Slashdot Mirror
AMD 'Venice' Core Shows Big Drop in Power Needs
dtjohnson writes "
Lost Circuits has carefully
measured the power consumption of four recent Athlon 64 cores and
has found that power consumption has been dramatically reduced in the
new 'Venice'
core from the relatively-low (compared
to Intel P4) numbers of the original 2003 'Clawhammer' core to less
than 30 watts under load and less than 10 watts for Windows at
idle. This huge power reduction was apparently accomplished by a
combination of 90 nm die shrink, Silicon-on-Insulator
technology, and something called 'dual-stress
liner technology' As Lost Circuits points out, power
consumption worldwide has been exploding as more CPUs come online and
the CPU power requirements increase so a significant power reduction
will reduce the burden on electrical grids everywhere."
Now, these numbers were completely extrapolated from the key cracking rates I saw generated on my Athlon 1200, and estimates based on published power consumption. But it pointed out to me that these distributed contests are not good for us, and they're not free. It personally cost me about $40.00 / year in electricity. So, I don't play the distributed computing games any more.
John
Excellent!
This is quite a welcome change from the days of the old AMD chips that would tan you as you worked.
Looks like its time for Intel to spend a bit more time looking at power consumption.
hooray for competition!
Starsucks
That really is a big drop THat's what they should put in those computers for third-world countries >.>
Show this to your friends and family that don't know what a real hacker is
You are thinking Pre-2000. AMD has run cooler than Intel, cycle per cycle, for a while now. This is in addition to getting more work done per cycle. The days of AMD being used to toast marshmellows was a loooong time ago. I have both in servers. My dual Xeon server is a fine example of a great system that requires 6000btu of AC to keep a room at 72F. Great box, great cpu, unreal heat.
Tequila: It's not just for breakfast anymore!
... shouldn't it also reduce the heat produced by processors, therefore extending processor life?
:)
Or, for an overclocked machine, extending the amount of time it takes for the processor to die?
Eviscerati.Org: All Hail the Eviscerati
Lowering the power consumption per core is a first step to upping the number of cores. I imagine that CPU power consumption for desktops will level out in the 100 W range and makers will add cores, cache, and clock speed to maximize performance within a given power budget. I could also see some innovators creating new cooling technologies to boost the power budget and thus boost the permissible CPU performance within that expanded budget.
Two wrongs don't make a right, but three lefts do.
Do they still let users overclock their cpu's? I know intel locked thier CPU's. I wonder if AMD still lets people play with their products more.
Rosco: "If brains were gunpowder, Enos couldn't blow his nose."
By the way -- Sony sucks!
Gamingmuseum.com: Give your 3D accelerator a rest.
"I think the #1 problem AMD must overcome is the relationship Intel has with Microsoft. AMD makes clone chips, Intel makes chips that fit into Microsofts OS. Intel and Microsoft share information about how the chip will work with the software."
AMD came up with x86-64. Microsoft was only willing to support one 64-bit extension to x86, so that's what Intel chips use; they are the clones now. And Intel is the one with compatability problems (eg DMA is broken with Intel x86-64 chip, which seriously hurts performance).
I don't support one over the other. They trade performance and price/performance crowns regularly and I'll buy whoever's ahead this quarter. Just sayin' that AMD not "just a clone" anymore.
I rarely criticize things I don't care about.
In other words, cautiously we project the current power consumption of all computers running somewhere in the order of at least 20 Hoover Dam power plants
.04% of the annual US oil consumption.
If 9000Mw/hrs are the equivalent of 4 Hoover dams and current estimate is 20 Hoover dams, then current consumption by CPUs is around 45,000 Mw/hrs.
This site quotes 10.9 cubic meters of oil per megawatt/hour.
If my math and sources are right, then CPUs alone, worldwide consume the equivalent of nearly 500,000 cubic meters of oil each year.
According to this site, one American barrel of oil is 0.15899 cubic meters.
That means that the power consumption of all the CPUs in the world equate to over 3 million barrels of oil/year.
Perspective? The US currently uses a bit over 20 million barrels of oil/day. So CPUs worldwide are using around the equivalent of
You need to get a solar chart for your area of the world, and look up the equivalent insolation in terms of hours. Around here, we get an equivalent of 3.5 hours of maximum sunlight per day, averaged over the course of the year. Assuming your numbers are similar, you'll need about (24/3.5)*200 watts worth of solar panels -- that's 1370 watts. Assuming you get a great deal, you might pay $2.25 per watt, uninstalled cost, so that's over $3000 just for the panels. You'd also have to build a mounting system and possibly install a small motor to keep the panels pointed in the optimum direction.
On top of that, you need a battery system to provide power during hours of darkness. I could continue BS'ing the numbers to figure out how many batteries you'd need but would rather not. Needless to say, it's going to be several thousand dollars for the whole system.
(Yes, I've done this before)
Microsoft's marriage to Intel has broken up on a few fronts. Most importantly probably are the facts that MS's servers are now running AMD chips (look for the recent articles about Microsoft's switch to 64-bit), and that the next Xbox (Xbox360) will apparently be sporting PowerPC processor(s). The "Wintel" moniker is now mostly defunct. Maybe WMD... :-)
rooooar
The G5 (PowerPC 970) has similar specs. I haven't seen new wattage numbers on the new 2.7GHz models, but I imagine they are similar. The PPC970 draws about 40 watts as far as I know.
Here's a recently updated performance benchmark on the G5.
More important are power draws and BTU's for the entire system. I've done some comparisons between the Apple Xserve and competing Opteron/Xeon/Itanium2 systems for customers, and the Xserve is usually better at total power consumption and generates less heat. Our numbers are published here. Performance numbers between cross-platform code running on an Xserve and on a comparable dual-CPU system are usually competitive, depending on what the test is. The Opteron can certainly win on synthetic benchmarks that test memory bandwidth due to the memory architecture, but most people don't actually need that bandwidth.
When testing actual customer code we're usually the same or better in performance, with lower power draw and less heat generation. As always, your mileage may vary.
- "When you want something with all your heart, the entire universe conspires to give it to you" -Paulo Coelho
They check occasionally and cache the result, so you won't get scanned every time you post.
Pathscale, Sun Studio 10. Both are great (commercial) compilers. Only make sense for "scientific" code running on a bad-ass Opteron clusters, though, biggest benefit is the parallelization support. It's a somewhat different market too, gcc supports just about everything out there, which makes progress in some areas slow (The SSA stuff in gcc 4.0 helps, but it's just a foundation for cool stuff). And it's what
people develop open source software for, even those people that don't know that much about writing portable code, so anything != gcc is a hassle.
AMD is also working with the GCC people too (including engineer hours on actually improving the code, I believe), there was a recent post on comp.arch about this.
I look into this now and then (Nanosolar might be promising) and I will be using solar power during day time hours only to lower power usage. I won't bother with batteries- they seem to be a system cost killer.
If you hook it up to your basic power, you need a special electrical switchbox. But you could hook it up to some peltier cooling devices and some LED light fixtures perhaps. And you can pick up a couple 100 watt panels for 400 to 600 dollars and only power your computer with them during the day.
She was like chocolate when she drank... semi-sweet at first and then increasingly bitter.
The coal argument is often used...however it is always used wrongly....the 300 year projection is based on current usage...not usage in trying to replace oil and further population growth. In which case coal is only projected to last 90 years at most. Still a fair while, but not as rosy as 300 years. http://www.physicstoday.org/vol-57/iss-7/p47.html# cap4
Laptop Reviews
> Windows NT would NOT work with Cyrix, it kept locking up.
I owned a Cyrix based computer with NT 4 - it had no locking issues. Cyrix tried to run the PCI bus out of spec. If you weren't selective about your expansion cards they could cause the machine to lock. But Cyrix CPUs worked fine with NT4.
All generalizations are false, including this one. Mark Twain
I suspect a socket 939 laptop/notebook will be available by May 2006. This is when I'll be ready to buy. Why would they put a 939 in a laptop? To offer a version with a dual core part. What I would like is this venice core part, a 12-14 inch WSXGA daylight viewable screen, and sound with a quality 24bit 192kHz ADAC with line in and line out, firewire, wireless, optical drive, enough battery for a good 4 hours of full tilt CPU work, and a 3 year warranty. Of course a sub 5lb package with a digital I/O port, two of those mini-pci slots, and a tough case and keyboard would be nice too.
Who knows, in 12 months, there may be a couple to choose from. There should be enough market for portable scientific, industrial, and medical to justify building the things.
I'd like to make a low-power consumption system and I've found it extremely difficult to find information on the heat generation of system components. Neither NVidia or ATI mention power consumption outside of mobile chips anywhere on their websites. Look on the boxes of the video cards? Nada. Review sites? Very slim info, and what little is out there conflicts. After spending a week off an on scouring the web I eventually got a 6600GT which in several articles was praised for being lower power. However, it still requires an supplemental power plug and generates a ton of heat.
If you read the article page 7 & 11 (idle vs burn-in and conclusions), you'll see that the winchester (first 90nm core) dropped consumption in half compared to newcastle, and venice improves upon winchester by 10-15% by 1)lower consumption at idle and 2) better IPS throughput on equal mhz.
I just realised how few Intel chips I've ever had. My 8086 was an Intel, as was my 386. My 486 was an Cyrix, and it was a POS, and after that, I've had nothing but AMD. I moved on to a K5, then a K6, an Athlon-XP, and now an Athlon64.
Up to and including the K6, it was purely a cost issue. I never had any issues with the AMDs, despite the heat. You just need to cool them better, and not try to save a few pennies on that side of things. That's been true since the K5.
With regards to Cyrix, they never made a decent X86-compatible chip in the lifespan of the company, this much is true, but AMD are not the same.
When the Athlon-XP came out, it was no longer a cost issue. Accept no substititutes, it had to be an Authentic AMD for me. That was based upon their proven track record of good service. YMMV.
I'm typing this on a Athlon-64 box that is as reliable as anything I've ever owned, and is currently running with a core temperature of 45C.
I could make that lower, but I'd need to turn my fans up. That temp is under load, mind you. Prescott cores run way hotter than that, I'm pretty sure.
Sign the FSF's Anti-DMCA petit
If AMD had some brains they would hire a few engineers to submit optimization patches to gcc for AMD processors.
They did - they paid SuSE to do the original work and some performance work. I'm not sure if that's still ongoing, though.
AMD64 is definitely on the GCC radar - it's now in the list of primary release platforms and they're taking AMD64 performance seriously for future versions. But it's slow progress and ICC has a big lead.