douglar wrote on 2025-07-17, 15:39:

This is a little off topic: I can understand finding sympathy for the Nurse Ratched character because she was assaulted at work by an convicted sex offender who was abusing the legal leniency granted occasionally granted to the mentally ill. However Nurse Ratched was hardly a good nurse or even a good person because she clearly lacked sympathy for her patients and her application of the rules she espoused to uphold was clearly arbitrary and manipulative. To use your words, she was also an asshole.

Sure, she was a bitter witch... But don't forget that the U.S. abolished those types of institutions altogether, which is terrible. I prefer to go back to the old system with mental hospitals that have Nurse Ratcheds, than the current system today. Bring back the straightjacket I say.

Better than just having insane vagrants all over the streets that's for sure

To be fair to Intel they did release the E7205 chipset to consumers (which was intended to be an expensive workstation chipset) with dual channel DDR support fairly soon after RDRAM proved to be a disaster.
I remember the RDRAM discussion back then, it was all about the insane price that never seemed to go down.
Reviews weren't really that bad until it was clear that it would never be anything but extremely expensive.
Now had Rambus managed to get the price more competitive maybe history would be a bit nicer to the first gen P4's.

That was a chicken-egg situation: it was so expensive because they didn't have the volumes and they didn't have the volumes because it was so expensive. Due to the royalties it would always be more expensive than equivalent amount of SDRAM, but that wasn't even the biggest factor. Intel tried to bootstrap that by ramming the Rambus down our throats regardless of price, by simply not supporting SDRAM on their higher-end parts.

We - literally - didn't buy it, and by 'we' I particularly mean big OEMs and corporate procurement departments - so where the real volumes are. Previously most had been Intel-only shops, but the price difference with other (mainly Via) chipsets became so huge that many of the staunchest holdouts flipped. This had another unintended consequence of strategic importance to Intel: by normalizing non-Intel chipsets: it vastly improved the prospects of AMD's Athlon CPU, which only had non-Intel chipsets. If you're reading this on a system with an AMD Ryzen CPU, you probably have to thank Intel's overreach with strong-arm tactics in the 2000 memory wars for AMD being around to make it...

It also fucked us now. There are only 2 chipset makers and despite both using a slightly modified pcie bus, they're not compatible even if you ignore the sockets.

dionb wrote on 2025-07-16, 10:11:

Even when i845GE came along, i850E still had a performance advantage - i850E with dual channel PC1066 RDRAM outperformed single-channel PC2700 DDR1; the i875P dual-channel DDR chipset was the first to actually beat the i850E over a year later. That was not because RDRAM couldn't go faster, it was because Intel had admitted commercial defeat and had stopped development of RDRAM chipsets.

The first Intel chipset to beat i850E was Granite Bay, which Intel ported to desktop from workstation branch for Xeon CPUs. Granite Bay had dual channel DDR266 support. Arguably, also SIS 645DX, which showed very impressive performance with DDR333.

Matth79 wrote on 2025-07-16, 01:13:

133 SDRAM is nicely balanced for a 133FSB and has good latencies. RAMBUS / RDRAM was a car crash, terrible latencies, so hot it needed heat spreaders that weren't just a fashion statement, and Intel shackled themselves to it for that generation of chipsets, and it was a wrong turn, but not the only wrong turn Intel made

I believe I recall the existence of RD to SD adapters. Btw. So if anyone is looking to ditch the RamBus.

https://www.overclockers.com/forums/threads/w … o-sdram.222963/

Sphere478 wrote on 2025-07-18, 13:51:

Matth79 wrote on 2025-07-16, 01:13:

133 SDRAM is nicely balanced for a 133FSB and has good latencies. RAMBUS / RDRAM was a car crash, terrible latencies, so hot it needed heat spreaders that weren't just a fashion statement, and Intel shackled themselves to it for that generation of chipsets, and it was a wrong turn, but not the only wrong turn Intel made

I believe I recall the existence of RD to SD adapters. Btw. So if anyone is looking to ditch the RamBus.

https://www.overclockers.com/forums/threads/w … o-sdram.222963/

I have that device. It's a proprietary Asus monstrosity, but if you have a supported board (Asus P3C-series, basically) it does what it's supposed to as well as can be hoped; I've not encoutered any MTH-specific instability when using it with my P3C-D, but performance is abominable as it's limited to 100MHz SDRAM and of course you get all the added latency of the MTH as protocol converter, so after discovering it worked it went straight back into my oddities cabinet and the two P3-933EBs on my P3C-D got their PC800 RDRAM back again.

dionb wrote on 2025-07-25, 11:16:

Sphere478 wrote on 2025-07-18, 13:51:

Matth79 wrote on 2025-07-16, 01:13:

133 SDRAM is nicely balanced for a 133FSB and has good latencies. RAMBUS / RDRAM was a car crash, terrible latencies, so hot it needed heat spreaders that weren't just a fashion statement, and Intel shackled themselves to it for that generation of chipsets, and it was a wrong turn, but not the only wrong turn Intel made

I believe I recall the existence of RD to SD adapters. Btw. So if anyone is looking to ditch the RamBus.

https://www.overclockers.com/forums/threads/w … o-sdram.222963/

I have that device. It's a proprietary Asus monstrosity, but if you have a supported board (Asus P3C-series, basically) it does what it's supposed to as well as can be hoped; I've not encoutered any MTH-specific instability when using it with my P3C-D, but performance is abominable as it's limited to 100MHz SDRAM and of course you get all the added latency of the MTH as protocol converter, so after discovering it worked it went straight back into my oddities cabinet and the two P3-933EBs on my P3C-D got their PC800 RDRAM back again.

So it used unused Rambus pins or something that only that board routed? Or it was purely a protocol converter and all lines were already usable?

Can you upload a high res of both sides? And board?

Think that board just took the RDRAM signals and supplies and fed into the MTH which converted signaling to SDRAM and back to the RDRAM bus. Some 820 motherboards that used SDRAM used the same MTH for that cause. But as dionb mentions, performance hit is ridiculous ...

Sphere478 wrote on 2025-07-26, 17:26:

[...]

So it used unused Rambus pins or something that only that board routed? Or it was purely a protocol converter and all lines were already usable?

Can you upload a high res of both sides? And board?

I'll see about pics this evening, but basically it's just an MTH-on-a-stick, so interfaced with the i820 MCH via PC800 RDRAM, buffered and converted that to PC100 SDRAM - just like the onboard MTH on boards like the Intel CC820 and Asus P3C2000 did.

Actually, there are perfectly good pics of it in the P3C-D manual, take a look on page 26: https://theretroweb.com/motherboard/manual/p3 … ad154924984.pdf

Oh, and the device really is proprietary: you can't just stick it into any 184p RIMM slot and expect it to work. I tried on another i820 board (FIC KC19+) and it registered a memory error, the same as sticking it into one of the 'normal' 184p slots on the P3C-D.

Compaq had something similar, albeit for the i840-based AP550 workstation - https://theretroweb.com/motherboard/image/com … 49701608684.jpg

PC Hoarder Patrol wrote on 2025-07-28, 13:50:

Compaq had something similar, albeit for the i840-based AP550 workstation - https://theretroweb.com/motherboard/image/com … 49701608684.jpg

Interesting - particularly given that these cards don't need a dedicated/special RIMM slot like the Asus one, but slot into the first regular RIMM with a C-RIMM next to it, as if it had been any other RIMM module.

Now I happen to have the motherboard of an AP550, so could conceiveably use these - but as you need to install two at the same time (and they would be a massive downgrade in performance and stability) and these things are obtainable but rather expensive I don't see myself testing these things anytime soon...

I’m recalling possibly a third adapter not yet shown. Can’t think of keywords for search though. Thoughts?

Image rehost of above links incase links go dead

I hope this thread is still active.
I once had a talk with a adjunct faculty of the university where I got my EE degree.
He used to work at Intel in the mid to late '90s to mid-2000s, and belonged to its platform division (i.e., PC chipset design).
I will keep the identity of this person private, but he used to belong to a company Intel acquired for their PC platform expertise.
Anyway, I once asked why Intel choose RDRAM, and his answer was that people above him (this person was in the middle management of Intel after the acquisition) I cannot remember who it was that was the head of the PC platform division made the decision to adopt RDRAM.
This person told his "boss" (the aforementioned PC platform chief) that RDRAM has little performance benefits since most typical P6 family memory traffic gets quite well cached by 2 levels of cache, and there is little long continuous burst transfer due to this where RDRAM may shine over SDRAM.
If I recall, read cache line fill is only 4 beats (4 transfers), and with such short burst, there is little benefit over SDRAM.
Another thing he noted was, even with RDRAM, fundamental DRAM access parameters (i.e., tRCD, tRP, tRAS, etc.) remain the same, hence, things do not get that much fundamentally faster.
Anyway, his boss got carried away by Rambus' hype against his advice, and we all know how things turned out for Intel between 1999 to 2002.
That's my 2 cents to this discussion.

Slightly off topic about RDRAM.
I have been working on a creating a replica of VIA Technologies VT5228C reference schematic mainboard for KiCad.
VT5228C reference schematic mainboard has Apollo Pro133A (VT82C694X) and VT82C686A "Super South" along with support for 133 MHz Slot 1 processors.
I have read the Apollo Pro133A chipset design guide and Intel 440BX chipset design guide, and both total around 140 to 160 pages, if one counts the OrCAD schematic.
Anyway, the ill fated Intel 820 chipset design guide is roughly double of the page count, and is far more detailed than the aforementioned two other design guides.
Besides that, Intel 820 chip set requires 28 ohm characteristic impedance for RDRAM traces versus 65 ohm for Apollo Pro133A and 440BX.
The recommended materials used in the 28 ohm characteristic impedance 4 layer PCB stack up is also different in terms of material thickness, and perhaps these lower volume, less familiar PCB fabrication methods might be the reason why mainboard and memory modules were so expensive back then.
I am not really an expert in PCB design, but just wanted to bring this up.

I can make that a whole lot easier for you.
See sig, and give me a bit to find one I prepared earlier.

Actually do you have a place to upload it all, or sendanywhere?

Have a look at the below, and let me know which from the LIST you want
I cbfd sorting which are apollo or not, so you tell me.

Heh. Pretty sure the only downloads are bots.

Users browsing this forum: Apple [Bot], ChatGPT [Bot], Claude [Bot], DuckAssist [Bot], MrCrocodile, myne, Nitram78, OpenAI [Bot], PetalBot [Bot], RetroPCCupboard, supergeek and 9 guests