Slashdot Mirror
The News From Computex, Including Non-Rambus P4s
M-Doggy writes: "With the festivities at Computex over in Taipei bringing in more out-of-towners
than the Chinese New Year, a plethora of new product announcements and sneak peeks
have hit the show floor. I noticed that AnandTech already has two articles up
covering the action. The
first one covers what was seen on the floor, including the AMD 760, NVIDIA's
Crush chipset, and more. The
second one has some interesting information regarding the non-RAMBUS solutions
for the Pentium 4 and even includes some preliminary benchmarks. Both speak of
the incredible politics behind the show; politics that rival even the recent events
in the Senate." (Read below for another snippet on those non-Rambus P4s.)
And Tuzanor writes: "Yahoo is reporting that Intel is releasing the i845, the first P4 chipset that doesn't use 'fast but expensive' Rambus memory. Funny, the story says that they will be using the "current standard DRAM chips" but says nothing of DDR RAM."
You are indeed correct that the $15 stuff is really bad and for low end systems. For my own stuff I typically purchase Micron or Siemens RAM which is decent stuff. Runs really nicely =)
$600 does seem steep until you consider the feature set.
If memory serves, that board has 2 onboard SCSI controllers, 2 onboard ethernet controllers, and an onboard AGP graphics device. If you bought the MSI board, and then the seperate plug-in devices to bring it up to that functionality, it would come out to about the same price.
So, the Tyan board is priced reasonably given its feature set. Of course, if you don't need all those features, then the less expensive board makes more sense for you.
--Lenny
AMD 760 boards yes, AMD 760MP no. Big difference between the single-proc 760 and the dual-proc 760MP. FYI.
Funny you should mention politics. Certainly RDRAM is better suited for the higher-bandwidth FSB of the P4 than the P3. However, when looking at DDR vs RDRAM performance on the P4, you should remember that Intel has no particular reason to make their DDR solution work very well, and a great deal of reason to show it performs worse than RDRAM. Just like their 815 (was that the one?) which performed worse than the ancient 440BX. No reason for it to do worse but for Intel not wanting it to perform well.
The enemies of Democracy are
Sure rambus is faster, but does it deliver a reliability+price/performance ratio that will attract corperate or even consumers?
:)
Faster than what? Sure, a dual-channel solution has more bandwidth than a single-channel (rah nvidia) DDR, but the latency sucks. This is why the benchmarks swing so wildly with an apps dependence on b/w vs latency. You can't replace one with the other, regardless of rambusinc hype.
Face it, today your bottleneck is the Video/Hard-drive (except for us using Ultra160 Scsi) and the bus to the outside world.
Sorry, but even with your scsi disk is still the biggest bottleneck (unless scsi somehow magically turns seek times into nanoseconds). I don't care how much data you can get me in a second -- if I (the cpu) have to wait ten million clock cycles before the first byte of data shows up, then I'm screwed.
Which I guess means I agree with you.
The enemies of Democracy are
Usually you do, except for the P4 processor. The P4 will probably never be what you want. See previous /. articles for long descriptions of the P4 problems. Rambus was just the tip of the iceberg.
If tits were wings it'd be flying around.
Sure rambus is faster, but does it deliver a reliability+price/performance ratio that will attract corperate or even consumers? Rambus has recieved a ton of bad press over the past 6 months, and with SDRAM priced at insanely low levels (Hey, I remember buying 16 meg of ram for over $1000.00 for a customers server) only the truely retarted would go rambus. IBM isn't even using rambus in the AVID non-linear video editing suites (Got a new one here less than 8 months ago, Yup PC-133 SDRAM in there) The latest Compaq servers dont use it...
Face it, today your bottleneck is the Video/Hard-drive (except for us using Ultra160 Scsi) and the bus to the outside world. (4 100BaseTX cards on a PCI bus is not living up to it's potential)
I dont care about faster, Give me 32 IRQ'S, how about an expansion bus that doesn't suck? How about a bios that isnt a P.O.S. out of the box?
my biggest gripe is the fact that we have been forced to use 16 IRQ's for way too damned long, they should have expanded it when they intorduced the PCI bus, now we have to wait forever to have a couple of free IRQ's on a new motherboard... (asus mobo's take 15 irq's out of the box)
There was a nice turnaround in hardware 2 years ago, but it is heading back into that pit of crap that made the late 90's hell on hardware hackers.
Do not look at laser with remaining good eye.
RDRAM sucks lots when coupled with the P4 because it has high latency. Coupled with the very deep pipeline the P4 has, this means that a branch misprediction will cost LOTS. Rembmber that the P4 fares well in very predictable environments such as multimedia.
From this point of view, SDRAM could help improve the P4's figures, since it has lower latency, thus branch mispredictions and random RAM accesses will be less penalized.
I'll wait and see.
Rumour has it that there will be be a version without the two Ultra/160 SCSI controllers.
Belief is the currency of delusion.
I wonder how long it will be until places like VA Linux and Penguin Computers have dual-athlon rackmount servers and deskside workstations for sale? :-)
Good luck with VA Linux and AMD. As far as I can tell they are as intel-only as Dell. If you ask them publically about AMD you get tight-lipped "No Comment" comments.
I hate to be one of those "Back in the day..." guys, but back in the day when I bought my first RAM upgrade for a PC, I payed more for 8MB than I would pay for 512MB today. Now that's fucking progress, mates.
Yeah well I remember paying over $80.00 for 8 *K* of memory. And data had to travel over the memory bus uphill both ways! You young whippersnappers!
The number of IRQs really isn't going to affect your performance. Remember most CPUs (x86 included) actually have only a single interrupt line going into them and they then handshake with an APIC to figure out what interrupt actually happened.
If you are worried about interrupt chaining, then there still isn't much time lag there. Handling an interrupt is usually a matter of just determining if your device caused the interrupt, saving the state of the device and firing off a backgroud operation to actually deal with the data. The only delay is running down the drivers chained off the interrupt to find the one that actually caused it - usually a matter of each one simply checking a single PCI register (ie 20 clocks).
Face it, the 'shared IRQ' is simply a non-issue any more (unless you have shitty drivers written by someone who learned to code in GW-BASIC and never read a design manual). Most RISC machines work happily with very few lines and the "16 Interrupts" idea is what is supported by the APIC and not the CPU anyhow, so get over it.
Fear: When you see B8 00 4C CD 21 and know what it means
If you don't know much about hardware then you probably won't want the "bleeding edge" technology. Stick with PC133, even though the RAM is the same price, the mobos that support it are more expensive.
ASUS is a good brand. I've had a number of them and they've always been rock solid with heaps of good features. They aren't the cheapest, but IMHO they are worth it.
Fear: When you see B8 00 4C CD 21 and know what it means
What sort of performance metrics do you have to show that shared interrupts are the source of your problem? My current laptop has damn near everything on interrupt 9 (as have most machines I've seen lately), and I've not noticed any performance problems at all. Smooth mouse, smooth video, DVDs play just fine without dropping frames, transfers over fast ethernet at >8M/s etc.
Most video capture problems are caused by disk latency or hitting a page fault at the wrong time, not interrupt latency. The average interrupt latency on most systems in somewhere under 1ms - if you are capturing at more than that rate then I suggest you need dedicated hardware. What model SCSI controller are you using? In many cases IDE is actually better value than SCSI - you can capture easily to an IDE drive at ATA/33 or better.
If you really want to know the largest cause of bottlenecks then I suggest you look at register spilling on the CPU. The x86 is horribly register starved and therefore hits the cache a LOT. If that's not where you want to look, check out things like the actual transfer rate to your drives, where the CPU time is going (system or user space), how loaded your I/O busses are, what bandwidth you have between North and South bridges etc. Shared interrupts really haven't been a problem for a long time, especially on "proper" operating systems like Linux.
Fear: When you see B8 00 4C CD 21 and know what it means
I'm buying a new mobo, athlon cpu, and memory in the next day or two. Should I get PC133 memory or DDR? It seems that DDR is much more expensive - is it significantly faster? I'm mainly using the workstation for PHP/HTML/SQL development, web surfing, and general office stuff.
Also, my friend recommended that I get an ASUS mobo. Was he right that this brand offers good price/performance?
As is probably obvious, I don't know jack about hardware. Any pointers about the above stuff would be greatly appreciated.
--
"my biggest gripe is the fact that we have been forced to use 16 IRQ's for way too damned long, they should have expanded it when they intorduced the PCI bus, now we have to wait forever to have a couple of free IRQ's on a new motherboard..."
This says two things 1) you don't understand what IRQ sharing/reentrant drivers are all about (which are NOT possible on the ISA bus, but are on the PCI bus/AGP port), and 2) that you seem to think you need to spend time configuring a PnP system. Don't try to configure a PCI system to force certain IRQs to certain devices, it won't work -- they do not need human intervention. You are obviously still scarred by the 1996 PnP implementations of ISA and OSes which are not samrt about resource allocation.
I have yet to see a modern system require more IRQs than it has because every modern PCI device can share them. I have yet to see a modern system require manual IRQ assignment to devices.
--
--
Internet Explorer (n): Another bug -- that is, a feature that can't be turned off -- in Windows.
From the ip-weenies dept:
In further news, Rambus, facing gallons of red ink in the 2nd quarter, have laid off 99% of their engineers. When asked for a reason for this perplexing move a Rambus spokesman stated, "The engineers were the cause of the whole problem! If we did not have to pay their salaries, we would all be millionaires! But now that Intel has violated our intellectual property by using DDR ram, we really need to tighten up and get to work." When then asked why Rambus kept their two lowest paid engineers the response was, "Well, we needed to keep them around in case one of our 500 lawyers has another annoying technical question. I hate those."
Why is it so hot? Where am I going? What am I doing in this handbasket?
And no, I'm not getting it confused with the spring water. (Hungry for swap, thirsty for Nanya?)
PC RAM has gotten just as cheap as CPU cycles and IDE disks, it seems. This is of course due the the new market for DDR RAM and RAMBUS, but even DDR RAM is remarkably inexpensive. I hate to be one of those "Back in the day..." guys, but back in the day when I bought my first RAM upgrade for a PC, I payed more for 8MB than I would pay for 512MB today. Now that's fucking progress, mates.
--
--
I like to watch.
But make sure that you're buying good integrated components, since you'll be keeping them for a while. Like the Intel STL2 (a dual-PIII board), which retails for about $600 but includes an Adaptec U160 controller, ATI video, and the oh-so-sexy Intel PRO 10/100 NIC. You may be able to do as well for less, but please research -- one "simiarly spec'ed" ASUS board does indeed include a SCSI controller, but it's a no-name SCSI2 controller, for fuck's sake.
$600 is very reasonable for a the first-available SMP Athlon board. People will pay for this board, trust me -- especially when they realize that they'll be saving major dough on higher-performing CPUs.
--
--
I like to watch.
Seems the 760 chipset is already available on the street: ugly pricewatch.com search query for the Tyan S2462. Price is about 580 give or take 20 it seems. (Of those shops I've actually directly dealt with essencom.com a few times; they seem to be a nice, reputable shop.)
(This suprised me as I thought what I've been seeing around the past few days were just beta boards.)
I wonder how long it will be until places like VA Linux and Penguin Computers have dual-athlon rackmount servers and deskside workstations for sale? :-)
--
News for geeks in Austin: www.geekaustin.org
News for Geeks in Austin, TX
Well, I've got an Intel Outside.
Seriously though, do we really need a cheesy slogan for AMD processors? People who choose AMD over Intel have more sensible reasons. IMHO even when AMD wasn't a serious competitor, I found the 'intel inside' logo as repulsive as *cringes* 'Designed for Windows 95'.
--
Escher was the first MC and Giger invented the HR department.
"AMD Anytime" sounds better. Consider this a modest gift from me to the marketing department of AMD. Use at will. Trademark it. Create dancing funny-clad people to go along with it. No strings attached.
Sure its better then earlier, but the T-bird and DDR still kicks butt :)
Probable impossibilities are to be preferred to improbable possibilities.
Aristotele
Those who ride the cutting edge often get cut by it.
The truth shall set you free!
Remember how when RDRAM first came out and it's benchmarks were abysmal? Rambus said that it was designed for the P4 moreso then the P3. I remember thinking that was a bunch of bs but it turns out to be true, not only that but the 10% performance difference (btwn RDRAM AND DDR)in that benchmark sounds about right if recall my impressions from some of Tom's comparisons.
I wouldn't go near a RDRAM based system but now that they have a viable platform for their product they'll never see the market penetration they now need. If I didn't hate Rambus so much I woulda considered an i850 RDRAM.
Anyone agree/disagree?
BOSTON SUCKS!
Seriously, I had to keep myself from yawning. The problem is that chip speed is far less important nowadays than it was 2 years ago. The bus speed matters a little bit, and it will be nice, but with recent Intel fiascos, I just don't see the point.
Perhaps I'm wrong, but I was recently pricing computers over at Gateway, since I got my son a copy of Black & White and I can't run it on any of my Linux boxen, so I decided to donate my old Win98 box to a friend and pick up WinME on a box that matches the demand specs of the game. And I looked at all the nice 1.1 GHz Intel chips and then saw that I could buy an 850 MHz AMD chip in a box for about $800 instead of shelling out $1600 for the "faster" Intel chip that I knew wasn't really faster.
So, what I'm saying is - so what? Until Intel finds some way to deliver faster Net access, it's all a bunch of hype about meaningless benchmarks that have no connection with my reality. And I'll still buy Linux for my servers, so I really just don't care.
But it is a cool trade show, I'll give you that.
--- Will in Seattle - What are you doing to fight the War?
I wouldn't touch Rambus with a 20 ft pole. Is it me, or do we always have to wait for the second generation of the Intel processor (ie P4 Northwood, P3 Cumine) before we even get remotely what we want?
People aren't demanding Rambus. They want something that's worth the cost!
To all of those that are not sporking this. . . All prices are subject to change without notice ;-) These prices come from our happy friends over at Pricewatch.com . :-) I like the prospects of that. For the price of One Rambus stick, I can put in nearly half a gigabyte of RAM. . . just think of how fast Doom will run. . .
Rambus - 64mb stick - $37
Sdram 133 - 64 mb stick - $9
Hmmm, that's odd, it would seem that it's nearly four times as much. And to check to make sure that it is a geometic expression. .
Rambus - 128 mb stick - $59
Sdram 133 - 128 mb stick - $15
So what do we learn from this class? Essentially that we are going to be able to have a lot more more memory running in our P4's
I am but mad north-north-west: when the wind is southerly I know a hawk from a handsaw.