows will support, basically forcing Intel to copy AMD in this regard, is also a potential embarassment.
One other thing is certain, if Intel can't reduce the power consumption of Prescott, they might run into some very serious problems maintaining stability at higher clockspeeds.
-
Have you read the xbit labs tests yet ?
http://www.xbitlabs.com/articles/cpu/display/prescott.html (http://www.xbitlabs.com/articles/cpu/display/prescott.html)
Quoted from the above link ...
"In conclusion to our discussion of Intel’s upcoming plans I can’t help mentioning the problem of 64bit extensions to IA32 architecture. Until quite recently, Intel has been denying the possibility to introduce 64bit extensions like x86-64 from AMD in its IA32 processors. However, the company’s position has become much more flexible lately: now the company’s officials say that 64bit extensions can be introduced as soon as there appears corresponding software, which will be able to use the new advantages. Keeping in mind that 64bit user version of Windows XP operation system is scheduled for the middle of this year, we dare suppose that the new Prescott core already has these 64bit extensions implemented, but they will remain deactivated until the right time comes. This way I wouldn’t deny that we might soon see Intel’s x86-64 processors in the market in the nearest future. Although, I wouldn’t also make any forecasts yet…"
My thoughts are .."Whats really going on intel ?"
-
I haven't read that review yet. I just noticed that quote from an Intel rep as a news item on Tom's Hardware, so I'm guessing I'm probably right.
-
guys, a really dumb question. when you speak of northbridge, prescott, willimette, etc . . . and i peek at intel site or the sites of online vendors, i don't see the cpu's referred to in those terms.
how does one know which is which?
-
Northbridge - Generic term for the portion of the motherboard chipset that contains the memory controller, bus to the CPU, and AGP controller. Linked to the southbridge by either the PCI bus (in the past) or a special bus (Vlink, Hypertransport, Mutol, etc)
Southbridge - Portion of the chipset containing the port controllers (keyboard, mouse, USB, etc) along with onboard sound and/or network if present. Some new Southbridges have Serial ATA support as well.
Northbridge and Southbridge are simply technical terms, not actual products.
Willamette, Northwood (A, B, C), and Prescott are Intel's internal designation for the various Pentium 4 variants. The marketing names are all various Pentium 4 varieties. (Like Pentium 4 with Hyperthreading Technology and 512kb Advanced Transfer Cache, which would be a Northwood 'C'.) Prescott will go by Pentium 4 'E' or Pentium 4 with 1 MB Advanced Transfer Cache.
Other Intel codenames for their CPUs are/were/will be Coppermine, Katami, Medocino, Klammath, Merced, McKinley, Tejas, Deschutes, P55C, etc.
Intel's chipsets also have codenames, for example your motherboard uses the "Granite Bay" chipset (E7205). Other chipset codenames were/are/will be Springdale, Canterwood, Camino, Atlanta, Seattle, Alderwood, etc.
AMD uses similar codenames for their CPUs: Thunderbird, Palamino, Throughbred, Barton, Applebred, Morgan, Spitfire, Mustang, and Thorton are all the past and present Athlon/AthlonXP/Duron core types.
Athlon 64s use the Clawhammer or Newcastle core. Sometimes this is just generically called Hammer, which covers both Sledgehammer (which was never released) and Clawhammer.
Edit: For reference your system uses a Northwood 'B' type Pentium 4. (B type Northwoods are the 533 MHz FSB parts.)
-
Tom's Hardware is still handicapping the Athlon 64 systems in key office and content creation type applications. Take a look at the hard disk setup for the test:
Can you post this to the THG boards or have you done this already? I haven't seen anything on this matter, would be a good topic for debate though! :)
-
Originally posted by Kaz
Can you post this to the THG boards or have you done this already? I haven't seen anything on this matter, would be a good topic for debate though! :)
dweeb :D
-
Originally posted by Kaz
Can you post this to the THG boards or have you done this already? I haven't seen anything on this matter, would be a good topic for debate though! :)
Read Blooms PC artitexture thread ...I punted that puppy for ya
-
I'm trying Fur I'm trying! :)
Thanks Roscoroo great reading there :aok THG is Tom's Hardware Guide http://www.tomshardware.com though and they could use a good swift kick in the *** to wake them up ;)
-
opps i musta been half asleep .... here i'll make up for it
anandtech has all the prescott reviews/ tech tests listed at the top of there page ..
http://www.anandtech.com/news/shownews.html?i=21422 (http://www.anandtech.com/news/shownews.html?i=21422)
that will link ya to about 10 sites or more :aok
-
Woah great find thanks! :aok
-
It looks like Intel confirmed yesterday that Prescott does have x86-64 (AMD compatible at that) support built in - but it will remain disabled for now. The Xeon with the Prescott derived core will have them enabled.