Aces High Bulletin Board
General Forums => Hardware and Software => Topic started by: SKurj on April 25, 2002, 11:06:11 AM
-
Oy
Gonna buy a new CPU probably early MAY, I am looking at the XP 2000+
Is there anything I should know before purchasin?
I am using a non-DDR mobo.
Intel fans stay out +) I cannot afford mobo upgrade at the time.
SKurj
-
You may want to say what motherboard you are currently using and what you are upgrading from.
AKDejaVu
-
A7V 133A, latest revision, and it will take the cpu as far as I know. I am upgrading from a Duron 800
SKurj
-
As far as I know the A7V133 can accept all the Palomino core Athlon XPs, which means up to an XP 2100. It will likely require a bios flash to support the higher speed CPUs. The next Athlon XP core Thoroughbred (sp) (which should be out by then, mobile Athlon XPs with this core officially launched a couple days ago) will undoubtedly require a lower core voltage; I do not know if the KT133a chipset does/could support them.
There are a couple things to mention here:
1. You won't get the added benefit of SSE instructions, which the Athlon XP added, unless you reinstall Windows. Depending on the applications you use, this could be very significant. Many drivers are optimized for SSE as well.
2. Your system would be significantly faster if you were to get a new motherboard with a newer chipset and DDR ram support. (Probably ~10 - 20 % faster overall with the same CPU ). This option would, of course, depend on your budget. Be sure whatever you buy can support the Thoroughbred Athlons when they do come out. (Yes, I read your post. ;) I just wanted to mention it. )
3. An XP 2000 class CPU is going to put out a lot more heat than a Duron 800, be sure you have good cooling. You are probably going to need a different heatsink.
-
if you get the amd athlon xp main board make sure you DO NOT GET WINDOWS XP WITH IT there are a lot of problems when used to gether trusst me i know (whe nyou set up a multiple sign in the paint msn messenger and the volume controls magicly dissapear)
-
The Athlon XP2000 is a powerful piece of hardware and would improve your PC performance greatly. However, unless you are planning to run a Database application with dozens of users or some massive algorithm it is mostly wasted power. Anything over around 1Gig is overkill for any home PC. There simply are no applications out there that require that kind of capability.
If I were you, I would look at a 1.2 to 1.5 Gig Athlon chip, which will provide you the Cache that the Duron lacks, and save the extra cash for a kick-ass video card.
Vol :)
-
too much power is like too much fun, too much money, too much love, too much...well you get the picture:)
Get as much as you can afford!
-
I have a quick question. Whats the difference between a
AMD Athlon T-Brid 1.4ghz 266 fsb
and a
AMD Athlon XP 1700+ 1.47GHZ
wheres the advantage in the XP series? I mean is the XP running at 1.7 or 1.4 ?
-
i believe its because of the XP's Quantispeed architecture is going to make it run alot faster then the T-bird, if i understood correctly it has 4 266mhz bus lines. i could be completly wrong. if i am im sorry and someone please correct me.
the 1.4 ghz rating is correct, the 1700 is supposed to indicate that it runs as fast or faster then the p4 1.7 ghz. the AMD is supposed to be alot more efficient than the p4 according to some people ive talked to and articles ive read. but like i said, i could be wrong as hell about this so you guys take it easy on me if i am. :)
-
The XP series of CPUs are sold under a rating system, rather than their clockspeed.
The 1700+ model actually runs at a clockspeed of 1.47 Ghz. Now that I've said that, read on, because there are some points that need made:
1. The XP 1700+ actually performs overall a bit better than a 2 Ghz P4 of the older "Willamette" design, and roughly better than a conjectural 1.7 Ghz "Northwood" core P4. Intel does not actually make a P4 1.7 Ghz with the Northwood core, those are at 1.6, 1.8, 2.0, 2.2, and 2.4 Ghz. You can tell a northwood from a willamette by the socket (478 vs 423 pins) and Intel uses an 'A' in the name for the 2.0 Ghz and below. All P4s at > 2.2 Ghz are northwoods. All Athlon processors perform much better than P4s at the same clockspeeds, AMD accounts for this by assigning a model number to them. You really should look at places like http://www.anandtech.com to get a true idea on how these CPUs stack up. The model rating uses a number of tests, and IMO is actually very conservative, but there are some areas where an Athlon will crush the P4, and some areas where the P4 is better than the Athlon.
2. The current Athlon XPs, which uses the Palomino core, vs the older Thunderbird core of the previous Athlon. Compared to Thunderbird (tbird), the Palomino core incorporates a couple improvements to the CPU itself which mean it performs a bit better than a Thunderbird would at the same clockspeed. Without detailing out exactly what they do (which would be a little beyond the scope of this post), these are support for SSE instructions (Intel created these, the P3 was the first CPU to use them), hardware prefetch, optimized die layout (reduces heat), larger TLB buffers (transistion lookaside), and a couple other minor things. The big one here is SSE instruction support, this allows the Athlon XP to take advantage of Intel P3 optimized programs AND programs which are optimized for AMDs own 3dnow instruction set. (Note, if you switch from a Tbird or Duron < 1GHz, you need to reinstall Windows to enable this function.) All told, a Palomino core Athlon XP will perform about 10 - 20% better than a tbird core Athlon at the same clockspeed. AMD invented a marketing name for these additions, calling it "Quantispeed." (Intel did the same for the P4, calling it "netburst" which btw has nothing to do with the internet and will not 'speed up your internet' by any amount whatsoever. The processor has ABSOLUTELY zero effect on your connection speed. I don't really know why Intel would want to confuse uninformed buyers with this.)
(Slash, I'm not sure what you mean by that "4 266 Mhz bus lines" thing. I'm pretty sure what you are confused with is what is known as the FSB (front side bus), which links the CPU to the motherboard chipset. The later tbird Athlons and all Athlon XPs use a 133 Mhz bus, but at double data rate (DDR), which is the equivilant of having a 266 Mhz bus.)
For the record, the XP 1500 + runs at a true clockspeed of 1333 Mhz, the 1600+ at 1400 Mhz, the 1700 + at 1466 Mhz, the 1800 + at 1533 Mhz, the 1900 + at 1600 Mhz, the 2000+ at 1666 Mhz, the 2100+ at 1733 Mhz. Can you see the pattern, there is a 66.67 Mhz true clockspeed difference between them. If you want to get really general with this rating system, you can gather that an Athlon would be around 33 % more powerful than a P4 Northwood at the same true clockspeed. That's making some really basic generalizations on typical PC usage and is full of loopholes, but it's not a bad assumption.
_____________________________ _____________________
I hope that answered some questions.
bloom25 - Electrical Engineer specializing in digital and analog circuit design. (Digital logic runs over into CPU design principles.)
-
HI,
I have had absolutely NO problems with my hardware setup. I built it myself. Here tis......
Athlon XP 1800+ (retail box, supplied heatsink and fan),
ECS K7S5A (SiS chipset) Motherboard,
512MB DDR RAM,
Radeon 8500 video,
Sounblaster Live 5.1 sound,
Altech Lansing ACS33 speakers,
NEC 19" monitor
Windows XP
The motherboard is only about $50 and is FAST and STABLE.
Total system cost was about $1000.
There are newer Motherboards now (my system was built 6 months ago) but you cannot go wrong with the above equipment.
thats my 2 cents
Patrick "Dawg1" Duncan
CO
VMF-182 Salty Dogs
Visit our Website (http://www.seaviper.com/saltydogs)
-
thanx guys
Bloom i am running XP Pro, ya think a reinstal would still be needed?
Yeah finances is the issue. I bought the 133A about a month before the 266 boards started appearing, and for the moment i can afford one or the other (cpu/mainboard)
Cooling is well looked after already by globalwin
BIOS already updated
if finances permit (and job #3 of 3 work out) i'll look at a new board.
The goals for this yrs upgrade are replacing cpu (duron 800 to XP2000+), vidcard(GF2 MX to GF4 Ti4400) and just maybe a monitor.
SKurj
-
skurj, I am quite pleased with my setup.
AMD XP1600
EPoX 8KHA+ MB
512 Mb DDR PC2100 RAM (this has more than doubled in price since I purchased in NOV last year.
Gainward Geforce 3 Original
onboard sound (dumped SBlive)
Volcano 6 fan and sink
several case fans
and all the other stuff
AH performance is awesome. runs IL2 well too. This setup should last me for quite some time. I also am using win2K as my OS. Highly reccomend it (or XP) for its stability and versatality.
-
To answer your orginal question I am running AMD2000+ on w2k and its fine.... cpu temps run normally at 50c which bothered me at first but I got used to it...... goes to 60c running AH...... have experienced no heat related problems.... actually have experienced no problems at all heat related or otherwise...
A7v266e MB raid0
-
Originally posted by vorticon
if you get the amd athlon xp main board make sure you DO NOT GET WINDOWS XP WITH IT there are a lot of problems when used to gether trusst me i know (whe nyou set up a multiple sign in the paint msn messenger and the volume controls magicly dissapear)
ummm ive been running Soyo Dragon + with XP1900 for 4 months now with Windows XP Pro. not 1 problem with either of them.
whels
-
maddog, your post is reassuring to me. Heat is a concern of mine. When I first built my machine I saw temps of 45C when running AH (I used artic silver as the medium), but now i frequently get to 56C, and when the machine is idle I drop to 47-48C. I am a little concerned, but if you are topping out at 60CV without probs...I feel better.
-
upgraded from 1.33 Tbird, geforce2 ultra & 512 ram on a IWILL KK266 to a XP2000, geforce 4 ti 4400 & 512 PC2100ram on a ASUS A7V266-E
still working the bugs out, will post benchmarks once it's stable..
-
I believe that like Win9x and Win2k, processor instruction support is determined and the OS configured accordingly when the OS is installed. I don't believe Windows XP would have changed that. I do know that SSE instruction support changes a registry flag, but I'm pretty sure there's more to it than that. Unfortunately I just don't know all the details here.
There won't be any problems putting the CPU in, you just won't be getting absolute top performance. (Though the difference between a Duron 800 and XP 2000+ w/o SSE support enabled will seem huge to you anyway.) The problem with Windows XP is that it could ask you to reactivate once you install the new CPU, and could also do the same when you add the video card. I personally refuse to use Windows XP or any software product with this "feature." (Office XP)
-
Originally posted by -ammo-
maddog, your post is reassuring to me. Heat is a concern of mine. When I first built my machine I saw temps of 45C when running AH (I used artic silver as the medium), but now i frequently get to 56C, and when the machine is idle I drop to 47-48C. I am a little concerned, but if you are topping out at 60CV without probs...I feel better.
Ammo,
ive seen the same temp rise. i 1st built this current system, i was getting 39C case and 42 C cpu temp under load.
now after 4 months or so break in, im 38C case and 49 to 50C under load cpu temp. cpu temp has gained about 10C over that time and dont have a reason why.
whels
-
Bloom,
I'm always wow'd by your posts....very detailed, lots of listed references
Do you build custom PCs?
-
Originally posted by LePaul
Bloom,
I'm always wow'd by your posts....very detailed, lots of listed references
Do you build custom PCs?
He's a "byte head":)
-
Skurj – you seem to be in good hands here :)
I have run AMD processors on Asus motherboards for three years now. If upgrading to another AMD processor, I can fully recommend Asus motherboards. Asus is dedicated to supporting AMD processors. They have just released a BIOS update to support the Athlon XP processor and their website is useful and informative.
The only thing against AMD processors is the temperature at which they run. They run VERY hot, and a heat sink/cooling fan is not an option – it is a necessity. Don’t just get any old fan; get the proper heat sink fan, and make sure the fan turns at sufficient speed to cool the processor. In addition to that, I also got myself a PCI slot exhaust fan. Asus provide a utility called Probe. It is a motherboard and CPU monitor. Keep this running at all times! It will alert you if things start hotting up, or if a fan speed drops below threshold. Just follow these tips, and your new processor should be fine! My Athlon is a 1.2GHz model, and it is fine for AH :D Don’t pay for something you don’t need.
-
Heyas
I have quite abit of experience building systems, I just don't always keep up to date on the latest developments. In my case thats not usually a bad thing... I typically stay 1 generation behind. This time i taking a leap to relative, as close to 1st generation as i've come.
Heat I am used to having overclocked every cpu I've owned. I run 47deg under load on the chip mid to hi 20's in case on my Duron at 975.
I am considering a new PS though. I am using an Antec 300W that came with the case. Think its maybe abit weak.
SKurj
-
I was concerned at first with temp but was told it was ok... AMD says die temp is 90 but it will shut down at 70 I think.... I was used to seeing my Intel 933 run at 38 which fostered my intial concern......
-
(Oh no, I feel another wall of text coming on... :D )
Athlons do generate a lot of heat, but to be fair, the Willamette core P4s puts out even more than an Athlon. Intel has the market power to force heatsink and power supply manufacturers to add special connectors and retaining clips to support them. AMD couldn't have gotten away with that.
The Athlon XP (Palomino core) incorporated some design changes to the shape of the core and some other circuitry changes that result in lower heat output and total power consumption than the Athlon Tbirds did. At the same clockspeed, a Palomino core Athlon XP puts out 20% less heat than a Tbird would. As clockspeeds have rose the Athlon XP 2100+ now puts out nearly as much heat as the Athlon Thunderbird 1400 does.
The XP 2100+ will probably be the last Palomino core Athlon, newer Athlons will be using a new core, codename Thoroughbred. The only difference as compared to Palomino, is the switch to a .13 vs .18 micron process. In case you were wondering, this .13, .18, .25, .35 micron process statement refers to the minimum possible transistor length. (A MOS transistor looks basically like a rectangle, they have both length and width. As manufacturing technology improves it has been possible to make the individual transistors smaller and smaller.)
Shrinking the transistors has many benefits:
1. Lower power consumption
2. Smaller die size (you get more chips off the same size wafer, thus lower costs)
3. Lower voltage requirements. (That's why a bios flash or a new motherboard may be required.)
4. Higher speed (there's more at work here, but suffice it to say that shrinking the transistors generally allows higher speeds to be reached.)
5. You can fit more transistors on a given size die.
Throughbred was released a couple days ago as the Mobile Athlon XPs 1600 - 1800+ for notebooks. Desktop users should see them very soon as well. This Athlon XP core will probably launch at 2000 - 2200+ (1800 Mhz) speeds and scale up pretty fast to around 2 Ghz true clockspeed (XP 2500+). The early versions will probably overclock pretty well, definately better than a Palomino core XP 2100+. If they are out when you do buy your CPU and your board can support them, this would be a good way to go.
If you wait a little longer, if AMD sticks to its roadmap, the Barton core Athlon XP will be released. (This is supposed to happen in Q3 this year.) At the very end of the year, AMD is supposed to launch their brand new CPU design based around codename ClawHammer and SledgeHammer. ClawHammer core CPUs are probably going to have a variation of the Athlon name, and the Sledgehammer core CPUs were just officially named the AMD Opteron processor. The Opteron will definately not be within the budget for most of us when released. It can run regular PC applications, but is much much more powerful than any other x86 processor design I know of. Some of the early benchmarks I've seen would indicate a 25 - 35% performance improvement over the current Athlon XP when running at the same clockspeed. The Clawhammer CPUs (which will probably be called something like Athlon 2) are supposed to launch at around a 2 Ghz clockspeed and the ClawHammer version will probably rate as an XP 3400+ at that clockspeed. The Opteron will be used in multiprocessor systems ( 4 CPUs to be exact ), so don't expect to be able to afford that one. ;)
There, now you all know what to expect for the rest of the year from AMD. :D
_____________________________ _____________________
LePaul, I have built computers for people. I mainly do it only for friends and family, when I have time. Recently I've been way too busy to even consider doing it though. I'm VERY VERY picky on what components I use in the systems I build, so it's not something I'd ever make any money doing. I always tinker and tweak the systems I built until they are as close to perfect as I can get them.
-
AMD TEMPS
I thought that you guys worried about you AMD CPU temps might want to see this stuff I read on the web.
http://www.anandtech.com/showdoc.html?i=1345&p=6
"Keep a careful eye on your core temperature. It is normal for 1GHz and faster Athlons to run at temperatures in the low 50s (Celsius), however by using a recommended heatsink/fan you can easily keep your CPU running in the 40s or maybe the high 30s. If your core temperature ever gets above 60C then you have reason to be worried. While that's within the CPU's operating limits, none of the currently available Athlons should ever run that hot if you're using proper cooling."
http://www.merc100.com/socketa_ocpg5.htm
"The technical data sheets for AMD Athlon and Duron processors show that they can operate up to a maximum core temperature of 90C before catastrophic failure. Obviously you will not want to run at temperatures close to that, and in all likeliness would not be stable at such high temperatures. The problem with these processors is that there is not a thermosister inside the core to give the true temperature of the core. Instead there is a thermosistor on the motherboard inside the socket, underneath the core. Because of this, the temperature readings that your BIOS and software report are severely under read. This means that if your BIOS reports a CPU temperature of 50C, the actual temperature inside the core is actually closer to 65-70C. Due to the design and compensation figured by each motherboard company, the amount of under read differs between the manufacturers. As a generalization though, the under read is usually about 15C. I tell you this so if you are running in the 50C range, and are experiencing lockups, you may want to look into additional cooling, instead of immediately expecting a RAM problem."
Enjoy!
r,
KRAT
-
Krat, that's a little outdated. One addition I forgot to mention for the Palomino core Athlon XPs was the addition of a thermal diode. Newer motherboards get the die temp right off the CPU, thus they read correctly. Older boards that originally used tbirds would still be off by a couple degrees, as they don't support this feature of the XP. It's important to note that no current AMD boards have a CPU powerdown on high temp function, even though Palomino (Athlon XP) and Morgan (Duron >1Ghz) are capable of doing so. (It's not very hard to implement, so I don't honestly know why it hasn't been done.)
-
bottom line educated ones... Should 56C (under heavy laod, ie AH) concern me?
-
Unless you have having problems, that's fine. It's a little warm, but well within reasonable limits. My 1300 @ 1339 Mhz Tbird runs at 45C idle and 59C full load (engineering mathematical apps). AH only can push it to around 55C.
I would periodically (once a month or so) take a look and be sure that temperature isn't coming up. My rule for Athlon Tbird systems is no higher than 60C full load.
Getting a heatsink off an Athlon tbird can be dangerous if you left that thermal pad on the heatsink (it glues itself to the core). Unless that temperature keeps coming up, I'd just leave it for now...
-
bloom--
I applied artic silver when I installed the sink. I completely removed and cleaned the thermal pad from the sink before mounting. I guess i should just leave it alone for now. I have been considering a better heatsink/fan for the processor recently though, maybe one of those copperbased jobs.
-
Anone have a good temp monitoring utility for Win XP?
Thanks,
DJ229
-
bloom-
since im bordering on clueless here, could you possibly explain the "quantispeed" to me? also, do you or anyone else know anything about ABS pc's? they appear to be using quality components and have gotten some good reviews. just curious before i drop 2 grand:D . thanks.
-
Slash, "quantispeed", like many other marketing names applied to PC components means absolutely nothing. ;) (Other examples are "netburst", "pentium", "hyperpipelined", "Athlon", "Duron", "Celeron".) None of these are real words. They use parts of real words to give you the impression that they mean something.
Athlon is similar to the latin for "performing unit" and Duron roughly translates as "lasting unit". "Netburst" is an Intel marketing term for certain parts of the P4 core architecture. "Quantispeed" is the AMD equivilant of that term. Marketing departments probably consume large amounts of pizza and beer, combined with a lack of sleep, to get themselves into the right frame of mind to come up with new totally meaningless but cool sounding words. :D
You aren't totally wrong with what you said above though, it is true that at the same clockspeed an Athlon will outperform a Pentium 4. Even though you know this, the general computer buying public often chooses a computer based ONLY on the clockspeed they see on the features list. Why an Athlon is able to do this is far beyond the scope of this post, and would require hundreds of pages on CPU design theory and architechural details of both processors.
I've seen a lot of analogies around which try to explain this in a somewhat meaningful way, but picture it this way: (I hope you like cars... :) ) A 454 Chevy V8 and a 350 Chevy V8 are both automobile engines. They both do the same job, and both interface with the same basic components (radiator, transmission, etc), but one is a "big block" and one is a "small block" engine. At the same RPM, which one is more "powerful"? What whould happen if the only rating system you had to compare the two engines was RPM though? Obviously the small block 350 can achieve higher RPMs than the 454 can, but that's not a totally valid comparison to decide which engine would be best in your truck.
I realize this is a silly way to try to describe this to you, obviously CPUs are not car engines, but it's not a totally inaccurate way of describing what's going on here. AMD's Athlon would be the 454 big block engine in the above example and the Intel P4 would be the 350. To explain to the public how the Athlon could perform equally with a P4, even though their clockspeeds were not the same, AMD invented a marketing term called "quantispeed". Basically they want you to think that the Athlon is somehow "turbocharged" and gets more work done per clock cycle. This is, in general, a true statement.
The Athlon is a kind of like a "brute force" type of processor. It can handle a higher number of instructions at once than the P4, and it's floating point math (FPU) unit is far superior to that of the Pentium 4. (The FPU is used for high precision mathematical operations.) Applications that make heavy use of the FPU almost always show the Athlon coming out ahead of the P4 in those tests. Most scientific, engineering, and Direct X games (and flight sims in particular) make heavy use of the FPU.
Compared to the Athlon, the P4 was designed to be capable of very high clockspeeds, but at the expense of how much it could process per clock. I don't know exactly why Intel chose this approch when designing the P4, but it doesn't hurt that it makes it very easy to scale up in clockspeed with much less effort than AMD requires to increase the Athlon's clockspeed. This ensures a fairly long lifetime of this design for future Intel processors. You may not know, but Intel's previous desktop processor design, the P6 core, was used under multiple names with only some fairly minor changes over its lifetime. (The P6, in all its various forms, was the heart of the Pentium Pro, Pentium 2, Pentium 3, and Celeron. These processors spanned from 1995 until 2002, as the Celeron is still based on this core.) I personally believe Intel wanted the successor to the P6 core to also be capable of being adapted and modified to serve as the heart of Intel processors for years. In that, they succeeded.
I don't want to mislead you and let you get the impression that the P4 is a "bad" performing design, remember that since Intel can easily boost up the clockspeeds they can match the performance of the Athlon at its lower clockspeeds to win over the enthusiast market and wow over the general public when if they were to see the huge difference between the clockspeeds the P4 and Athlon run at. When Intel released the P4 2.4 Ghz, that was the first time in over a year and a half that their top of the line processor edged out the top of the line Athlon for first place. The Athlon XP 2100+ still definately wins the "bang for the buck" race though.
AMD response, to keep from losing the business of the general (uneducated in computers) public was to invent the term "quantispeed" to describe how, for example, a 1467 Mhz Athlon can outperform a 1700 Mhz Pentium 4.
The CPU business changes very quickly as to which is the top performing CPU at a particular moment. In just a few days (May 2, from leaked information) AMD will be releasing a new Athlon XP core at higher clockspeeds. Expect Intel to respond soon after with a 2.533 Ghz P4 to match AMD.
At the end of this year, AMD will be releasing it's sucessor to the Athlon XP, or more exact 2 replacements. One will likely sell under the marketing name Athlon XP-64 and the other was just named the AMD Opteron processor. The lower performance Athlon XP-64 is supposed to launch at a true clockspeed of 2 Ghz, but will carry at least an XP 3400+ rating, or the equal of a 3.4 GHz P4 Northwood. The Opteron will be even more powerful, but it will be out of the price range for the average consumer on launch. Intel has a new P4 core (actually 2) which they plan to release on mid-2003 to respond to AMD. It's a never ending cycle for the next couple years at least.
I hope that's helped you some. :)
Unfortunately I don't know anything about ABS computers. (I build my own computers.)
-
Bloom25! I've just read your post above, and what a cracking read it was :) I read every word and learned a lot.
My own processor is the Athlon 1.2GHz. I upgraded to this from my previous AMD-K6-3 450MHz, and in Warbirds it made a hell of a difference. In Aces High, the Athlon seems fine. What would you recommend as a processor for Aces High to someone in the market to upgrade, and why?
I also have a question about late model graphics cards. I understand there's now a GeForce 4 with anti-aliasing, whatever that is. You've told us a lot about processors - can you now tell us about graphics cards?
-
If I may expand on this bloom25.
The clock rates of any given processor today is the speed of the core unit in the processor, but does not give you any idea of the amount of time (T-states) it takes a given processor to actually fetch AND execute any given instruction.
The number of T-states it takes to execute an instruction is difficult to measure today. This is due to the processor pipeline architectures, which is probably more key to the actual performance differences between Intel and AMD.
Remember, all the processors are handicapped by the speed of memory/RAM access, as it is significantly slower than the CPU core is running.
Where the processor clock rate helps the most, is in the FPU (if the FPU is running at the processor clock rate). Floating point is expensive, in terms of T-states. Intel has really never been known for having a fast FPU. It is no surprise AMD's FPU blows it away.
One of the tradeoffs in the design architectures of the two companies offerings occurs at the pipeline level of the CPU. This pipeline is where all the action takes place before the instructions are actually executed. Intel has typically a larger pipeline than AMD. This is done to help with predictive branching, at some expense though. If the branching is positive, then in that one instance, Intel may roll past AMD, but this is an iffy proposition. On the other hand predictive branching in a large pipeline can invalidate a significantly larger number of instructions in the processor pipeline and thus take a performance hit as the pipeline now needs to be fetched again before the processor can get back to work. Of course, the processors continues to work as the instructions come in, but this is much slower than being able to walk through the pipeline, as it were.
The larger pipeline also adds more T-states to any given instruction. Think of the processor pipeline as a tube, with one end being the memory and the other end being the CPU core where the intruction gets executed. The time it would take to roll a ball through the tube gets longer as the tube gets longer, if all other things are equal (i.e. the ramp of the tube and weight of the ball).
AMD, typically, has a much smaller pipeline than Intel, at the same clock rates. This contributes to a little more performance in real-time applications, as the T-states required to actually execute the instruction is less than in comparable Intel designs. In non-real time applications, such as databases, Intel might perform better than AMD as the longer/deeper pipeline would work to thier advantage. Depends on the data. AMD's more efficient FPU would help them overcome any advantage Intel would have if the database contained a siginificant amount of floating point math.
Now, with Intel's Northwood core, they have taken an elegant, but bruteforce approach to overcome some of thier inefficiencies. Moving to a 13 micron process allows Intel to reduce the number of clock dividers/multipliers in the processor and boost the peformance of various units in the CPU. I do not know if they actually did this or not, but I suspect they may have and it would make perfect sense. The lower mass of each gate would allow for faster switching times.
Intel also gained some other advantages when they went to the 13 micron process. Lower power consumption, which means less heat to dissipate. The ability to clock at significantly higher speeds, with better manufacturing yeilds (this is pretty significant by the way).
At this point in time, the performance differences between Intel (Northwood) and AMD is pretty negligible. Of course, tomorrow is another story.
Don't let the marketing buzz words drive you to make a decision. People get caught up in those things (which is what marketing is supposed to do), but they are pretty irrelevant. Just like video cards "Hardware T&L". The truth about that buzz word; with todays CPU speeds a well written application (game) can do the T&L significantly faster than the video card can. Even though you get some parallel operations going on, it is still not significant enough to help.
-
thank you bloom25 and Skuzzy. that was more than helpful. breaking it down into pickup truck engines is very helpful to us "redneck" types.:D once again i appreciate your responses.
i am smarter for reading them.
-
I was going to talk about pipelining and t-states Skuzzy, but analyzing t-states isn't totally accurate either. (Hardware prefetching, TLBs, and branch prediction throw in a big monkey wrench, decoding of some particular x86 instructions in rom.) If you guys wanted the exact number of stages in the pipelines of the Athlon XP and P4, it is 11 for the Athlon XP and 20 for the P4. This means, worst case, that the Athlon at the same clockspeed will be much "faster" than the P4 in executing instructions. (The pipeline is only 5 stages in the first Mac G4, and 7 or 8 in the newer models. This is one of the main reasons the Macs have been limited to 400 mhz to now 1 Ghz while Athlons and P4s were at 1 Ghz +. The architecture of a G4 is totally different than a PC processor so there's even more to get into there, but it makes a valid point.)
Skuzzy, on May 2nd AMD will release the Thoroughbred core Athlon XPs, which are on .13u. (As far as I know, that is the only change, but that in itself will reduce power consumption and allow higher clockspeeds.) AMD plans on making some L2 cache changes and a FSB increase with their "Barton" core in Q3 this year.
beet1e, what did you want to know about anti-aliasing? I.E. How does it work? What does it do? These are totally different questions, but knowing how it works helps in understanding what it does to improve picture quality. (Note, I don't know nearly as much about graphics chips as CPUs, but I do know basically how AA works ... or at least how it used to work.)
Hehe, what did you think of my auto engine analogy Skuzzy? :D (You'd be amazed how much it helps in explaining how very technical things work to relate them to cars.)
As for FPUs, the big difference between the P4 FPU and the Athlon FPU is a cost cutting meaure Intel implemented to keep die size down (and make costs lower). A FPU has to handle addition, subtraction, multiplication, and divide operations. (Along with load and store to memory operations, but we won't talk about that.) The main difference between the Athlon/Duron design FPU and P4 design (and actually P3) FPU broadly falls into the somewhat marketing and somewhat engineering term "fully pipelined." Basically what's going on here is that in order to do a multiplication or divide operation, a P4 must make use of both it's addition (subtraction - they are the same operation to a computer with one minor difference) unit AND it's multiplication unit. Thus the P4's FPU is not "fully pipelined." In other words, its multiplication and addition units share some common units within the CPU. This means a P4 must wait for an addition operation to finish before it can start processing a multiplication operation. Note, a multiplication operation takes FOREVER to do. (Unless it's a power of 2, that's just a shift operation. BTW a P4 lacks a fast barrel shifter, so these operations are also significantly faster on an Athlon.)
The Athlon has fully independant multiplication and addition units in its FPU. This means that not only do they not wait on each other, the Athlon can actually do both a multiplication and an addition operation AT THE SAME TIME. This makes a HUGE difference under some circumstances. As long as the result from the multiplication operation isn't needed for the addition operation (or visa versa), they can be done simultaneously. A game with a high precision flight or physics model (which needs to do FPU math to crunch out equations) will most likely be faster on an Athlon than on a P4. From my own (very limited) testing, this seems to hold true for AH, but either CPU is so fast that it's not like a P4 user will have an unplayable experience in AH. Some newer games, like Il2 and Commanche 4 show a significant lead on Athlon systems as the resolution is increased for this very reason.
-
Bloom25,
I would love to know where you got your knowledge from. I never realised there was so much complexity in my little Athlon processor which I can hold in the palm of my hand.
I spent a considerable part of my working life as a COBOL programmer, and also wrote some Assembler routines for IBM’s mainframe operating system, MVS (Multi-programming Virtual Storage, since repackaged with other components and re-released as OS/390). In Assembler, each line of code represents one machine instruction, unlike COBOL, where one statement can generate 20-30 or more machine instructions. I even remember coding a SLL R4,1 Assembler instruction to shift left logical the bits in Register 4 by one position, so I know exactly what you meant about the longwinded nature of that multiplication process. If multiplying by a number which was not a power of 2, it would be necessary to load the value to be multiplied into an even-odd pair of registers, and load the multiplier into another register. The reason for needing two registers for the value to be multiplied was because the system knew that multiplying two registers together could result in a multiplicand value which could only be accommodated in a doubleword, or two registers. Much of the longwindedness of such multiplication operations stemmed from having to set up at least three registers to perform the operation, and that could mean storing their contents first, and restoring them afterwards. If I knew I was dealing only with smaller values, I opted for the halfword variants of Multiply instructions, and avoided having to use register pairing, the result fitting into a single register because the Max halfword multiplier value would be 32,767, the leftmost bit being reserved for the sign. Better still was to use the Decimal instructions, which did not use registers at all. I don’t know whether Intel or AMD processors support decimal instructions in the same way as IBM’s mainframe processors?
I remember reading about “Mr. Intel” in TIME Magazine a few years ago. He is a Hungarian gentleman who escaped Hungary with his wife around the time of the 1956 uprising. He went to America and changed the world with his invention of the processor chip. The very first version of a semiconductor chip was shown in the TIME article. It’s amazing to think that what we have today sprang from that...
If anyone would like to blow their minds on IBM Assembler for OS/390, check out IBM's Principles of Operation (http://publibz.boulder.ibm.com/cgi-bin/bookmgr_OS390/BOOKS/DZ9AR007/CCONTENTS?DT=20010725142528) manual.
-
Quite correct bloom, it is very difficult to get the t-state information from a pipeline full of operations, but one can derive the t-state information from a single operation.
Basically, a program could invalidate the cache, then execute a single instruction and wait for the return of the operation before moving on.
A little tricky, but it can be done.
Yes, the barrel shifter in the Intel processors is not very good. Go figure, when Motorola introduced the 68020 family processor and FPU (long time ago), they had a barrel shifter thay took one t-state to complete any shift operation, regardless of the bit count. I remember seeing the Intel guys jaws hit the floor.
Thanks for the information about the AMD release. I hope they can get thier yeilds up.
-
Originally posted by beet1e
I remember reading about “Mr. Intel” in TIME Magazine a few years ago. He is a Hungarian gentleman who escaped Hungary with his wife around the time of the 1956 uprising. He went to America and changed the world with his invention of the processor chip. The very first version of a semiconductor chip was shown in the TIME article. It’s amazing to think that what we have today sprang from that...
As I remember, Jack Kilby of Texas Instruments and Robert Noyce of Iowa (later Intel) were credited with the development of the integrated circuit. They shared this credit in the form of the 2000 Nobel Prize for Physics with Zhores Alferov (Beloruss)sp?)) and Herbert Kroemer of CA.
Maybe the guy you are thinking of, known as "Mr. Chips" was Swiss-born Jean Hoerni, a brilliant engineer that used silicon for the first time combined with an insulating method known as the planar process. He ended up making Fairchild Industries into a $100 billion company. Passed away in '97 at 72. That him?
dh
-
Horn,
The guy who I read about was, I believe, Andras Grof who at some time changed his name to Grove. I managed to find this article (http://news.com.com/2009-1001-823656.html?tag=cd_mh) about him.
-
In case you were wondering, this .13, .18, .25, .35 micron process statement refers to the minimum possible transistor length.
I don't believe this is correct bloom. As a rule, it refers to the transistor gate length.. not the transistor length. Even then... it stopped being accurate in that regards about 3 years ago. Now, it probably more reflects a metal 1 line width than anything else.
AKDejaVu
-
BTW... for all you physics guys out there... the wavelength of the light being used to generate these structures is .248 micron (some .193 micron too... but not in anything on the market from Intel).
AKDejaVu
-
That sounds right DejaVu. I didn't even give it a thought and just assumed the measurements of the ethchings were done via the width of the trace (as it were).
.248 micron...thanks,..oh oh. I am a geek! Who is using .193? Heavy stuff there and very costly too.
This has been a fun thread.
Isn't the length of the line fixed by the capacitance value in conjunction with the propagation delay of the signal/current in order to insure smooth transitions of power from on to off to on states? It's been a while since I did circuit design.
-
Originally posted by beet1e
Horn,
The guy who I read about was, I believe, Andras Grof who at some time changed his name to Grove. I managed to find this article (http://news.com.com/2009-1001-823656.html?tag=cd_mh) about him.
Ahhh, cc. He and Noyce were both at Fairchild in the 60's--they founded Intel in '68. Dr. Grove has authored some awesome books on the semiconductor as well as one about management that is a great read. I've met him twice (my whole family worked for Intel at one time or another) and he's not only incredibly bright, but also charismatic, which always suprised me as he's such an uber-geek.
Here's more:
http://www.intel.com/pressroom/kits/bios/grove/bio2.htm
dh
-
Horn,
I read the article whose link you posted, and I now see that the article I must have read was the 1997 TIME Man of the Year profile. Gawd, was it that long ago?!
-
Yes, I meant gate length... :o (I'm pretty bad at reading over my posts to make sure I didn't mess up something simple. :( ) I hate to admit it, but most of my knowledge is based on work with a .5um (HP) and .35um (TSMC) process, so I'm not totally clued in on particular issues dealing with newer processes. With these two processes, the .5 um, for example, refered to the minimum gate length, and the minimum spacing between contacts and metal layers was .25 um (if I remember right). You don't really have to worry about it, because when you run DRC (design rule check) it will tell you if you violated any design rules for that particular process.
Beet1e, most of the stuff I've learned has been through CPU architecture classes, digital logic design classes (most of my interests are in ASIC design which are coded in VHDL, but many of the concepts hold true for both CPUs and Asics), and a LOT of personal interest in the area that give me the drive to actually attempt to "read" books on the subject. I've also done some work with various microcontrollers (Intel 8051/8031, PICs, etc). (Trust me, engineering books are neither easy to read and really aren't the slightest amount of fun either... )
If you want a general overview at a slightly higher level than what I posted above, I think there are some good articles at http://www.arstechnica.com . (Animal posted a link to that site, and I think there were some articles there about the K7 (Athlon) core. I don't know how accurate it is, but from the few articles I looked at it looked pretty good.)
Most of the design I do (unless it's very basic) is actually done via a programming language called VHDL. This language is similar to C in structure, and is very easy to read if you have any knowledge in programming. It allows you to define "components" which are somewhat similar to objects in other programming languages (Java comes to mind). It allows you to remove yourself to some degree from the actual hardware and focus more on behavioral aspects. For example, given the correct library for the type of logic gates you are using, you can just type: a <= vector1 + vector2; and when this code is compiled and synthesized you will actually get an adder (what type depends on whether you wanted it optimized for speed or area). You can do structural coding as well, and the synthesis tool will make a number of optimizations for you if you code it correctly. (And gates replaced with Nand gates, ors with nors, etc)
This language was developed for the DoD and the first version became available in 1987 (VHDL '87), it was revised in 1993. There is some current work to extend it to analog circuits as well. Nearly ALL Asic (application specific integrated circuits) designs are now done to some degree using VHDL or Verilog (similar to VHDL, but not as good IMO).
-
I'm not sure if this is what you were getting at Skuzzy, but I can tell you that the interconnects themselves account for much of the delay in modern circuits. As with just about everything else, it all comes back around to RC basics.
These lines have fairly high resistances, and there is often considerable capacitive coupling between metal layers. (I was looking for some exact values, but I didn't find anything new enough to be very useful.) It's also worth noting that the contacts themselves have considerable resistance, and unfortuantely that varies widely. (For the HP .5 um process I did a cascode amplifier design with, I'm thinking it was between 25 - 100 Ohms per contact. )
-
Close enough bloom.
I got out of hardware design when I realized TTL designs were going the way of the dinosaur. I did a few ASIC's, and many PAL's, but found the work rather boring. As a matter of fact, that is one of the reasons I got into software design. I found it much easier to write the simulation software for the hardware designs and much more mentally challenging. I wrote a circuit design simulation tool when I was doing those ASIC's and PAL's which allowed me to do basically one shot designs. Made it really boring after that though, but allowed me to migrate into software design.
I still keep my hand in it, by doing work around the house. My cat door was a fun project. I built a 4 bit microprocessor which controls the cats door.
Attached a magnet to her collar and when she comes within 6 inches of the door, it opens for her. Pretty basic sensing device coupled with some digital controls for the servo stepper that operates the door.
Hehe, I probably have about 5 thousand transistors in use in my home for all the things I have built.
Still, I like to keep my feet in it all. Of course, I dont find engineering manuals boring at all. They fascinate me. Most people would find them incredibly boring though. I remember getting my first TTL logic parts manual from TI. Thought I was in heaven.