That's still poor; I'm using the Free ATI drivers from the 2.4 kernel and XFree86 4.3.0, and I get about 800fps with glxgears (lousy benchmark, BTW). On my two year old P4 2.4 and Radeon 7500.
That's not surprsing, since you do get 3d acceleration for the radeon 7500, but none at all for the radeon 9800, for which no documentation is available unfortunately to the dri developers. (This project here, http://r300.sourceforge.net/ has a driver for r300 which is being developed by reverse-engeneering ATI's driver, but it's not usable currently.)
(and btw, 80fps seems to be way too slow even for software rendering, maybe the software renderer is synced to screen refresh...)
No card with DX9 support has specs available under any terms that permit an Open Source code release.
Not true. The i915 GMA (although it streteches the word graphics "card" a bit) might not be very fast, but it is a fully DX9 compliant chip. It has open-source drivers (including support for "pixel shaders", GL_ARB_fragment_program).
Hypterthreading doesn't cut the L2 cache in half. At least not statically, it is possible that if you're unlucky it will indeed look like each process will only have half the cache.
You're correct that the P4EE 3.4Ghz is out, but the 3.6 is not. In fact, this chip would probably be difficult for intel to manufacture, since the P4EE is still based on Northwood, and the highest speed grade of that core is 3.4Ghz. The Prescott core (P4E) currently goes up to 3.6Ghz, but no chips with additional cache exist so far.
IBM, for some horrible reason, doesn't think you need a a good graphics adapter: their 12" X40 uses Intel Extreme Graphics 2. Dell's 12" Latitude also uses this horrible Intel graphics hack
What do you have against Intel Extreme Graphics 2? If you compare it to your own requirement list, you can see it's all that's needed - you don't have 3d games on your list. For all basic tasks (office, dvd) that chip is quite good, with the advantage it saves some even some power compared to separate chips.
Also good luck getting Linux to run right with the Intel Pro(tected source code) Wireless LAN and the Intel Extreme(ly proprietary) Graphics adapter.
It is funny that you mention that. Most wireless lan chips and graphic chips do not have open-source drivers at all, but you can get an open-source driver easily for both the intel wireless lan chip (without firmware, but I don't think there is any wlan chip which you can get open-source firmware) and the intel extreme 1/2 (or even the GMA900 for that matter) graphic chip (btw the GMA900 is the only one of the "modern" chips I know of which has open-source drivers). Calling Intels graphic chip "Extremely proprietary" is really really far off the mark (now if you'd have said "Extremely slow" for 3d I wouldn't have objected), the "proprietary factor" of a GF FX5200 is about 1 million times higher...
won't run. cpu would be just acceptable, but the game requires a graphic card with some flexible fragment pipeline - either full "pixel shaders" (ARB_fragment_program) or some of the vendor specific extensions (it has different codepath for GeForce/GF2/GF4MX, GeForce3/GF4 and ATI Radeon 8500/9000-9200). There was supposedly also a generic path for "dumb" cards, but JC stated it is gone from the final game.
You make that sound like intel uses slower busses on Xeon setups just to fool their customers. I don't think that's the case at all. If you have only one cpu, it is much easier to increase the bus frequency, as it only needs to connect 2 chips (northbridge & cpu), trace length is quite short. But with 2 cpus (Xeon DP), you now have 3 chips to connect to the same bus. Not only are the traces going to be longer, but there is also more load on the bus. For quad setups, it's even a lot worse - 5 chips to connect, and you really start getting routing problems.
I believe this is the reason that current P4 use FSB800 (soon FSB1066), current Xeon DP use mostly FSB533 (the FSB800 Xeon DP (Nocona) has already been paper-launched), and Xeon MP are stuck with FSB400. Of course it is very counterproductive, as the available bus bandwidth per cpu really goes down the toilet (a Xeon MP in a quad setup has only 1/8 of the bandwidth available on average than a single P4!).
IIRC ATI cards can also do the single-pass two-sided stencil thingy - at least r350 (radeon 9800) and newer. You're right though they do not support "UltraShadow". There were some quotes some time ago, that "the NV3x are basically built for Doom III" and this is mostly why afaik. Actually I'm surprised that the GFFX5950 doesn't win always against the Radeon 9800, despite that missing feature (and despite the subpar ATI OGL driver).
Same thing that happens when your air cooling fan fails... your processor catches fire (that was a sweet video...)
Funny, but still wrong. That video you're probably refering to showed what happened when the heatsink is removed, not the fan is failing. BIG difference.
Doesn't mean that the cpu couldn't get damaged if the fan fails (and the system doesn't shut down), but it will take minutes to reach really high temperatures, and it certainly won't be such a spectacular death...
This law they are trying to pass will allow companies to patent the software (intellectual property) that they have written?
They already get copyright on the code they write, you seem to confuse that with patents...
Just like any other company that does R&D (the most costly aspect of producing a product) what is the problem with it being patented?
The problem is software is basically algorithms. You should not be allowed to patent business methods, mathematics et al. This just serves the big guys with lots of lawyers (since it's already impossible to write any piece of software without violating some trivial patents).
Let them recoup their costs. If I write some software, I be damned if someone is going to tell me how to market/sell/utilize it (short of national security breach).
You don't need a friggin patent to market/sell/utilize the software you wrote in any way you want. Copyright (which you get automatically) already grants you everything you need.
Well you're correct, though this specific display really wants 2 dual-link connectors, unless you're satisified with half the refresh rate - of course it has a REALLY high resolution.
Actually, no, the Apple Cinema Display 30" only needs 1 dual-link DVI connector (with the new Mac Geforce 6800 you can hook up two of them) - this thing needs two dual-link connectors, unless you want to reduce the refresh rate in half...
Transmeta's implementation works also for current generation efficeon chips, since it's 100% "firmware based" (in their codemorphing software). Source (german):
mczak
How is it apples and oranges? Why would I build a cluster with four way opterons when I could get a faster cluster using two way? If there was a $670 Itanium2 processor for clusers, I'd use that. But there isn't, and you have to look at what exists.
Actually, there is an Itanium2 "for clusters" (ok, intel says it's for workstations). Itanium2 DP is in fact much cheaper - the only difference to the Itanium2 MP is it has less L3 cache (only 1.5MB vs. the 1.4Ghz Itanium2 MP's 4MB), so its SpecFP score will no longer look that impressive;-)). And it can only be used in dual systems (not sure if this is really true, but it's advertized at least that way).
Are you saying that clusters built with Opteron 248 chips don't work, because they're not "Big Iron" chips? That's absurd.
They used quad-cpu nodes for their clusters for some reason probably (only half the nodes, more easily manageable, less space...). So it's probably fair to assume that if they would have used Opterons, they would have wanted quad-cpu nodes too.
Of course, Opteron 848 vs. Itanium2 MP is still much cheaper (as is Opteron 248 vs Itanium2 DP).
It is pin-compatible, but intel says you need a new stepping (b-step) of their chipset. I have no idea if these new steppings are already in use, or if it might run even with the old stepping unofficially. Also, the bios might not like it.
At least a 2.13Ghz P-M (with 533Mhz FSB) is on the roadmap. Not THAT much of an improvement from the now available 2Ghz part though. Still quite impressive nonetheless.
That's not really a fair comparison. You can get a Mobile Athlon XP-M 2600+ (mainstream) for $110, and it has a TDP of 45W. Yes, the Pentium-M still uses less power than the Mobile Athlon, but it's nowhere near as dramatic as your numbers suggest - and the price difference is even larger.
Does this mean if you're not doing a lot of number crunching, your battery life will be less than a Banias system?
Yes. However, the difference is rather small (keep in mind the cpu is not the only thing drawing power, so that 1.5W difference is really not that much).
Maybe the cache has something to do with this... whatever. I would be more impressed if the power requirements were lower across the board. And the lower peak doesn't seem that low.
No, it's not because of the cache (actually, banias/dothan have neat tricks to reduce power draw of the cache). It's a direct result of 90nm vs. 130nm. Peak power draw goes down, because of the lower voltage needed for switching in the transistors. But idle current goes up, because if you reduce structure size, you get more and more leakage current (which, btw, is a huge problem nowadays - leakage current was basically 0 just a short while ago, but it goes up exponentially with smaller structures).
Apple lossless codec is included in that comparison (guess there has been an update...). According to that comparison, it's nothing to write home about. Almost the same compression ratio as FLAC (very slightly worse), but slower (30% or so). Can't see a reason why you'd prefer it over FLAC (maybe it's faster on Macs, don't know).
2. The ATI Radeon X800s will require two power rails also. So stop dreaming about a "power efficient" part and buy a new PSU:(
That's so far just an INQ rumour. The X800 Pro will NOT have two molex connectors, noone seems to know if the X800XT will have two. Personally, I doubt it. Then again, I doubt the GF6800U really needs two molex connectors - some reviewer tested it with a 300W PSU (apparently of good quality, the key here is it must be able to supply most of that 300W on the 12V rails) with only 1 molex connector (using a y-cable)... The measurements indicate it doesn't really use that much more power than previous generation anyway.
I disagree though that dual-slot is inelegant. I'd rather prefer a dual-slot card which is somewhat quiet, than a loud single-slot card. Of course, IMHO it would have been a far better solution to have an exhaust hole directly venting out the hot air (seems like a stupid idea NOT to do this with a 2-slot design).
On PAL DVDs, the compressed format is MPEG-1 Layer 2
This is not quite correct. It might be true that players must support PAL DVDs with MPEG-1 Layer 2 audio, however pretty much all PAL DVDs you can buy use AC-3 audio instead. At least I have never seen a PAL DVD which uses mpeg audio...
addressing 4):
I've seen several sites using compiler benchmarks recently. Quick summary: The P4 Prescott is about the same speed as the P4 Northwood, if compiling isn't done in parallel both don't have the slightest luck against Athlon 64 (can't even beat Athlon XP). However, if something like make -j3 is used, the P4 are very close to the Athlon 64.
Some link (Visual Studio, can't remember the other which used gcc):
http://www20.tomshardware.com/cpu/20040318/athlon- fx53-28.html
Unfortunately the pin layout will be different - the whole point of socket 939 is to make it possible to build 4-layer boards, which supposedly requires a new pin layout - it seems to be impossible with the current socket 940 chips.
need to correct myself, the 1499$ version doesn't have a FX 5200 Go nor a FX 5700 Go - it has integrated intel extreme graphics, about as slow as it gets... (and it doesn't have a P4 3E neither)
The about 2000$ version has the FX 5700 Go and the P4 3E.
Slashdot Mirror
Hypterthreading doesn't cut the L2 cache in half. At least not statically, it is possible that if you're unlucky it will indeed look like each process will only have half the cache.
You're correct that the P4EE 3.4Ghz is out, but the 3.6 is not. In fact, this chip would probably be difficult for intel to manufacture, since the P4EE is still based on Northwood, and the highest speed grade of that core is 3.4Ghz. The Prescott core (P4E) currently goes up to 3.6Ghz, but no chips with additional cache exist so far.
Calling Intels graphic chip "Extremely proprietary" is really really far off the mark (now if you'd have said "Extremely slow" for 3d I wouldn't have objected), the "proprietary factor" of a GF FX5200 is about 1 million times higher...
maybe it will have aalib support for those people still running hercules monitors ;-)
won't run. cpu would be just acceptable, but the game requires a graphic card with some flexible fragment pipeline - either full "pixel shaders" (ARB_fragment_program) or some of the vendor specific extensions (it has different codepath for GeForce/GF2/GF4MX, GeForce3/GF4 and ATI Radeon 8500/9000-9200). There was supposedly also a generic path for "dumb" cards, but JC stated it is gone from the final game.
You make that sound like intel uses slower busses on Xeon setups just to fool their customers. I don't think that's the case at all. If you have only one cpu, it is much easier to increase the bus frequency, as it only needs to connect 2 chips (northbridge & cpu), trace length is quite short. But with 2 cpus (Xeon DP), you now have 3 chips to connect to the same bus. Not only are the traces going to be longer, but there is also more load on the bus. For quad setups, it's even a lot worse - 5 chips to connect, and you really start getting routing problems.
I believe this is the reason that current P4 use FSB800 (soon FSB1066), current Xeon DP use mostly FSB533 (the FSB800 Xeon DP (Nocona) has already been paper-launched), and Xeon MP are stuck with FSB400. Of course it is very counterproductive, as the available bus bandwidth per cpu really goes down the toilet (a Xeon MP in a quad setup has only 1/8 of the bandwidth available on average than a single P4!).
IIRC ATI cards can also do the single-pass two-sided stencil thingy - at least r350 (radeon 9800) and newer. You're right though they do not support "UltraShadow". There were some quotes some time ago, that "the NV3x are basically built for Doom III" and this is mostly why afaik. Actually I'm surprised that the GFFX5950 doesn't win always against the Radeon 9800, despite that missing feature (and despite the subpar ATI OGL driver).
Doesn't mean that the cpu couldn't get damaged if the fan fails (and the system doesn't shut down), but it will take minutes to reach really high temperatures, and it certainly won't be such a spectacular death...
Well you're correct, though this specific display really wants 2 dual-link connectors, unless you're satisified with half the refresh rate - of course it has a REALLY high resolution.
Actually, no, the Apple Cinema Display 30" only needs 1 dual-link DVI connector (with the new Mac Geforce 6800 you can hook up two of them) - this thing needs two dual-link connectors, unless you want to reduce the refresh rate in half...
Transmeta's implementation works also for current generation efficeon chips, since it's 100% "firmware based" (in their codemorphing software). Source (german):
mczak
Of course, Opteron 848 vs. Itanium2 MP is still much cheaper (as is Opteron 248 vs Itanium2 DP).
It is pin-compatible, but intel says you need a new stepping (b-step) of their chipset. I have no idea if these new steppings are already in use, or if it might run even with the old stepping unofficially. Also, the bios might not like it.
At least a 2.13Ghz P-M (with 533Mhz FSB) is on the roadmap. Not THAT much of an improvement from the now available 2Ghz part though. Still quite impressive nonetheless.
another very recent interview with Dave Orton, quite interesting IMHO: http://www.beyond3d.com/interviews/daveorton/
Apple lossless codec is included in that comparison (guess there has been an update...). According to that comparison, it's nothing to write home about. Almost the same compression ratio as FLAC (very slightly worse), but slower (30% or so). Can't see a reason why you'd prefer it over FLAC (maybe it's faster on Macs, don't know).
I disagree though that dual-slot is inelegant. I'd rather prefer a dual-slot card which is somewhat quiet, than a loud single-slot card. Of course, IMHO it would have been a far better solution to have an exhaust hole directly venting out the hot air (seems like a stupid idea NOT to do this with a 2-slot design).
addressing 4):- fx53-28.html
I've seen several sites using compiler benchmarks recently. Quick summary: The P4 Prescott is about the same speed as the P4 Northwood, if compiling isn't done in parallel both don't have the slightest luck against Athlon 64 (can't even beat Athlon XP). However, if something like make -j3 is used, the P4 are very close to the Athlon 64. Some link (Visual Studio, can't remember the other which used gcc): http://www20.tomshardware.com/cpu/20040318/athlon
Unfortunately the pin layout will be different - the whole point of socket 939 is to make it possible to build 4-layer boards, which supposedly requires a new pin layout - it seems to be impossible with the current socket 940 chips.
need to correct myself, the 1499$ version doesn't have a FX 5200 Go nor a FX 5700 Go - it has integrated intel extreme graphics, about as slow as it gets... (and it doesn't have a P4 3E neither)
The about 2000$ version has the FX 5700 Go and the P4 3E.