Lot of stuff to check out, so grab a brew...
We're going to take a closer look at the current PC platforms, and try to get a little more useful info than we'd get from your typical site review.
Ok, the common hardware for all platforms are: 120GB WD 8MB hard drive, stock Radeon 9700 Pro (Catalyst 2.5, default Performance settings), SoundBlaster Audigy, and Windows XP Professional SP1.
For the AMD setup, I used the 8RDA nForce2 motherboard and the 2400+ processor. For single-channel tests, a 512MB DIMM is used, and for the dual-channel tests, a pair of 256MB DIMMs.
For the Intel setups, a 2.4B processor was used, an Abit BE7-RAID for 845PE (single 512MB), a Gigabyte 8IHXP2 for 850E (two 256MB), and a Gigabyte 8INXP for Granite Bay (two 256MB).
I'm going to start out by saying I cheated...I was originally going to do a no-mods, stock-fans write up (i.e. what any Joe-blow could pull off if he just ran out and bought stock parts). Turns out, this would have been really boring, because frankly, the stock Athlon fan sucks. It just can't handle high voltage (well, for overclocking your computer it can't...if you want to use your computer case to cook food, then the stock Athlon fan is great!). I went and grabbed a Thermaltake copper jobbie...added a lot to the overclockablity of the 2400+, and a lot of noise. Furthermore, I added a little fan to the northbridge heatsink, as this let me pass Prime95 at the highest FSB's (that heatsink gets HOT). For the P4, I stuck with the stock Intel heatsink. Oh well...I guess it's to be expected that a little extra effort goes into Athlon overclocking, so I don't expect this little bend of the rules to take away from the comparison too badly
The tests run are Sandra 2002 Memory Bandwidth, Comanche default 800x600, Quake III 1.17 demo001 default Fastest, Jedi Knight II jk2ffa High-Quality 1024x768, Unreal Tournament Flyby and Botmatch with Anand's HQ scripts at 1024x768, and 3DMark 2001 SE default.
Let's get after it...
Stock Comparison:
All four are very nice stock performers....very competitive. I'm really impressed here by how well RDRAM still holds up today...I mean we're talking tech that's over two years old now, and it's still leading the P4 pack (probably the biggest reason so many companies pulled out of launching a Granite Bay solution...I mean, most computers shipped aren't overclocked, after all). The XP/nForce2 combo is a particularly strong UT2K3 performer...very nice.
Let's crank it up a bit...
150 FSB Comparison:
All scale pretty damn well. Of note to me is the nature of the scaling between the two memory types. The DDR setups seem more D3D friendly (UT2K3 flyby and 3DMark), with better scaling compared to RDRAM. RDRAM is quite kind to everything else, particularly OGL/Carmack-creations.
Turn that dial a little more...
156 FSB Comparison:
That memory-tendency continues...note that the GB board caught up with the RDRAM platform in D3D, and just eeked past it, while the rest of the bunch all made up ground as well. RDRAM continued to be quite affable to everything else...might be a real monster with Doom III, and photo/video editors sure dig it. All these platforms are quite nice, in my opinion.
That's where we'll end the direct scaling comparisons, cuz that's where RDRAM quits, and I'm almost out of room with the 2400+ clockspeed. We'll take a deeper look at the remaining solutions, starting the the XP/nForce2...
A closer look at the XP/nForce2...
Thank heavens for the nForce2, is all I've got to say. I love it...I don't know what I'm more impressed with, the board or the chip here. If I'd have had to stick with a VIA chipset here, this would've been called 'An Intel Massacre Comparison...'
AMD overclocking with this combo is definitely more fun...I ran more benchmarks with this combo than the other three combined. The ability to adjust the multiplier and tweak the FSB is just sweet, and when you throw in the differences between single and dual-channel, your options become quite numerous. On the flipside, it's also more frustrating...I was greeted with the dreaded corrupt OS load three times while trying to push this sucker, and had no such pleasure on any of the Intel platforms. I consider an imaging program a necessity for overclocking (I love you, Ghost).
Let's play...
Single-channel versus Dual-channel:
I just wanted to see this at a couple different levels. You wouldn't think there would be a big difference going dual-channel with the XP, because single-channel simply matches the Athlon's FSB bandwidth. And there isn't a very big difference either. Still, there IS a difference, and it scales as well. Let's look at this further...
Different levels of 2000MHz:
I ran dual-channel (DC in the charts) at stock 15x133, 12x166, and 10.5x190, all which gave me 2000MHz. 190 was it for dual-channel...this was where the OS liked to puke. Perhaps not coincidentally, 190 was also as far as I could get in DIMM slot 3 single-channel (the farthest DIMM, and the channel that you have to populate to enable dual-channel with multiple DIMMS). Maybe it's not particularly dual-channel operation, but the nature of the DIMM slot itself, or the distance of the slot/traces from the chipset? In any event, using DIMM 2 (the closest to the chipset) let me hit 200 FSB (10x200, 2000MHz) in single-channel mode. Note that this is the winner of this little 2000MHz shootout...single-channel at 200 beat dual-channel at 190 in everything but Comanche and UT2K3 botmatch. Also interesting is how very close 190 DC is to 200 SC...you could say that dual-channel is worth approximately 9-10 FSB, and be pretty damn close to correct.
As for the overall picture...WOW! Look at how much performance we wrang out've 2000 Athlon XP MHz here, by simply manipulating the multipliers and the FSB! Very impressive indeed...but this is also where something started to bug me a bit as well. I'll return to that in a few places as we continue. 2400MHz at 1.92v was the best I could squeeze out've this little proc, which I think is pretty damn good for mid-range air cooling (i.e. not using a Delta hair dryer and losing my sanity from the noise). Anything higher, and heat ruined the party. Still and yet, a strong showing...let's look:
XP at 2000 and 2400MHz:
Since single-channel gave me the top overall results, I ran all the results again in single-channel mode to get an accurate picture of what's happening. The left two columns are at 2000, to see what the 200 FSB brought to the party, and the right two columns to see what the extra 400MHz of clockspeed is giving up. The total gain from 2000/133 to 2400/200 is simply delicious. The part that concerns me is column one versus column 3...what 400MHz of raw clockspeed does. Anybody notice that the boost to 200FSB alone (column 2) does a great deal more than adding 400MHz of raw clockspeed in everything but Comanche (the most CPU-dependent of these benches)? Good, glad you were paying attention
Yep, that's what's buggin' me...I'll get into that again, but for now, Intel's feeling lonely...
A closer look at Intel...
I'll just break down my opinions by platform...
845PE: I dig it. For the 'cheap' alternative, it sure brings it nicely. Yes, it's the slowest, but it's the cheapest, and it's very flexible for overclocking. If you have crap memory, you can run lower ratios (though I'd recommend getting good memory), you're not really limited on FSB, and it's really stable. This is the board I'd tell people sitting on the fence to get if they're waiting out next-gen tech...you can't go wrong here. Cheap, fast, overclockable, stable...all good.
850E: Got a soft spot for this tech...a long time favorite for the P4 enthusiast. It's still the top performer for stock-runners and up-to-150 FSB overclockers. If you still have this platform, I don't see any reason to switch to any of the others in here...wait until at least springdale or later. Your upgrade path is a chip with a higher multiplier. For example, say you've got this 2.4 at 2808/156 (your limit). Don't go buy another board/memory...just go nab a 3.06 chip when they drop. Bam, 3450 or better (about as much as you can hope for on air). As a matter of fact, if you look at the benchmarks, I'd even recommend this combo to those of you getting a 3GHz chip and using air cooling...you're very unlikely to pass 150 FSB anyway. That's the thing with Intel platforms...you basically let your muliplier and your cooling dictate which choice is best for you. With the 850E and the right chip, you can probably even wait past Springdale, and jump on the Prescott/DDR II bandwagon.
Granite Bay: I frankly don't know what to think about this thing, even after weeks of use. I can see why it didn't get a wide-spread release. Still, I'm going to have to call it the 'platform of choice' for the enthusiast at this moment in time...and I'll note, that this is a very brief (and expensive) moment. You get RDRAM-like bandwidth at higher FSB's than you can hit with the 850E. And you pay for it.
I'll note a few things that I've noticed about GB (off-2.4 topic)...it does, as Michael at Lost Circuits mentioned, perform better with Hyperthreading with 1GB installed versus 512MB (this is weird, because it doesn't happen on the other platforms). Furthermore, memory timings make a MUCH bigger difference on this platform than ANY other (more than double so). In essence, this means that, if you get this board, you should not only spring for more memory, but GOOD memory...hey, you're already dumping over $260 on the board, why the heck not, right?
One last note, since I was playing with a 3.06 and Hyperthreading...Hyperthreading actually scales well. At stock speeds, enabling HT put a slight hit on most the benchmarks. However, when overclocking, every single Hyperthreaded bench beat it's non-HT counterpart after 140 FSB. Pretty sweet...
Ok, back to the numbers...here's the max I could squeeze out've the three platforms with a 2.4B:
Maximum Intel:
There ya go (stuck the stock GB on the end for comparison). I was able to pass overnight Prime95 1 FSB higher on the 845PE than GB. Points of interest: Damn, the 850E is close to the 845PE at 3:4, even at these levels, and giving up 200MHz. Damn, GB has some nasty bandwidth! Damn, the 850E is close to the GB in everything but 3DMark (and even that difference is only 2%). Damn, these are all pretty damn close! Did I say damn yet?
GB is the winner (well, performance winner), plain and simple, so let's bring back in the Athlon...
'You always saved the best for last'...:
Can you say parity? I put the stock numbers versus the max numbers for each in there, to get an idea of what we're dealing with here.
Either way, that's impressive. Very close all around. A special nod to the XP/nForce2 for pulling this off with a LOT less cash (then again, look how close the 845PE was to GB, at equal money to the nForce2...).
I'd call the XP/nForce2 the clear winner, but...
Where's the upgrade path? Yes, I consider this important.
Let's take a closer look (naturally):
Athlon/P4 Clockspeed/FSB scaling...
It's been widely held that the Athlon does more clock-for-clock than the P4, opening all sorts of arguments. How accurate is that statement, though? Let me explain this chart (I love excel and formulas). What we have here is an equal 12.5% boost in clockspeed and FSB for both. Besides each bench is the percentage gain for that action (quite interesting info in it's own right). The composite gain is the sum of those six values divided by six. The efficiency figure is simply the composite gain divided by the actual clockspeed/FSB boost.
Both actually impressed me...I figured both would be much less that the 80+ percents they both gained. However, note that the P4 definitely gets more out of the deal. What's up with that? Let's delve into that further...
Athlon/P4 pure clockspeed scaling...
Pure clockspeed. To explain the chart, that's a 20% gain in raw Athlon clockspeed versus a 27.5% raw clockspeed gain for the P4 (note, no hyperthreading). Let's just jump right into what's important (which helps explain what's been bugging me about the Athlon all along): from raw clockspeed gains, the P4 gets 60.11% efficiency from the deal, where the Athlon gets less than half, at 45.80%. This shocked me, really, given my preconcieved notions of the two processors. It's been said that the P4's longer pipeline is more 'future thinking'...you know what? Maybe that wasn't BS.
What does it all mean?
Well, for the Athlon, this means I get a helluva lot for my dough. However, what else I'm I going to get? Look at where I'm at...2400MHz, 200FSB. Damn nice. But what can I look forward to? If I went out and dropped $400 US on a 2800+...guess what? I actually may not get ANYTHING out of the deal...I'm already at 200 FSB, the best I can do (without more modz, naturally). I might actually get ZERO if I can't push the clockspeed higher. BUT (and that's a big ol' butt)...who cares? I'm going to get less than half my efficiency out of that...and what can I expect, anyway? 2500MHz, 2600MHz? Wow...
Yes, that's what bums me out a bit. Maybe Barton??? I don't know...it's going to be a boost in cache, which should help, but who's to say it'll actually clock as high? It might not, given the extra heat...now what?
In any event, I'm impressed overall with this combo. The negatives for me are the heat and noise (me personally, I'd be clocking this down 100MHz, because it can do that with significantly less voltage and heat...maybe even a quieter fan!), and the upgrade path. Making an educated guess, the best I could hope for as an upgrade with this setup as is would be Barton at around 2500-2600MHz...and that's hardly a given on this cooling.
Which brings me back to the P4. Very nice results with all three platforms, made even more impressive considering this is a 'lowly' B0 chip with near-silent stock Intel cooling. I've got obvious HERE AND NOW upgrade paths (I say that in bold in protest to AMD's recent trend of paper-launching...PLEASE don't screw us with Barton!).
Very nice scaling, and the BIG wildcard...Hyperthreading.
Dammit, Hyperthreading is SO hard to describe...you've seen web sites try to quantify it, but it's REALLY hard to describe, because it's not really benchmarkable (is that a word?
I. LOVE. IT. If you actually USE a computer, you will LOVE Hyperthreading. Everything SNAPS, is the best way I can describe it. Probably the best example I can give is...say I'm playing a game online with my buds, and dammit!, my favorite TV show is coming on. I did mention this 9700 Pro is an All-in-Wonder, right?
With hyperthreading, I alt-tab out, and hit record. Then I run Comanche (a heavy-CPU game). No sweat, everything's beautiful.
Guess what? I turn Hyperthreading off, and do the same...Comanche is a lot jumpier, but worse, I get a whopping 38% frame loss in the capture (this is HORRIBLE). I tried the same feat on every platform...between 32 and 40% frameloss...with Hyperthreading, zero.
This is the best way I myself can explain hyperthreading...not to sound dramatic, but it really does make the impossible possible. I consider it THAT big of a deal...this is an awesome upgrade path to ANY of the P4 platforms I've looked at here. THIS is why I can't declare a winner myself...Hyperthreading, plain and simple. It's not a small deal. It might be the most significant proc-tech I've seen in years (very exciting).
So you'll have to do some work yourself...what's important to you? Hopefully, some of this helped in your decision...
Ok, I couldn't quit there
P4 future...
That's the P4 3.06 with hyperthreading stock, versus Vapochilled 3750.
Damn, it still scales well, no? We're actually hitting the limits of the awesome 9700 Pro here...the vid-dependant benches (UT2K2 flyby, 3DMark) are holding us back from +90% efficiency. VERY nice...had to get a little extreme for ya
This was a lot of work, and a lot of fun. I invite conversation on this...
Cheers
Reply With Quote
WOW
Keep the good info comin
Bookmarks