There is a company out there that has an FPGA in a 940 pin socket. What about putting a GPU in it? Dual channel memory, HT link to the main processor, HT link to a DAC from the GPU [make mobos with fixed DACs on the board].
It takes one [public] violator and boom no more blogs. It's that way with many "benefits".
A friend of mine works at Broadcom and he mentioned how they used to spring for supper if you worked late. Then people started working till 6pm to get the supper and then take off. End result, no more free suppers. Even though the majority of the people taking the free meals were in fact working they still took the benefit away.
It's best not to indiscriminantly talk about your employers past or present without some thought towards how what you are writing could be perceived...
Generally anything that is physically safe and prevents you from accomplishing your goal would be something acceptable in my books. The griefers on online games I'm talking about usually sacrifice the goals of the game to ensure that you can't participate.
I guess you can grief in the real world except there people tend to frown on the behaviour. And since the griefer can see the negative attention they can be less likely to continue it.
Consider the pool you're playing in. Most online RPG players spend considerable time in their "investments" [which in and of itself isn't "bad"] and forgo the usual sociallizing that most others take part in.
So you're basically playing with a bunch of anti-socials, who are almost certainly young and really aren't playing the game to relax or have fun. They're on a mission. So of course they'll grief and cheat to get ahead. They think they're accomplishing something when all along it's just supposed to be a minor diversion.
I stopped caring about online games after I played SOCOM for the first time. Playing against people who would only target you even though you had 8 other people on your team and would hunt you down even if it didn't help accomplish the mission was just annoying. Specially since you know for a fact the kid plays 8 hours a day and didn't have college [and a job] to go to. I just wanted to have fun and a reasonable chance of participating and instead I was targetting because I was newbie and never given a chance to really play. Basically each round I would be alive for 30 seconds then have to wait 10 mins for the next round. After a few hours I got so frustrated that I never signed on again.
Might sound like "whining" but many people share the same experience because it's not that hard to really make the game anti-fun. All it takes is a handful of pricks who don't get why the game exists and want to ruin it for anyone else as a sign of their l33t skillz.
Sarcasm aside I was trying to give you friendly advice. Believe it or not but companies like IBM hire firms to scour the net for their brand. They report potential violations and what not. You may not want to work with IBM in the future, but your company [assuming you're not self-employed] may want to. Or, god forbid, you may actually want to take IBM up on an offer in the future as a contractor, etc, etc....
The best thing to do is just not speak as if you know the intentions and motives of the company and/or its employees. You can talk about your work and your company but you have to follow various rules [both legal and H.R. related].
But, if you really want to keep speaking as an agent of IBM. Someday you may cross the line and get some nice legalese in the mail.
You're not under an NDA means you're not privy to any NEW information. It doesn't mean you can all of a sudden disclose everything you learned previously. If you gave out the secret sauce for Tivoli you can rest assured IBM would be all over you for that.
Just saying it's not wise to speak on their behalf. Saying things like
"It was my personal impression that people didn't care"
is generally better than
"It was IBMs position that nobody cared"
You have to disclaim that you're not speaking on behalf of the company, even if you're an ex-employee. [Hint: I learned this lesson a couple of weeks into working for my current employer. Even if what you say sounds innocent they will still jump all over it.]
As a former IBM employee, you should know better than to talk about internal politics and what not like that. Heck, I work at IBM as a vendor and even I'm closely scrutinized for what I say [not that I really have anything substantive to say publically in any case].
That aside, most employees of most companies are not really fully aware of the legal ramblings they're involved in. Out of site out of mind...
My god....mon dieu!! etc... 158$ must turn into what? 500$ retail? Have fun losing that phone.
Of course it'd be nice if cell companies both offered this monstrocity of a money pit and the el-cheapo phones that companies like Moto make as well. You know, that whole "free market" thingy...
Well it could be the gentoo/xorg combo but afaik I'm still locked into the xorg-nv driver right now.
Which is ok because I'm too busy writing my book to miss the 3D accel.:-( (that and I have consoles for gaming).
Mostly I like my cards (NV43 iirc) because they were decently priced, work over PCI-E (a must for PCI TV tuner owners), don't need fans and provide good performance. Most of the time I'm in 2D mode anyways so it doesn't matter. But accelerated bitblts are nice and handy (specially for overlays) so I need more than a dumb framebuffer.
The problem is most architectures are fairly locked down. For instance, putting another MUL engine in would be nice except MUL only reads from two registers and only outputs to two registers. A fourth pipe would raise retire latency, etc, etc, etc. In the AMD world at least most common ALU opcodes are already directpath. Larger caches could help but they raise the latency, so would raising the set-associtivity. More TLB entries are always welcomed, etc...
As an employee of one of the chip companies I couldn't really go into any other ideas without treading on NDA boundaries. Lets just say both camps have uses for more transistors. Don't worry:-)
4x4... You mean a pair of 285s and a tyan mobo? Aside from the lack of dual SLI that's basically a 4x4 setup.
If you actually need CPU power you can already buy Opteron gear today. Also while I wouldn't mind playing with a 4-core Intel kit I still love my 2x285 kit.
The problem is the hardware is always in a state of flux and just incremental improvements. Your GeForce 7800 is probably based on the same HDL source as the 6600 with appropriate changes. This means that legacy symbol names from one project creep up into the new space. You get odd names, combined with lack of comments and documentation [compliance] leads to hardware with "oops" that the drivers have to work around.
Things could vastly improve for the customer if they stopped pretending that they know best. I know for a fact that companies like ATI and Nvidia spend a good deal of time [re: cost of the video card] in DRM technologies. Basically they don't give two shits about you as a customer so long as you
a) feel inadequate with your 75W GPU and buy the next best thing next quarter
b) fully comply with their "dominance" of your machine, force you to run windows, force you to use their bloaty drivers, use their drm, etc
I tolerate Nvidia solely because their kernel modules work decently [well not anymore as they're not keeping pace with xorg development]. Opening the 2D and 3D accelerators to the public can only serve to make the hardware more popular.
Their value is in the hardware and the ability to develiver it. Not the interface that puts a triangle on the screen.
The problem is manufacturing not so much the design.
Suppose they got the HDL for a super-duper 16 pipelined vertex shading GPU today. Could they still roll them out to end users for less than $500 a pop? Probably not.
As another poster mentioned the IA64 is not a general purpose processor with decent pipeline control, register renaming, branch prediction, etc.
This means your typical random access applications [e.g. desktop] do not behave as well.
IA64 is really meant for the case where you can handtune that 5% of the code that takes 95% of the time in assembler. Who cares if your GUI is inefficient if you're pulling 4x the FLOPS as an Opteron in your HPC application? However, for a web server where you are doing less simd and more branching, loops, small ops, the Opteron arch [well and Conroe] helps.
It's like the traditional Transmeta vs. Intel/AMD. Intel and AMD spend a lot of hardware on runtime optimizations [e.g. prefetching, branch prediction, CISC => RISC decoding, register renaming, etc] so that it can cope with the current runtime situation [e.g. workload] whereas Transmeta optimized your task more statically [with less forethought to branches, cache, etc]. The net result was you could get some performance from Transmeta but the MIPS/MHz just wasn't there. Similarly IA64 *can* really fly if you hand tune your application AND it can take advantage of the architecture.
It isn't just a matter of having a good compiler [which admitedly GCC isn't there] but having a good problem.
24MiB of 6T SRAM is 1,207,959,552 transistors just for the cells. Not including tags, address decoders, etc...
By comparison, an Opteron uses 113,246,208 transistors for the 2048+256KB of cache [assuming they use a 6T structure which I don't know for a fact since I'm not privy to the details and technically I couldn't say even if I were, so don't assume what I said is verbatim fact, yada, I hate disclaimers] and the 4MB Duo (total of 4096+128KB of cache) needs 207,618,048 transistors for its cache.
Even with a sufficient market, right now that chip is going to be expensive. It's huge and the yield can't be that high.
But that's kinda the problem. The cost and the fact that's its a very niche processor will never make it as common place as Xeon or Opteron processors. Specially when both Intel and AMD are pushing SSE you can get a lot of the parallel SIMD like benefits of IA64 on x86.
The problem with the IA-64 platform is there are so few applications for which it is well suited. And even for problems it IS well suited it's a matter of figuring out how to extract the desired performance.
That said, a well tuned IA-64 application can smoke the best offerings from x86 world [on both sides of the fence]. But a $3700 USD price tag may push people away. Specially since processors like the Opteron 285 are nearly half the price and way more flexible.:-)
I'm not tooting my own horn. I'm saying certification is meaningless and that people will use ANYTHING so long as it meets their needs and budget. My code happens to be free and apparently they don't care that I haven't forked over the required 10 grand or so PER RELEASE to get it verified [per platform too btw].
Yeah a shared L2 is nice and all. Talk to me when you work in cluster computing though. Or have to run more than WinXP and your fav FPS.
And technically HT == hypertransport. HTT == hyper threading technology. Intel has never had HT [or similar] technology in x86 lines.
The types of tasks I use my box for don't really bode well for the whole "let's have our cores share access to a L2 thereby creating effectively a freaky fast FSB with contention issues".
I like my smaller but independent 1MB L2s on my Opterons because it lets me run tasks [independent tasks I may add] full steam ahead. And because my 2P box has two memory controllers I can get a shitload more bandwidth than any 2P Conroe could ever dream of.
Like I said, there are things one camp is good for, and things the other is. To unilaterally state that Conroe will whip AMDs butt is just a load of ignorant baloney.
Slashdot Mirror
There is a company out there that has an FPGA in a 940 pin socket. What about putting a GPU in it? Dual channel memory, HT link to the main processor, HT link to a DAC from the GPU [make mobos with fixed DACs on the board].
That'd be hella cool.
Tom
It takes one [public] violator and boom no more blogs. It's that way with many "benefits".
A friend of mine works at Broadcom and he mentioned how they used to spring for supper if you worked late. Then people started working till 6pm to get the supper and then take off. End result, no more free suppers. Even though the majority of the people taking the free meals were in fact working they still took the benefit away.
It's best not to indiscriminantly talk about your employers past or present without some thought towards how what you are writing could be perceived...
Tom
Generally anything that is physically safe and prevents you from accomplishing your goal would be something acceptable in my books. The griefers on online games I'm talking about usually sacrifice the goals of the game to ensure that you can't participate.
I guess you can grief in the real world except there people tend to frown on the behaviour. And since the griefer can see the negative attention they can be less likely to continue it.
Tom
This will get modded down but ...
Consider the pool you're playing in. Most online RPG players spend considerable time in their "investments" [which in and of itself isn't "bad"] and forgo the usual sociallizing that most others take part in.
So you're basically playing with a bunch of anti-socials, who are almost certainly young and really aren't playing the game to relax or have fun. They're on a mission. So of course they'll grief and cheat to get ahead. They think they're accomplishing something when all along it's just supposed to be a minor diversion.
I stopped caring about online games after I played SOCOM for the first time. Playing against people who would only target you even though you had 8 other people on your team and would hunt you down even if it didn't help accomplish the mission was just annoying. Specially since you know for a fact the kid plays 8 hours a day and didn't have college [and a job] to go to. I just wanted to have fun and a reasonable chance of participating and instead I was targetting because I was newbie and never given a chance to really play. Basically each round I would be alive for 30 seconds then have to wait 10 mins for the next round. After a few hours I got so frustrated that I never signed on again.
Might sound like "whining" but many people share the same experience because it's not that hard to really make the game anti-fun. All it takes is a handful of pricks who don't get why the game exists and want to ruin it for anyone else as a sign of their l33t skillz.
Tom
Sarcasm aside I was trying to give you friendly advice. Believe it or not but companies like IBM hire firms to scour the net for their brand. They report potential violations and what not. You may not want to work with IBM in the future, but your company [assuming you're not self-employed] may want to. Or, god forbid, you may actually want to take IBM up on an offer in the future as a contractor, etc, etc....
The best thing to do is just not speak as if you know the intentions and motives of the company and/or its employees. You can talk about your work and your company but you have to follow various rules [both legal and H.R. related].
But, if you really want to keep speaking as an agent of IBM. Someday you may cross the line and get some nice legalese in the mail.
Tom
You're not under an NDA means you're not privy to any NEW information. It doesn't mean you can all of a sudden disclose everything you learned previously. If you gave out the secret sauce for Tivoli you can rest assured IBM would be all over you for that.
Just saying it's not wise to speak on their behalf. Saying things like
"It was my personal impression that people didn't care"
is generally better than
"It was IBMs position that nobody cared"
You have to disclaim that you're not speaking on behalf of the company, even if you're an ex-employee. [Hint: I learned this lesson a couple of weeks into working for my current employer. Even if what you say sounds innocent they will still jump all over it.]
Tom
As a former IBM employee, you should know better than to talk about internal politics and what not like that. Heck, I work at IBM as a vendor and even I'm closely scrutinized for what I say [not that I really have anything substantive to say publically in any case].
That aside, most employees of most companies are not really fully aware of the legal ramblings they're involved in. Out of site out of mind...
Tom
My god....mon dieu!! etc... 158$ must turn into what? 500$ retail? Have fun losing that phone.
Of course it'd be nice if cell companies both offered this monstrocity of a money pit and the el-cheapo phones that companies like Moto make as well. You know, that whole "free market" thingy...
Tom
Well it could be the gentoo/xorg combo but afaik I'm still locked into the xorg-nv driver right now.
:-( (that and I have consoles for gaming).
Which is ok because I'm too busy writing my book to miss the 3D accel.
Mostly I like my cards (NV43 iirc) because they were decently priced, work over PCI-E (a must for PCI TV tuner owners), don't need fans and provide good performance. Most of the time I'm in 2D mode anyways so it doesn't matter. But accelerated bitblts are nice and handy (specially for overlays) so I need more than a dumb framebuffer.
Tom
I'd rather have a 100GB 4200RPM then a 40GB 7200RPM :-)
Plus that's what the 2GB of memory is for.
Tom
The problem is most architectures are fairly locked down. For instance, putting another MUL engine in would be nice except MUL only reads from two registers and only outputs to two registers. A fourth pipe would raise retire latency, etc, etc, etc. In the AMD world at least most common ALU opcodes are already directpath. Larger caches could help but they raise the latency, so would raising the set-associtivity. More TLB entries are always welcomed, etc...
:-)
As an employee of one of the chip companies I couldn't really go into any other ideas without treading on NDA boundaries. Lets just say both camps have uses for more transistors. Don't worry
Tom
[STD disclaimer...]
4x4 ... You mean a pair of 285s and a tyan mobo? Aside from the lack of dual SLI that's basically a 4x4 setup.
If you actually need CPU power you can already buy Opteron gear today. Also while I wouldn't mind playing with a 4-core Intel kit I still love my 2x285 kit.
Tom
It'd be less of a problem if they
1. Spent more time to document the damn hardware
2. Opened the interface to the public.
The problem is the hardware is always in a state of flux and just incremental improvements. Your GeForce 7800 is probably based on the same HDL source as the 6600 with appropriate changes. This means that legacy symbol names from one project creep up into the new space. You get odd names, combined with lack of comments and documentation [compliance] leads to hardware with "oops" that the drivers have to work around.
Things could vastly improve for the customer if they stopped pretending that they know best. I know for a fact that companies like ATI and Nvidia spend a good deal of time [re: cost of the video card] in DRM technologies. Basically they don't give two shits about you as a customer so long as you
a) feel inadequate with your 75W GPU and buy the next best thing next quarter
b) fully comply with their "dominance" of your machine, force you to run windows, force you to use their bloaty drivers, use their drm, etc
I tolerate Nvidia solely because their kernel modules work decently [well not anymore as they're not keeping pace with xorg development]. Opening the 2D and 3D accelerators to the public can only serve to make the hardware more popular.
Their value is in the hardware and the ability to develiver it. Not the interface that puts a triangle on the screen.
Tom
The problem is manufacturing not so much the design.
Suppose they got the HDL for a super-duper 16 pipelined vertex shading GPU today. Could they still roll them out to end users for less than $500 a pop? Probably not.
Tom
For the price you pay for IA64 boxes you're going to be learing IA64 assembler fairly quick.
From what I heard handtuned bignum code flies on the older IA64 [doing RSA-1024 private key ops in ~500K cycles].
But when you pay $3600 for a processor you want to get the maximum benefit of it.
Tom
As another poster mentioned the IA64 is not a general purpose processor with decent pipeline control, register renaming, branch prediction, etc.
This means your typical random access applications [e.g. desktop] do not behave as well.
IA64 is really meant for the case where you can handtune that 5% of the code that takes 95% of the time in assembler. Who cares if your GUI is inefficient if you're pulling 4x the FLOPS as an Opteron in your HPC application? However, for a web server where you are doing less simd and more branching, loops, small ops, the Opteron arch [well and Conroe] helps.
It's like the traditional Transmeta vs. Intel/AMD. Intel and AMD spend a lot of hardware on runtime optimizations [e.g. prefetching, branch prediction, CISC => RISC decoding, register renaming, etc] so that it can cope with the current runtime situation [e.g. workload] whereas Transmeta optimized your task more statically [with less forethought to branches, cache, etc]. The net result was you could get some performance from Transmeta but the MIPS/MHz just wasn't there. Similarly IA64 *can* really fly if you hand tune your application AND it can take advantage of the architecture.
It isn't just a matter of having a good compiler [which admitedly GCC isn't there] but having a good problem.
Tom
24MiB of 6T SRAM is 1,207,959,552 transistors just for the cells. Not including tags, address decoders, etc...
By comparison, an Opteron uses 113,246,208 transistors for the 2048+256KB of cache [assuming they use a 6T structure which I don't know for a fact since I'm not privy to the details and technically I couldn't say even if I were, so don't assume what I said is verbatim fact, yada, I hate disclaimers] and the 4MB Duo (total of 4096+128KB of cache) needs 207,618,048 transistors for its cache.
Tom
Even with a sufficient market, right now that chip is going to be expensive. It's huge and the yield can't be that high.
But that's kinda the problem. The cost and the fact that's its a very niche processor will never make it as common place as Xeon or Opteron processors. Specially when both Intel and AMD are pushing SSE you can get a lot of the parallel SIMD like benefits of IA64 on x86.
Tom
It's ok, all Intel needs is people to buy the equivalent spec_int box [a mere half million dollars USD will do] and they'll be smooth sailing. :-)
Tom
The problem with the IA-64 platform is there are so few applications for which it is well suited. And even for problems it IS well suited it's a matter of figuring out how to extract the desired performance.
:-)
That said, a well tuned IA-64 application can smoke the best offerings from x86 world [on both sides of the fence]. But a $3700 USD price tag may push people away. Specially since processors like the Opteron 285 are nearly half the price and way more flexible.
Tom
I'm not tooting my own horn. I'm saying certification is meaningless and that people will use ANYTHING so long as it meets their needs and budget. My code happens to be free and apparently they don't care that I haven't forked over the required 10 grand or so PER RELEASE to get it verified [per platform too btw].
Tom
I am, by suggesting that the "testing centres" are scamsters who want you to keep thinking that with a FIPS seal your product is actually secure.
FIPS certification is largely meaningless outside of RNG and EM testing.
Tom
In my case, my code is used everywhere AND I make no money off it. :-)
Tom
CERTIFICATION is a SCAM!
It means nothing other than your implementation of an algo is correct. It doesn't mean you used it right.
Tom
Yeah a shared L2 is nice and all. Talk to me when you work in cluster computing though. Or have to run more than WinXP and your fav FPS.
And technically HT == hypertransport. HTT == hyper threading technology. Intel has never had HT [or similar] technology in x86 lines.
The types of tasks I use my box for don't really bode well for the whole "let's have our cores share access to a L2 thereby creating effectively a freaky fast FSB with contention issues".
I like my smaller but independent 1MB L2s on my Opterons because it lets me run tasks [independent tasks I may add] full steam ahead. And because my 2P box has two memory controllers I can get a shitload more bandwidth than any 2P Conroe could ever dream of.
Like I said, there are things one camp is good for, and things the other is. To unilaterally state that Conroe will whip AMDs butt is just a load of ignorant baloney.
Tom