Cross platform benchmark comparisons. — Joe
SUMMARY: The case for an upgrade depends on where you start from – the more current your system, the less the benefit.
Choices – sometimes good, sometimes confusing. If you’re thinking of upgrading and wonder which way to go and what you’ll get, what follows may shed some light on what to consider.
As Ed pointed out in 2002 Double Trouble, the high-performance end is likely to be quite competitive in 2002 and our long-range plan has been to increase Intel coverage in 2002. To my delight, one of our Intel readers very kindly sent me a PIV and an ACORP motherboard that runs SDRAM. Shortly thereafter, Lucky Star sends me their PIV motherboard that uses DDR (detailed review in the works). I will use these primarily to test PIV heatsinks, but since I had them, I figured I might as well do a few comparisons.
Let’s get this one out of the way first:
Benchmark
| PIV SDRAM
| PIV DDR
| % Diff
|
SiSandra CPU
| 3190/2047
| 3187/2055
| NA
|
SiSandra MM
| 6723/8261
| 6724/8266
| NA
|
SiSandra Memory
| 681/682
| 1102/1228
| -38/-44%
|
Quake
| 149.1/144.0
| 182.3/176.5
| -18/-18 %
|
3D Mark 2001
| 3568
| 4126
| -14%
|
Let’s call this the RAMBUS Performance Tax and leave it at that – obviously, if you are upgrading, run – don’t walk – away from the PIV SDRAM option.
OK, let’s get to some real choices.
If you were walking in to COMP USA, for example, and saw two systems side by side, one a PIV 1700 and an XP 1700+, you might think they were comparable systems. The salesman whispers in your ear “AMD is fooling you – it only runs at 1466 MHz and the PIV at 1700 MHz – more is better!”
“DAMN!” you say, “Show me how much AMD is fooling us – let’s run a few benchmarks!” and you find:
Benchmark
| XP 1700+
| PIV 1700
| % Diff
|
SiSandra CPU
| 4109/2033
| 3187/2055
| 29/-1%
|
SiSandra MM
| 8170/9402
| 6724/8266
| 22/14%
|
SiSandra Memory
| 793/897
| 1102/1228
| -28/-27%
|
Quake
| 190.3/188.2
| 182.3/176.5
| 4/7%
|
3D Mark 2001
| 4506
| 4126
| 9%
|
Well now, a real interesting set of benchmarks! (If you were AMD, wouldn’t you de-emphasize MHz?) What hits you in the eye is the almost 30% difference in bandwidth in favor of the PIV. However, even with this disadvantage, the AMD’s XP1700+ manages to eke out some modest (probably invisible on the screen) gains vs the PIV 1700.
But since you know more than the average “walk into COMP USA” type, you want to “peel back the layers” of this onion, so you ask to and run both systems at the SAME SPEED:
Benchmark
| XP @ 1400
| PIV @ 1400
| % Diff
|
SiSandra CPU
| 3930/1929
| 2617/1683
| 50/15%
|
SiSandra MM
| 7797/8972
| 5525/6789
| 41/32%
|
SiSandra Memory
| 802/859
| 1101/1223
| -27/-30%
|
Quake
| 184.9/182.9
| 161.8/161.7
| 14/13%
|
3D Mark 2001
| 4751
| 3766
| 26%
|
Double digit differences are significant and something you might see on screen. The salesman screams “LIES all LIES! I can run them both at 1400 MHz and you’ll see a different picture!” He takes you to the back room and shows you a PIV running at 13 x 130 FSB (1400 MHz) with this comparison:
Benchmark
| XP 10.5×133
| PIV 13×130
| % Diff
|
SiSandra CPU
| 3930/1929
| 3179/2043
| 24/-6%
|
SiSandra MM
| 7797/8972
| 6714/8250
| 16/9%
|
SiSandra Memory
| 802/859
| 1096/1237
| -27/-31%
|
Quake
| 184.9/182.9
| 218.0/211.9
| -15/-14%
|
3D Mark 2001
| 4751
| 4274
| 11%
|
Of course, what he neglects to tell you is that the PIV is running the PCI bus 30% faster¹ than spec while the XP is not, and that the PIV he is using is an engineering sample with an adjustable multiplier (not available to the public) – but minor details.
You then say “I have a T-Bird 1400 @ 1650 (11×150), Iwill KK266+ using SDRAM – let’s see how it stacks up:
Benchmark
| XP 1700+
| PIV 1700
| T-Bird 1650
|
SiSandra CPU
| 4109/2033
| 3187/2055
| 4599/2282
|
SiSandra MM
| 8170/9402
| 6724/8266
| 9049/11294
|
SiSandra Memory
| 793/897
| 1102/1228
| 631/710
|
Quake
| 190.3/188.2
| 182.3/176.5
| 172.4/169.9
|
3D Mark 2001
| 4506
| 4126
| 4257
|
“No fair!” the salesman screams, “You’re overclocking your system. You can overclock the PIV or XP also”.
“This is true, and I’ll probably get a 10-20% gain over the T-Bird. The REAL question – Can I see it on screen?”
Fade out on the salesman’s blank stare.
Depends:
- If your system is not more than one year old, why bother?;
- If you MUST upgrade now and you are not into overclocking, then a similarly rated PIV or XP based DDR system will do you fine;
- If you are into performance, wait for the 0.13 micron CPUs;
- For the avid overclocker, I think it comes down to this – if you’re OK with varying only FSBs, then Intel; if you want to vary both multiplier and FSBs, then AMD (I like more options than less).
As 2002 unfolds, we will do more testing, but I don’t think you’re going to see a “knockout” CPU from either Intel or AMD. If AMD makes enabling multiplier adjustments all but impossible, it may be a real toss-up.
¹For example, note that Quake at 218.0/161.8 = 35%, roughly in line with the 30% bus increase.
TESTING NOTES: All tests were conducted using 256MB Crucial DDR or SDRAM at 2522 memory settings; in some cases, 3DMark2001 would not run with these settings and was detuned to Normal. Video card was a LeadTek Geforce 2 Pro, 32 MB. SiSoft Sandra benchmarks used the 2001te version, Quake was run at 600 x 480 and 3DMark2001 at 800×600, 16 bit, sound enabled on all motherboards.
No replies yet
Loading new replies...
Member
Join the full discussion at the Overclockers Forums →