Author: Robert Hyatt
Date: 15:11:25 02/10/00
Go up one level in this thread
On February 10, 2000 at 17:03:30, Albert Silver wrote: >On February 10, 2000 at 16:50:10, Robert Hyatt wrote: > >>On February 10, 2000 at 14:35:24, Peter Kappler wrote: >> >>>On February 10, 2000 at 10:52:38, Ed Schröder wrote: >>> >>>>On February 10, 2000 at 10:13:16, Thorsten Czub wrote: >>>> >>>>>On February 10, 2000 at 01:04:50, Lonnie Cook wrote: >>>>> >>>>>>On February 09, 2000 at 23:30:14, Thorsten Czub wrote: >>>>>> >>>>>>>On February 09, 2000 at 23:00:35, Tom Kerrigan wrote: >>>>>>> >>>>>>>>Integrated = on-die. So it should be running at the same speed as the core. But >>>>>>>>the design of the L2 cache may have increased latencies, so it's possible that >>>>>>>>it won't be as fast as some people think it "should" be. >>>>>>>> >>>>>>>>-Tom >>>>>>> >>>>>>>hm. >>>>>>>whatever - could be a killer-cpu for cstal. >>>>>> >>>>>>I have run CSTal a little Thorsten and I was getting 50-60K on mine >>>>> >>>>>on 1000 ? >>>>> >>>>>that would be 3.3 - 4 times faster than on my k6-3/400 ! >>>>>nice nice nice. >>>> >>>>That's because the KryoTech 1000 Mhz L2-cache runs at 400 Mhz. I hope >>>>my new toy will arrive tomorrow :-) >>>> >>>>Ed >>> >>> >>>Hi Ed, >>> >>>When will we get a chance to see this machine in action in a GM Challenge game? >>> >>>--Peter >> >> >>I would hope "never". This is not exactly the most reliable thing to do to >>hardware. Remember that for every system hang you see, there were hundreds of >>thousands of cases where an instruction produced the wrong result but was in >>"user-land" where it didn't crash anything. >> >>It is incredibly risky to overclock when 'it counts'. > >Even at -40 degrees? Furthermore, I remember reading at Tom's Hardware that one >could adapt the system to use other chips, so it isn't limited the chip it comes >with. If one used a chip that was naturally faster, wouldn't the dangers of >overclocking be gradually minimized, especially at that temperature? > > Albert Silver The basic clock frequency for a chip comes from adding up gate delays in a path from X to Y inside the chip. If you go too fast, you latch data before it settles. If you go too slow, you just go too slow. If you ramp up the clock beyond spec, you _must_ do something to speed up the gate delays. One way is to up Vcc. That ups the heat. Which kills the chip lifetime. You can do various tricks to conduct the heat away (ie kryo approach). But no matter what you do, you can't make the gates function much faster. In some cases the vendor intentionally underclocks the chip, so that they can lower the voltage to avoid a serious heating problem. In that case, you can run the clock back up if you are willing to spend the money to cool it properly. In other cases (ie L2 cache) you simply can't push it any faster no matter what you do. There are several variables, including temperature, Vcc, manufacturing process, gate delays, clock frequency, etc. Running the clock frequency up without addressing anything else is asking for trouble. Running the clock frequency up, running the voltage up, and dropping the temp via a kryotech cooling device isn't a sure thing either. As there is definitely a frequency beyond which the best ICs of today simply won't function. We still have to overcome resistance, capacitance and inductance, each of which adds heat or delay to the thing. I had a student here that was running an AMD and was getting odd results on rare occasions, while it produced perfect answers on an Intel box. He wrote a pretty good cpu diagnostic in assembly, which tried to test every instruction multiple times with various types of data (ie all 1's, alternating 1/0's, sliding 1's, etc.) He found that his slight overclocking was simply bogus. His machine "seemed" stable, but only because an instruction has to screw up in the O/S to crash things. A bogus operation in a program might or might not cause it to fail. I don't like the risk, personally. Ed had problems with his kryo machine. Lonnie has had hangs with his. I prefer "rock-solid" myself...
This page took 0.04 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.