Secret of Intel Core Duo CPU

I have 2 systems running with Q6600 Intel Core Duo Quad cores. They can crunch a lot of numbers and last price I saw for them online for only $290-310 each. Both these systems use a Asus P5B-VM uATX motherboard, 2GB of DDR2-800 RAM, Enermax Liberty 400w modular power supply, 80GB SATA drive and Ubuntu Linux 7.04.

The Sneaky Secret of the Core Duo
I have since discovered the secret of the Core Duo - it is a rather smart (but sneaky) marketing move by Intel. My first Core Duo was a E6300 running at 1.8Ghz. I remember reading all the magazine reviews that showed the 1.8 kicking the butt of AMD's running at 2+Ghz. It was like Intel had found a new secret sauce for making processors. The magazines played along and all bemoaned how AMD was doomed unless they could counter this brilliant new secret sauce.

But I now know the secret. At work I happen to have a few Pentium D Dual-Core running at 3.6GHz - a nice number; just happens to be twice the rated speed of the E6300. However, if you run some old-fashioned MIPS/FLOPS hardware benchmarks - the kind magazines NEVER run anymore - you will find the Core Duo at 1.8Ghz pretty much matches the Dual-Core PD at 3.6Ghz at basic integer tests, and does only 50-60% as well at the floating point tests. So clearly, the Core Duo E6300 has portions of the chip running at 1.8GHz and portions running at 3.6Ghz (double-clocked). Such technology is easy these days - 10/100Mhz Ethernet hardware runs with a 25Mhz crystal and uses a clock multiplier to gain the 100Mhz cycles. Intel must be using a 1.8Ghz crystal and clock multiplier to run portions of the chip at 3.6Ghz. This also makes sense given the Core Duo concept came out of Intel's "mobile" design team - people who realized that running different portions of the chip at different speeds helps cut power usage and heat generation.

The really brilliant (& somewhat risky) marketing move was to call a chip like the E6300 a "1.8GHz chip" even though it ran at 3.6Ghz ... this is what caused the big media back-lash against AMD. Had the magazines tested the E6300 as a 3.6Ghz chip ... the test results would have been disappointing compared to a true 3.6Ghz Pentium D dual-core. It would have shown the Core Duo as a chip which sacrificed performance for lower power.

However, since it was called a 1.8Ghz chip, all the tests were showing up as 30 to 70% "better than expected". Magzines had no problems with the apples-to-oranges comparison of a 1.8Ghz Intel out running a 2+Ghz AMD since the AMD had (on paper) a higher clock rate. But what they didn't understand was that they were actually comparing a slightly crippled 3.6Ghz Intel against the 2+Ghz AMD; of course the crippled 3.6Ghz chip would beat out the 2+Ghz AMD

The Cost of Power

Summary: Some musings comparing work accomplished by my computer to my personal out-of-pocket costs for the electricity to feed it 24-hours a day - something I dare say very few home computer users look at.

Last month I upgraded a Celeron D 2.53GHz media server to a Core 2 Duo 1.8Ghz. When not using the server for "media" (watching DVD or recording broadcast TV), I run BOINC/Rosetta distributed science jobs on it. Since the Celeron D was functional, I moved it to an old chassis & updated the power supply - both have efficient, after-market power supplies.

As a hobby (and part of what my Mother would call our inherited Scot's blood) I enjoy using an AC power meter to evaluate the cost of running appliances. My meter is from http://www.brandelectronics.com/ and it shows some interesting facts, such as that my Cox digital cable box consumes 24-watts when powered "ON" ... and 23-watts when turned "OFF" :-)

Obviously, the Core 2 Duo - running 2 jobs at once - contributes more credits to BONIC projects than the Celeron D. But I was interested in comparing what I gain given the monthly costs to run my now unnecessary Celeron D.

Computer Summary:

Core 2 Duo: 1.8GHz, 1GB DDR2-800 RAM, 320GB SATA drive, nVidia 7100 (fanless) 400w power supply

• Rosetta Benchmarks; fp=1744 int=3656 (since dual, means maybe fp=3488 int=7312)
• When Idle: CPU temp = 70 DegF, AC power usage = 105 watts
• When both cores at 100%: CPU temp = 100 DegF, AC power usage = 129 watts

Celeron D: 2.5GHz, 512KB PC2100 RAM, 30GB PATA drive, nVidia 6300 (fanless) 350w power supply (it had 1GB RAM, but 1-of-2 sticks went bad)

• Rosetta Benchmarks; fp=764 int=1677
• When Idle: CPU temp = 100 DegF, AC power usage = 98 watts
• When sole CPU at 100%: CPU temp = 125 DegF, AC power usage = 134 watts

I was at first pretty shocked that the Core 2 Duo - even with both CPU at 100% - used less total wattage than the Celeron D. Especially since every time you pick up a computer magazine there are dire warnings about needing a 600w, 800w, or even 1000w supply in a "modern" computer. By the way, a good AC power meter also tracks maximum power - which turns out in my case to be from 140 to 150 watts max when either the Core 2 Duo or Celeron systems first boot up.

Sonce both systems eat about the same power, just rounding the wattage to 130 watts burned 24-hours per day amounts to from $7.50 to $13.00 per month. This ranges includes my Minnesota kwh charges of about $0.08 per KWH and also my California charge of about $0.14 respectively. I wonder how many people understand they pay that much per month to run their computer 24-hours a day? Over a year that totals from $90 to $160 per computer - and this is JUST the computer. I'm not including the wattage used by monitors, printers, Ethernet switches or the DSL/cable router hardware. Plus with the computers running in a cool Minnesota basement, I don't have to include the extra air conditioning load they'd create in a hot climate like my Southern California home.

So now for the true "musing" - if I average the last 10 Rosetta jobs handled for each computer:

• Core 2 Duo: average 10594 seconds and 36.87 credits granted per job
• Celeron D: average 10406 seconds and 22.75 credits granted per job

However, since I'm looking where my $7.50 (or $13.00) per month goes I have to remember the Core 2 Duo runs 2 jobs at once for this same wattage so really one could say I am "paid" an average of 73.74 BOINC credits for each pair of 10600 second jobs that the Core 2 Duo runs. So the Core 2 Duo gives me almost 4 times the BOINC credits for the $100 spent a year on electricty to feed my hungry computer with both cores at 100% load 24-hours a day. Of course, even if the CPU throttled back to idle I'd still be paying about $80 per year to run the computer 24-hours per day.

So should I still run the Celeron D? Should I upgrade it to something closer to the Core 2 Duo? The upgrade cost me close to $450 once one considers the cost of the CPU, the new motherboard, and the new DDR2 RAM. This is an interesting question without a simple answer ... yes, running the old Celeron D doesn't cost me any more from a hardware stand-point ... but I am paying good money out of my pocket for the power.

So what is the real cost of power?

Beware of 4-pin fans (Intel or BTX fan)

I got a rather shocking education in new Intel technology. It seems to be something created for BTX motherboards. Let me step back a bit.

Most of us are familar with the 3-pin fans - 12vdc, RPM feedback, and ground. Motherboards adjust the fan speed by using pulse-width-modulation (PWM). In effect, they send pulses of 12vdc into the fan and the overall speed of the fan is related to the percentage of the time the fan sees 12vdc compared to no voltage. The problem with this idea is many fans "buzz" or vibrate slightly due to the jolts of power speeding up the coasting fan.

So apparently Intel came up with the 4-pin fan or "variable" fan. Instead of "punching" the fan with full 12vdc pulses, the 4th pin is used to send a low-voltage PWM and the fan internally uses this to set a variable speed from its minimum RPM at no pulses to its maximum RPM at a near 100% PWM. Sounds good - but when you plug this into a 3-pin plug it means the 4th pin has no pusles and if you're unluck the fan just stays at its minimum RPM. I learned this the hard way. It seems even the normal PWM of the 12vdc supply doesn't seem to speed up the fan. The fan manufacture offer this work-around: just short the unused 4th pin to the 2nd pin and the fan goes full blast jet-turbine level. Hmm, so it seems WORST case such a 4-pin Intel/BTX-style fan is either quiet but too slow, or super-fast and noisy.

So make sure you look at the photos or product carefully. Unless you have a mobo that supports such 4-pin fans, you don't want them. Online vendors make this a bit more confusing by often refering to fans connectors as 3/4 - which usually means it is a 3-pin connector with a 3-to-4 pin Molex-style power converter.

A64 Temperatures

Interesting differences:

The PcChips mobo which properly manages the Athlon64 temperature has a Zalman-clone "cyclone" style copper cooler. It runs at about 2200 RPM and is fairly quiet. The Athlon64 idles at between 75 and 79 degrees F. When pressed to high load it will be running in the 85-87 degree F range. In rare situations I've seen it hit the low 90's. Not bad for a modern CPU. The Speedfan utility can query both the mobo's view of CPU temperature, as well as the AMD K8 CPU offers direct PCI access into the CPU's view of it's own temperature. The direct K8 value tends to be within 2 degrees F higher or lower than the mobo. This gives me fairly good confidence in both values.

In contrast the Asus mobo with the same Athlon64 CPU always saw the temperature at 70 to 73 degree C - and was always running the CPU fan of the CoolMaster copper cooler at minimum RPM of 1400. Direct SpeedFan access to the K8 via PCI put the temperature in the 100 to 110 degree F range. This is a worrying difference - but given my room temperature was in the 72 to 74 degree F range it's rather hard to believe the CPU was really running COLDER than room temperature. So the Asus has to go - I RMAed it back to newegg.

I ordered as replacement an interesting Jetway mobo with an Nvidia chip-set and a mix of PCI, AGP anf PCI-Express slots. If it works with my NVidia 6800 AGP card, this mobo (with 2 PCI-Express slots) will make a good server that can still be in use 3 or 4 years from now.

AMD Athlon 64 3000+ (51-watt)

Summary: to upgrade yet reuse my existing RAM and AGP card, I upgraded to an old socket 754 AMD Athlon 64. Given total cost (with shipping) was under $100, I'd say result was successful.

You all know how computer "fixes" tend to creep and grow. This upgrade started because I decided my worker-bee system (acting as file server, DVR, and media player) was starting to stress its old Celeron CPU - even with a whiny stock cooler I'd seen the temperature hit 170+ degrees F. So I figured why not upgrade the AMD Athlon XP in my for-fun system and hand-me-down the XP CPU to be the worker-bee? The XP rarely hits 105 degree F.

Well, as much as I'd love moving to a spank'n new AMD X2 dual core (blah-blah-blah), that would mean $300+ for new DDR2 RAM, $150 for an AM2 motherboad, $200 for new PCI-express graphics card worth owning, not to mention the $100-300 for a good CPU. Plus with Intel's latest DuoCore Extreme beating AMD's X2 by 30% to 70% I suspect this next year is going to be an amazing technology race. Next summer the dual-core (or quad-core) CPU will be perhaps a few 100% faster and take less power.

So I started by looking through motherboards for something with an AGP slot (for my ATI Radeon X1600 AGPx8) and using DDR400 / PC3200 RAM. It turned out the "newest" motherboards to support this combination were the socket 754. This limited me to an AMD Athlon 64 - something one doesn't see for sale much. I finally found a nice $58 deal at
newegg (dot) com; adding an old S754 motherboard means the entire upgrade is in the $80 to $100 range.

Digging a little deeper, it turns out this particular A64 model is actually something in the Mobile Athlon family, so it burns a maximum of 51-watts, instead of the standard 89-watts of earlier Athlon 64 processors. This struck me as a fortunate "accident" since cooler means quieter and that is something I value.

Basic specs from AMD's web site: Details for this AMD Athlon 64

• Processor AMD Athlon™ 64, Model 3000+ (P/N ADA3000AIK4BX)
• Operating Mode: 32/64, Stepping E6
• Frequency: 2000Mhz, HT Speed: 1600
• Voltage: 1.40V Max Temp: 65°C
• Thermal Power: 51W (older models were 89w)
• L1 Cache: 128KB, L2 Cache: 512KB
• CMOS Technology 90nm SOI, Socket S754

So ... where is the $$ "creep" in project? Well, it started with a $29 copper after-market cooler since I wanted to leverage the value of my new low-watt CPU. After getting it running, this sweet little A64 idles at about 77 degree F and heats up to about 85 degrees F max when fully loaded! Upon power-up, it is satisyfing to hear the turbine-whine of the cooler drop to silence in a few seconds as the motherboard takes control of the cooler fan. In fact, for the first few days I had this urge to keep looking at the fan to make sure it wasn't faulty because it runs so quiet.

My worker-bee system is in a mini-ATX case and only had PC2100 RAM. So I started looking at min-ATX Socket A motherboards for my recylced XP ... long-store short I just decided to buy a second A64 with S754 mini-ATX mobo and another 1GB of PC3200 RAM. Since my worker-bee system runs 24/7 I liked the idea of the quieter fans. So I guess my initial $100 budget has expanded to 2 x $100 for 2 CPU and mobo, $60 for 2 nice after-market coolers, and $200 for 1GB of DDR400 RAM. At least I did suppress the desire to upgrade my for-fun system to 2GB RAM. Hopefully next summer I can re-upgrade the for-fun system to an dula-core AM2.