Actually, Cadence most definitely did not start the way you describe, from a PhD candidate's original research. Cadence then, as now, was built by merger. The original merger was ECAD and SDA, followed shortly by the acquisition of Gateway Design (Verilog-XL), HLDS, CCT, Ambit, Quickturn, Simplex, Get2Chip, etc.
As a cynical EDA user, I would tell you that Cadence very very rarely (if ever) develops something from internal R&D.
An addendum to this: many geeks probably don't realize how often people lie in real life. When you poker, you are forced to lie (if you want to be successful in the long haul), and try to puzzle out when other people are lying.
If you think it would be _trivial_ to write a client, you're dead wrong. Go to U of Alberta website and read some of the papers, then tell me how easy it would be. I have been playing poker pretty seriously for 5+ years now, before the wonderful surge in popularity, and have plenty of CS background. It's not going to be as easy as everyone thinks.
You comments on BB/hour are interesting though, my understanding is that most bots can barely win at the micro-limits (0.05-0.10) much less a 2-4/3-6 game. I suspect they would get killed at 25-50 (although the 20-40 I played in Barcelona played essentially like a Bay Area 3-6 or 6-12, simply amazing stuff). The nice thing about bots is that they can be deployed en masse -- if you could actually beat a 2-4 game for 2BB/hour, and you ran a dozen of them, you're making about $100/hour, 24 hours a day. that's pretty nice.
Lastly, the scariest thing in the whole article is the guy who sells software to enable team play. That very threat is why I don't go near internet poker. It'd be too easy for me and 3 of my friends to sit at the same table and have an irc client open on the side sharing hand info.
Can you imagine knowing 4 sets of hole cards, and probably always having one player at either UTG/+1 and one player at button/cutoff?
Hmm... I assume you and I have the same car, mine is an '02 with quattro. Across 31630 miles I am averaging 23.61 miles, according to the palm pilot mileage tracker applet I use. It looks like the highest per-tank mpg was around 28, and the low around 21 (hurrah, EPA estimate). I do an 80-mile roundtrip commute each day which is about 75+ hwy and 5 pretty grueling city (in and out of San Francisco). I also average 80mph on the highway, which might explain some of it....
If you care about floating-point performance, regardless of how much I dislike the boxes, Itanium has the best SPECfp results (by far) -- look at the SPEC website if you want to see what I mean.
Opteron has the best overall integer performance at the moment, and probably the best price/performance ratio. I believe it will thrash a Xeon on integer or FP, due to the doubling of the internal register file for both. But for max-floating-point ops per second, Itanium is still king. It's true for hspice in my line of work (which mostly consists of FP matrix math), it should hold true for you too.
I would try to get a set of eval systems from someone like HP, if I was you. They sell all 3 architectures -- tell them you will buy whichever one runs your research app the fastest and they will probably be willing to do the demo (or could get you some time ssh'd into one to at least try it out on your code).
I don't know what's worse, that I knew you did your math wrong because I immediately recognized 1048576 as 2^20, or that I intuitively knew you typo'd the price as $32 instead of $22.
So let's say I have a massive computing need for, oh say, chip design. Chip designers, like Texas Instruments, Cirrus Logic, or General Semiconductor, require massive amounts of CPU time and even more memory.
Almost. Speaking as a chip designer (7 years designing ASICs and building the HW clusters to do so), where Sun is getting killed is 2 places: Absolute CPU performance, and price/performance.
For EDA (chip design) applications, SPARC sucks, big-time. Opteron 248s and Xeon 3.2's (SPECint=1500 and 1300, respectively) destroy 1.2GHz SPARCs (SPECint=~700). They also cost between 1/2 and 1/5 of the price, depending on config.
Sun used to have blow-the-doors-off good I/O performance compared to PC's (back in the SBus days, when PC's were ISA). Now they use the same PCI as the PCs, and have roughly similar performance. Granted, the PC should probably have a SysKonnect GE card to drive the wire hard, but whatever, it's still cheap compared to Sun GigE cards.
The big Fire V480/V880 can hold plenty of memory -- that's not the problem -- but it's a waste because the CPU is dog-slow, so your job still takes forever to finish. It used to be the only choice though, if you wanted 64-bit support for the tools. (If you have the choice between a job that takes 4 days, or one that crashes due to out-of-memory after 2, you can guess what most people pick...)
I'm really looking forward to Opteron machines that hold more than 16GB. At that point, stick a fork in Sun for chip design use, they're done.
Except "accelerating through a yellow" tickets are double (or more) of "running a red" tickets in California. In San Francisco (where I live), there are tons of red-light cameras already, and the fines are steep (for the very reasons you mentioned). Equipping them with radar is probably the next step for SF-DPT anyway, since tickets are the primary source of revenue for the city.
You clearly are not from california: When it rains here, everyone slows to 3mph regardless of traffic conditions, rain volume, visibility or any other factor. You'd think there was a foot of snow on the road.
Or study EE/CompEng (for example at Stanford you can major in EE with sub-specialization of computer software, or major in CS with sub-specialization of computer hardware -- they are very similar in terms of courses but the theory-classes you take will be different).
Also, at the grad level, consider programs in Distributed Systems in addition to networking.
ObPlug: Stanford has some excellent higher-division CS and EE classes in computer networks, and distributed systems. Stanford's MS programs are heavily professional-oriented (1 year, coursework-only, no thesis) and worth considering post-undergrad -- they are very hard to get into though. If your employer is an SITN member you can take the classes via distance learning, without having to be admitted to a full-time degree program.
As a person who plays poker a LOT, let it be mentioned that bluffing has a specific value: it works in conjunction with the times you actually have a good hand and win the pot. If you almost always raise when you have, say, a pair of aces, you can get some people to fold when you raise with garbage. If you can get everyone to fold, then you were successful. But if even _one_ person disbelieves you and calls your bluff, you're screwed.
"Naked" bluffing, where you don't have a hand history to back it up and make it convincing, that's a quick way to lose a lot of money. Go figure, it's SCO's strategy.
Alright I need to call this out now. These Xserves are not really all that impressive (yeah it's great that apple builds 1U hardware but for god sakes it's the usual overpriced crap).
I just priced your $3000 machine for real. The $3K machine has a restriction of 10 simultaneous users. Eliminating that (hoorah for non-free-software!) bumps up the price to $5750 for this:
Dual 2GHZ PPC G5 (512kL2, 1GHz bus, etc) 2GB DDR400 ECC SDRAM (2x1) 80GB SATA Dual GigE Applecare Premium (which gives next-day support outside of business hours and 4-hour during business hours) Unlimited-users OSX licensing
Compare that to a Dell 1750 for $4796, priced yesterday on Dell.com: Dual 3.2GHz P4 Xeon (1MB L2, 533 FSB) 2GB DDR266 SDRAM (2x1) 73GB 15K RPM Ultra320 (a better drive!) Dual GigE 3yr 4-hr onsite support, 24x7 (add 500 for S/W support but good luck) No OS (Get an unlimited-users linux Distro, dude!)
Yes, Virginia, Apple's fancy slideware shows that they linpack 9.0 GFLOP versus 8.2 GFLOP (for a Dell 1750 2x3.2 of all things!). But if it's integer perf you care about, the 970 (G5) should SPECint around 1160 (since no one provides the data except IBM @ 1.8GHz) -- compare with 3.2GHz P4-533 Xeon SPECint of 1220-1283.
Wow, marginally worse performance for an extra grand per machine -- what a deal!
This is pretty much equally true in Silicon Valley. Our startup did some layoffs in the ASIC team -- they all had jobs in less than 2 weeks (GOOD jobs too). As the poster said, RTL'ers are dime-a-dozen but the hard-science stuff still pays. Also, chip verification is a good area, since it's dirty work that is hard to do really really well.
And if you really want to do CS, just do systems stuff instead of HCI or GUI design. Companies still are hiring skilled developers for device drivers or embedded systems. I would recommend against focusing all your studies on Web-stuff....
Sony VAIO Z505R. It's about 4 years old so only a Pentium II 366, replaced the HD with an IBM 30GB 2.5" for about 100 bucks, quieter too. 192MB memory, 12inch screen, about 4.5 pounds with the triple battery. No builtin wireless but one PCMCIA slot for your flavor of choice (a/b/g). I love it. It's perfect for xterms/emacs/mozilla/xmms.
Might I recommend ebay to you? (either the R, RX, S or SX will probably fit the bill, the J or H models I think are P3-500ish procs). Expect to need to replace the drive, they were originally in the 6-10GB size.
"I look forward with much anticipation to what IBM does with this chip in the next 12-18 months"
From following these processor wars for the better part of a decade I have learned one thing well: Never, ever, ever bet against Intel in that kind of timeframe. You do realize that 4GHz+ "P5" chips should be available by then? (probably 3.6GHz in your 12 month timeframe, and as high as 5GHz on the 18 month end)
If AMD manages to put much pressure on Intel via the Athlon64 this fall, expect Intel to turn up the wick sooner rather than later. That is the beauty of the x86 market, real knock-down drag-out competition. The customer really does benefit.
Slashdot Mirror
Actually, Cadence most definitely did not start the way you describe, from a PhD candidate's original research. Cadence then, as now, was built by merger. The original merger was ECAD and SDA, followed shortly by the acquisition of Gateway Design (Verilog-XL), HLDS, CCT, Ambit, Quickturn, Simplex, Get2Chip, etc.
As a cynical EDA user, I would tell you that Cadence very very rarely (if ever) develops something from internal R&D.
An addendum to this: many geeks probably don't realize how often people lie in real life. When you poker, you are forced to lie (if you want to be successful in the long haul), and try to puzzle out when other people are lying.
That turns out to be a pretty valuable skill.
what's up ddt! drw from karl's linuxnet....
good luck with the acting.
and for the record, Abuse was just waaaay ahead of its time.
That's just about right.
If you think it would be _trivial_ to write a client, you're dead wrong. Go to U of Alberta website and read some of the papers, then tell me how easy it would be. I have been playing poker pretty seriously for 5+ years now, before the wonderful surge in popularity, and have plenty of CS background. It's not going to be as easy as everyone thinks.
You comments on BB/hour are interesting though, my understanding is that most bots can barely win at the micro-limits (0.05-0.10) much less a 2-4/3-6 game. I suspect they would get killed at 25-50 (although the 20-40 I played in Barcelona played essentially like a Bay Area 3-6 or 6-12, simply amazing stuff). The nice thing about bots is that they can be deployed en masse -- if you could actually beat a 2-4 game for 2BB/hour, and you ran a dozen of them, you're making about $100/hour, 24 hours a day. that's pretty nice.
Lastly, the scariest thing in the whole article is the guy who sells software to enable team play. That very threat is why I don't go near internet poker. It'd be too easy for me and 3 of my friends to sit at the same table and have an irc client open on the side sharing hand info.
Can you imagine knowing 4 sets of hole cards, and probably always having one player at either UTG/+1 and one player at button/cutoff?
ouch ouch ouch.
Not if the next 15 lights in a row are timed (like they are in SF)....
Or perhaps the last to make the green (or yellow)....
Hmm... I assume you and I have the same car, mine is an '02 with quattro. Across 31630 miles I am averaging 23.61 miles, according to the palm pilot mileage tracker applet I use. It looks like the highest per-tank mpg was around 28, and the low around 21 (hurrah, EPA estimate). I do an 80-mile roundtrip commute each day which is about 75+ hwy and 5 pretty grueling city (in and out of San Francisco).
I also average 80mph on the highway, which might explain some of it....
What about the babes?!
(I love yahoo's sense of humor with the URL, btw -- "glamour"?!)
Top speed is all about gearing, plain and simple. I'm sure with a reasonably tall gear package they could hit absurd top speeds in an F1 car.
If you care about floating-point performance, regardless of how much I dislike the boxes, Itanium has the best SPECfp results (by far) -- look at the SPEC website
if you want to see what I mean.
Opteron has the best overall integer performance at the moment, and probably the best price/performance ratio. I believe it will thrash a Xeon on integer or FP, due to the doubling of the internal register file for both. But for max-floating-point ops per second, Itanium is still king. It's true for hspice in my line of work (which mostly consists of FP matrix math), it should hold true for you too.
I would try to get a set of eval systems from someone like HP, if I was you. They sell all 3 architectures -- tell them you will buy whichever one runs your research app the fastest and they will probably be willing to do the demo (or could get you some time ssh'd into one to at least try it out on your code).
-c
I don't know what's worse, that I knew you did your math wrong because I immediately recognized 1048576 as 2^20, or that I intuitively knew you typo'd the price as $32 instead of $22.
I love binary.
Almost. Speaking as a chip designer (7 years designing ASICs and building the HW clusters to do so), where Sun is getting killed is 2 places: Absolute CPU performance, and price/performance.
For EDA (chip design) applications, SPARC sucks, big-time. Opteron 248s and Xeon 3.2's (SPECint=1500 and 1300, respectively) destroy 1.2GHz SPARCs (SPECint=~700). They also cost between 1/2 and 1/5 of the price, depending on config.
Sun used to have blow-the-doors-off good I/O performance compared to PC's (back in the SBus days, when PC's were ISA). Now they use the same PCI as the PCs, and have roughly similar performance. Granted, the PC should probably have a SysKonnect GE card to drive the wire hard, but whatever, it's still cheap compared to Sun GigE cards.
The big Fire V480/V880 can hold plenty of memory -- that's not the problem -- but it's a waste because the CPU is dog-slow, so your job still takes forever to finish. It used to be the only choice though, if you wanted 64-bit support for the tools. (If you have the choice between a job that takes 4 days, or one that crashes due to out-of-memory after 2, you can guess what most people pick...)
I'm really looking forward to Opteron machines that hold more than 16GB. At that point, stick a fork in Sun for chip design use, they're done.
you are totally correct. it's pretty sad how little sense of history most tech journalists have these days.
"taped out" means adhesive tape, not magnetic.
Except "accelerating through a yellow" tickets are double (or more) of "running a red" tickets in California. In San Francisco (where I live), there are tons of red-light cameras already, and the fines are steep (for the very reasons you mentioned). Equipping them with radar is probably the next step for SF-DPT anyway, since tickets are the primary source of revenue for the city.
You clearly are not from california:
When it rains here, everyone slows to 3mph regardless of traffic conditions, rain volume, visibility or any other factor. You'd think there was a foot of snow on the road.
Mod parent up, grandparent down! He's exactly correct.
Absolutely!!!
Or study EE/CompEng (for example at Stanford you can major in EE with sub-specialization of computer software, or major in CS with sub-specialization of computer hardware -- they are very similar in terms of courses but the theory-classes you take will be different).
Also, at the grad level, consider programs in Distributed Systems in addition to networking.
ObPlug: Stanford has some excellent higher-division CS and EE classes in computer networks, and distributed systems. Stanford's MS programs are heavily professional-oriented (1 year, coursework-only, no thesis) and worth considering post-undergrad -- they are very hard to get into though. If your employer is an SITN member you can take the classes via distance learning, without having to be admitted to a full-time degree program.
As a person who plays poker a LOT, let it be mentioned that bluffing has a specific value: it works in conjunction with the times you actually have a good hand and win the pot. If you almost always raise when you have, say, a pair of aces, you can get some people to fold when you raise with garbage. If you can get everyone to fold, then you were successful. But if even _one_ person disbelieves you and calls your bluff, you're screwed.
"Naked" bluffing, where you don't have a hand history to back it up and make it convincing, that's a quick way to lose a lot of money.
Go figure, it's SCO's strategy.
Thank you for reading my post.
Alright I need to call this out now. These Xserves are not really all that impressive (yeah it's great that apple builds 1U hardware but for god sakes it's the usual overpriced crap).
I just priced your $3000 machine for real. The $3K machine has a restriction of 10 simultaneous users. Eliminating that (hoorah for non-free-software!) bumps up the price to $5750 for this:
Dual 2GHZ PPC G5 (512kL2, 1GHz bus, etc)
2GB DDR400 ECC SDRAM (2x1)
80GB SATA
Dual GigE
Applecare Premium (which gives next-day support outside of business hours and 4-hour during business hours)
Unlimited-users OSX licensing
Compare that to a Dell 1750 for $4796, priced yesterday on Dell.com:
Dual 3.2GHz P4 Xeon (1MB L2, 533 FSB)
2GB DDR266 SDRAM (2x1)
73GB 15K RPM Ultra320 (a better drive!)
Dual GigE
3yr 4-hr onsite support, 24x7 (add 500 for S/W support but good luck)
No OS (Get an unlimited-users linux Distro, dude!)
Yes, Virginia, Apple's fancy slideware shows that they linpack 9.0 GFLOP versus 8.2 GFLOP (for a Dell 1750 2x3.2 of all things!). But if it's integer perf you care about, the 970 (G5) should SPECint around 1160 (since no one provides the data except IBM @ 1.8GHz) -- compare with 3.2GHz P4-533 Xeon SPECint of 1220-1283.
Wow, marginally worse performance for an extra grand per machine -- what a deal!
(yes I hate Apple fanboys, sorry)
This is pretty much equally true in Silicon Valley. Our startup did some layoffs in the ASIC team -- they all had jobs in less than 2 weeks (GOOD jobs too). As the poster said, RTL'ers are dime-a-dozen but the hard-science stuff still pays. Also, chip verification is a good area, since it's dirty work that is hard to do really really well.
And if you really want to do CS, just do systems stuff instead of HCI or GUI design. Companies still are hiring skilled developers for device drivers or embedded systems. I would recommend against focusing all your studies on Web-stuff....
I need to put something in here to confound the lameness filter, so this will have to do.
Sorry, I really, really needed to vent.
Hey, it's my laptop!
Sony VAIO Z505R. It's about 4 years old so only a Pentium II 366, replaced the HD with an IBM 30GB 2.5" for about 100 bucks, quieter too. 192MB memory, 12inch screen, about 4.5 pounds with the triple battery. No builtin wireless but one PCMCIA slot for your flavor of choice (a/b/g). I love it. It's perfect for xterms/emacs/mozilla/xmms.
Might I recommend ebay to you? (either the R, RX, S or SX will probably fit the bill, the J or H models I think are P3-500ish procs). Expect to need to replace the drive, they were originally in the 6-10GB size.
"I look forward with much anticipation to what IBM does with this chip in the next 12-18 months"
From following these processor wars for the better part of a decade I have learned one thing well: Never, ever, ever bet against Intel in that kind of timeframe. You do realize that 4GHz+ "P5" chips should be available by then? (probably 3.6GHz in your 12 month timeframe, and as high as 5GHz on the 18 month end)
If AMD manages to put much pressure on Intel via the Athlon64 this fall, expect Intel to turn up the wick sooner rather than later. That is the beauty of the x86 market, real knock-down drag-out competition. The customer really does benefit.