• Please review our updated Terms and Rules here

Very cheap P4 (socket 423) mobos on ebay

Socket 423 handles only the very first Pentium 4s getting up to 2 GHz. Note, if I remember the look of the slots correctly, that motherboard needs RAMBUS RIMMs which will be very expensive.
 
There were adapters to get a Socket 478 P4 to work in a 423 socket, but, like a lot of early P4 systems, the performance would be underwhelming. And, of course, you need the proper Dell case for these.

Even early 478 boards were a bit disappointing; I'm thinking of the early IBM Netvista systems--best used for their cases.
 
It's a Dell board and unless you have the case it might take some smithing to get that board to fit right because of the proprietary riser bezel.
 
If you want an early RAMBUS P4 get a whole system, they will be collectable someday. I don't like early P4's much but I do have at least one with SDRAM (Thinkcentre with p4-1.6 I think).

Those boards are cheap because they are not meant for ATX cases and used funky power supplies as well.
 
The SDRAM was used on the older P4's that were slow anyway, the later 3ghz LGA775 P4's with DDR were not too bad.
 
The SDRAM i845 motherboards were noticeably slower than the i845 RAMBUS equivalents with the same CPU but SDRAM was the way to get large amounts of RAM affordably. Looking at a 2001 ad, 256 MB SDRAM cost 25% of the 256 MB RIMM and finding any RIMMs was a matter of calling lots of vendors because supplies were just that lacking. Putting together a tolerable XP system required SDRAM until DDR became common. Whatever performance benefits high speed RAMBUS provides are overwhelmed to the slow down from swapping.

One other drawback with the empty motherboards is that any memory slot that doesn't get a RIMM will need a memory terminator which seem to be even harder to find today.
 
RDRAM being paired with the Pentium 4 was a really terrible idea. Netburst was a weak architecture to start that required lots of low latency cache and low latency main memory to perform halfway decent, neither which it had until late in the Pentium 4 life and DDR2. Only the final Prescott and Cedar Mill cores with 2M of cache and DDR2 had any sort of decent performance. The final Tualatin PIII-S 1400 wasn't soundly outperformed by the Pentium 4 until it reached speeds nearing 2 GHz. A dual Tualatin system could hold ground with even faster Pentium 4s that had Hyper Threading.

RDRAM was touted for its high bandwidth, but the problem is that it also had very high latency compared to SDR and DDR, not something being a good combination with the Pentium 4. It was also a problem in the Nintendo 64, the latency to main memory was so high that programmers often had to stream data directly from the ROM because it was orders of magnitude faster.

As for the availability of CRIMM terminator modules, you can still get them pretty easily on Ebay. I did a search a few minutes ago and found around 100 listings for CRIMMs.
 
Chuck(G) said:
Well, IIRC, it was a gold mine for some lawyers. :) Rambus seemed to be on a "let's sue everybody" toot for awhile.
Click to expand...

Well, doesn't RAMBUS have a headlock on the entire memory industry? They own he coop and everyone else has to stand with hat in hand and beg please.
 
Agent Orange said:
Well, doesn't RAMBUS have a headlock on the entire memory industry? They own he coop and everyone else has to stand with hat in hand and beg please.
Click to expand...

They've done it in bad faith, they're basically a scummy patent troll company.

They joined the JEDEC in the mid 90s and when they didn't agree to reasonable patent and licensing fees as required by JEDEC, they left. But not before drawing up patents based on publicly discussed information at JEDEC on upcoming memory technologies. So quite a few of their patents are from other companies and people that they had nothing to do with, which they used in litigation later on against said companies.

Eventually they went crying to the courts using many of the bad faith patents against the memory industry. Eventually many of them were invalidated due to rambus destroying evidence before trial and other scummy practices, but not before siphoning millions of dollars from the memory industry and Nvidia.
 
GiGaBiTe said:
... The final Tualatin PIII-S 1400 wasn't soundly outperformed by the Pentium 4 until it reached speeds nearing 2 GHz. A dual Tualatin system could hold ground with even faster Pentium 4s that had Hyper Threading.
Click to expand...

My experience is that my dual Pentium III-S 1.4GHz IBM x330 1RU server isn't soundly outperformed (where 'soundly outperform' means 'at least twice as fast') by ANY Pentium 4. I use the old Byte UnixBench suite for basic comparisons since it has a good mix of CPU-heavy and I/O-heavy benchmarks; the x330 gets a single-CPU score of 201.7 and a dual-CPU score of 324.6; I have several old Dell PowerEdge SC1425's with the late 64-bit capable Netburst Xeon's, which are basically Pentium 4's with lots of L2; at 3.2GHz, with HT off, I get a single-CPU score of 275.5 and a dual CPU score of 491.4 (running 64-bit code; I would expect 32-bit code to run faster by some percentage).

To 'soundly outperform' according to my definition the single-CPU score would need to be at least 403.4 or the dual-CPU would need to be at least 649.2; even the fastest NetBurst CPU made, running at 3.73GHz, probably can't touch a single-CPU score of 403.4. A dual-CPU dual-core box might be able to beat the multi-CPU score needed of 649.2, but I think upgrading to a Core 2 Duo will be required to reach the 'soundly outperform' threshold.

In contrast, a Dell PowerEdge 1950 (Core 2 Quad-derived dual Xeon E5410 @ 2.33GHz) scores 647.6 single-core, 2384.4 8-core, and my third-gen Core i7-3740QM (2.7GHz) laptop gets a single-core score of 1336.8 and an 8-core score of 4203.2. One of the major UnixBench components is Disk I/O bandwidth and latency, so multiple CPUs don't scale linearly. Also, just for grins and giggles, I ran UnixBench on an SGI Altix 350 system way back in 2012 and posted to Nekochan; this is a 30-CPU system, running 1.5GHz Itanium CPUs, and it racked up a single-CPU score of 444.0 and a multi-CPU score of 1170.6; the slow SATA drive tied to a single node really kills this system, as the single-CPU Dhrystone of 419 relative to the 30-CPU dhrystone of 12553 illustrates. So the Altix 350's IA64 does 'soundly outperform' the 1.4GHz Pentium III-S, at nearly the same clock speed.

The people who were grossly underwhelmed were the Socket 423 early adopters who bought 1.4GHz P4s, which were measurably slower than even the 1GHz Pentium III's for which they were supposed to be upgrades. It was hilarious.

As for the availability of CRIMM terminator modules, you can still get them pretty easily on Ebay. I did a search a few minutes ago and found around 100 listings for CRIMMs.
Click to expand...

I have two on my desk right now here at work.... people who come in my office who are somewhat computer-saavy are very confused by memory modules with no chips on them.....and I enjoy watching the reactions....
 
lowen said:
even the fastest NetBurst CPU made, running at 3.73GHz, probably can't touch a single-CPU score of 403.4. A dual-CPU dual-core box might be able to beat the multi-CPU score needed of 649.2, but I think upgrading to a Core 2 Duo will be required to reach the 'soundly outperform' threshold.
Click to expand...

The "fastest" Netburst by clock speed was the Prescott-2M Pentium 4 672 at 3.8 GHz, and its Xeon equivalents. But the raw clock speed doesn't have the grunt to back it up. For Intel to ramp up clock speeds on Netburst, they had to increase the already ridiculously long pipeline in every previous core revision at 20 stages to 31 stages in the Prescott and Cedar Mill. The much longer pipeline introduced pipeline stalls that significantly hampered performance and required even higher clock speeds to close the difference.


The Pentium 4 EE 3.73 GHz based on the Prescott-2M was actually slower than the former Gallatin 3.46 GHz Pentium 4 EE that preceded it due to the longer pipeline.
 
I'm actually referring to the Xeon 5080, although the passmark for the EE 965 is some better. The EE 955 at 3.46 is some slower. Both the EE 965 and the Xeon 5080 run at 3.733 GHz.
 
GiGaBiTe said:
They've done it in bad faith, they're basically a scummy patent troll company.
Click to expand...
Troll or not - they make the rules and if some upstart company decides to go it alone, it's just a matter of time before RamBus's squad of lawyers haul them into court. If there ever was a cause to invoke the Sherman Antitrust Act this would be it. Must be a lot of 'good stuff' going out the back door over at RamBus to the right people.
 
You must log in or register to reply here.
Back
Top