Why didn't AMD think of this?

[Brian23]

AMD should put the Athlon 64 on a slot card like they did with the slot A. On the card it should have 4 memory sockets for whatever type of memory the particular processor wants. The slot should be a Hypertransport standard slot such that any hypertransport processor could be plugged into the motherboard.

This would allow everyone to keep their motherboard when they upgrade. The PCIe/AGP/PCI controller would be in the motherboard chipset which would only need changed if you wanted new features. (SATA-III or USB-3 for example.) Also, when/if AMD choses to go to DDR2 or 3, then they don't have to require a new motherboard. Also memory timings could be tighter because AMD would know exactly the trace length to the memory chips.

It seems like a win-win solution. Why didn't they think of this?

 

[overst33r]

they make more money this way...i think

 

[aka1nas]

- Extra latency between the CPU card and the rest of the system?

- Not cost effective as most people don't upgrade.

- Pretty sure there would need to be some extra core logic and protocols associated with being able to use ANY CPU and memory. You would probably need a chip on the card to convert the data to a standard format and send it across the HT bus and then another chip to receive it in whatever protcol you defined. This would add even more cost and increase latency or add overhead.

- they do sort of do this already. Vendors like Sun have Opteron setups with the mobo in two pieces in which one part is expansion slots and the other the CPU and RAM. They aren't cost-effective really though.

 

[Brian23]

Originally posted by: aka1nas
- Extra latency between the CPU card and the rest of the system?

Wrong. The CPU is already using the HyperTransport bus, it's just not on a plug in card. It would run exactly the same.

- Not cost effective as most people don't upgrade.

More people would upgrade if you tell them that they can keep their motherboard.

- Pretty sure there would need to be some extra core logic and protocols associated with being able to use ANY CPU and memory. You would probably need a chip on the card to convert the data to a standard format and send it across the HT bus and then another chip to receive it in whatever protcol you defined. This would add even more cost and increase latency or add overhead.

No. Like I said, the opteron, athlon 64, sempron, and some Athlon XP-Ms use hypertransport already. There is no other connection between the CPU/memory and the rest of the system. (besides the power supply.)

- they do sort of do this already. Vendors like Sun have Opteron setups with the mobo in two pieces in which one part is expansion slots and the other the CPU and RAM. They aren't cost-effective really though.

They would be cost effective because it would GREATLY reduce the complexity of the motherboard. All the high frequency traces are between the CPU and the memory.

 

[jswjimmy]

also we would need new cases

 

[thanatos355]

"More people would upgrade if you tell them that they can keep their motherboard. "

how many mom and pop dell owners out there do you think even KNOW what a mobo is?

 

[Brian23]

Originally posted by: jswjimmy
also we would need new cases

Why? We had ATX cases with Slot A, Slot 370 and we still have them.

 

[imported_whatever]

Originally posted by: jswjimmy
also we would need new cases

not if it was a card that plugged in where PCI cards currently go, or was just not supported by the case (like slot based CPUs in the past)

 

[Brian23]

Originally posted by: thanatos355
"More people would upgrade if you tell them that they can keep their motherboard. "

how many mom and pop dell owners out there do you think even KNOW what a mobo is?

They know the difference when you tell them that they can swap their CPU/memory card for X dollars vs. telling them they need to upgrade their motherboard, cpu, memory, and reinstall their OS and programs because the new motherboard has different drivers. It's cheaper plus less work.

Sure you can change CPU, mem, and mobo without reinstalling, but it's a driver nightmare plus you won't get as good performance since you can't fully remove the files/drivers/registry entrys the old hardware left behind. If you could just replace a CPU card, then all the PCI/AGP/PCIe/IDE drivers can stay exactly where they are.

 

[jswjimmy]

i guess it would fit in 90% of cases. but i have seen some that it would not. i guess if you could fit a high end video card in it it would work but still it would not work in some cases.

my friend got a long card and it would not fit in his case. it hit the hard drive rails

 

[Brian23]

Originally posted by: jswjimmy
i guess it would fit in 90% of cases. but i have seen some that it would not. i guess if you could fit a high end video card in it it would work but still it would not work in some cases.

Show me an ATX case that won't accept an ATX motherboard.

 

[jswjimmy]

Originally posted by: Brian23

Originally posted by: jswjimmy
i guess it would fit in 90% of cases. but i have seen some that it would not. i guess if you could fit a high end video card in it it would work but still it would not work in some cases.

Show me an ATX case that won't accept an ATX motherboard.

its not the fitting the mobo its fitting the card. ive seen some cases the you can even get your finger between the pci cards and the case door. i guessing this card would be a lot bigger than a pci card

 

[Aenslead]

Originally posted by: thanatos355
"More people would upgrade if you tell them that they can keep their motherboard. "

how many mom and pop dell owners out there do you think even KNOW what a mobo is?

Point.

 

[imported_stmok]

Originally posted by: Brian23
AMD should put the Athlon 64 on a slot card like they did with the slot A. On the card it should have 4 memory sockets for whatever type of memory the particular processor wants. The slot should be a Hypertransport standard slot such that any hypertransport processor could be plugged into the motherboard.

This would allow everyone to keep their motherboard when they upgrade. The PCIe/AGP/PCI controller would be in the motherboard chipset which would only need changed if you wanted new features. (SATA-III or USB-3 for example.) Also, when/if AMD choses to go to DDR2 or 3, then they don't have to require a new motherboard. Also memory timings could be tighter because AMD would know exactly the trace length to the memory chips.

It seems like a win-win solution. Why didn't they think of this?

(1) Because it costs too much to produce a card/cartridge solution compared to a socket solutions.

(2) Because a similar idea is being done by ASRock!


The mobo that supports Socket 939 now, but allows you to upgrade to Socket M2 via CPU upgrade card. The upgrade card has the Northbridge/RAM/CPU/VRMs/etc. The mobo has a bunch of jumpers to tell it to use which solution (onboard or upgrade card).

But wait, there's more! It has AGP and PCI-Express Graphics slots. The AGP slot IS native, so unlike current solutions, there is no performance loss due to implementation. Its in fact competes quite well to AGP only mobos.

Don't believe me? Have a look.
=> http://www.ocworkbench.com/ocwbcgi/newspro/viewnews.cgi?newsid1117278681,68697

The chipset being used is a ULi solution. But it handles itself quite well to even nForce3/4 solutions. The reference mobo being benched here.
=> http://www.ocworkbench.com/2005/uli/m1695/g1.htm

So this mobo offers you the option to keep your AGP video card and use Socket 939 now.

In the future, you can upgrade to Socket M2 and PCI-Express Graphics.

In case you're wondering, yes it does support Dual-core A64 X2s as well.
And yes, you can use AGP and PCI-Express video at the same time.

 

[Brian23]

Originally posted by: stmok

Originally posted by: Brian23
AMD should put the Athlon 64 on a slot card like they did with the slot A. On the card it should have 4 memory sockets for whatever type of memory the particular processor wants. The slot should be a Hypertransport standard slot such that any hypertransport processor could be plugged into the motherboard.

This would allow everyone to keep their motherboard when they upgrade. The PCIe/AGP/PCI controller would be in the motherboard chipset which would only need changed if you wanted new features. (SATA-III or USB-3 for example.) Also, when/if AMD choses to go to DDR2 or 3, then they don't have to require a new motherboard. Also memory timings could be tighter because AMD would know exactly the trace length to the memory chips.

It seems like a win-win solution. Why didn't they think of this?

(1) Because it costs too much to produce a card/cartridge solution compared to a socket solutions.

(2) Because a similar idea is being done by ASRock!


The mobo that supports Socket 939 now, but allows you to upgrade to Socket M2 via CPU upgrade card. The upgrade card has the Northbridge/RAM/CPU/VRMs/etc. The mobo has a bunch of jumpers to tell it to use which solution (onboard or upgrade card).

But wait, there's more! It has AGP and PCI-Express Graphics slots. The AGP slot IS native, so unlike current solutions, there is no performance loss due to implementation. Its in fact competes quite well to AGP only mobos.

Don't believe me? Have a look.
=> http://www.ocworkbench.com/ocwbcgi/newspro/viewnews.cgi?newsid1117278681,68697

The chipset being used is a ULi solution. But it handles itself quite well to even nForce3/4 solutions. The reference mobo being benched here.
=> http://www.ocworkbench.com/2005/uli/m1695/g1.htm

So this mobo offers you the option to keep your AGP video card and use Socket 939 now.

In the future, you can upgrade to Socket M2 and PCI-Express Graphics.

In case you're wondering, yes it does support Dual-core A64 X2s as well.
And yes, you can use AGP and PCI-Express video at the same time.

The only "problem" with that solution is that it already has the CPU socket on the board. You're paying for a motherboard that has a lot of engineering in it that may not be used if you put a CPU card in it. If AMD put the CPUs on a card then the CPU pcb would be CHEAP due to MASS production. There wouldn't be motherboards with different length traces so memory stability issues would be completely nonexistant. (I'm not implying they have a problem now, I'm just stating that a more controlled memory interface can only help.)

 

[redhatlinux]

Every Heard of COAST, (cache on a stick) great IDEA, failed bad. YOUR obvious not a mobo designer. TRACES on a mobo that are just a few mm longer than "perfect" WILL cause problems. ADD a Socket and you will slow thw system down by a TON.

BTW, the 60nm technologie is over 4 years old and only NOW being seen by the General Public.

I consulted for AMD and a buddy of mine worked at the Austin plant.

THE DOC

 

[CheesePoofs]

The motherboard manufacturers would be the ones who could do that, not amd. AMD just makes the processors. SOmething similar to this has already been implimented with the boards that allow you to use s754 or s939 with add-in card. Just take that mobo design and change it a lot and you've got yoru idea.

However, it would never make any money. You have to remember the percentage of the market enthusiasts make up. Its tiny, and no OEM would ever bother buying a board like that when they could just buy one big board to cover it all.

 

[redhatlinux]

You still don't understand, when you go OFF CHIP for ANYTHING, it equals sloooow. I know EVERTHING about socket converters, used to build my own, mobo makers (preferred vendors work side by side with AMD !!!, I know I am one !!!!

 

[batmanuel]

Let's not forget this new weird little board that is coming from ECS.

This mobo supports LGA775 natively out of the box, but then you can buy an upgrade card to change the CPU to a Pentium M or even s754 or s939 AMD chip. It seems to be based around similar technology as the board mentioned earlier.

I just don't think it will ever catch on with the enthusiaist crowd, though. The biggest problem I can see is that it doesn't look like you could fit an XP-120 or a Zalman CNPS7700 on one of those upgrade cards without ripping the slot right off of the mobo when you moved the case.

 

[redhatlinux]

Look, there is ALWAYS room for niche market company's, but did anybody here experience Cache on a STICK .... great idea bak in the day. CPU cache memory, looked very similar to main memory of today, "plugged" in to the mobo. Case vendor's even made a 'easliy removable panel', for those "not trained at opening their BOXES ... he he. I built a system for my bro-in-law. OOOPs, tried to upgrade it to larger cache... NOGO, the timing was that critical. HYPER-TRANSPORT is SOOOOOOO SLOW compared to CPU speeds, VERY fast for a BUS, useless for a CPU.

SLOCKETS, socket adapters have been around for ever.... I've been posting in this forum, under a different ID for ever ... find out why certain manuf SLOCKETS only worked with certain mob's .... TIMING, VOLTAGE,Noise EMI component quality, etc etc. No WAY JOSE.

 

[Jeff7181]

Originally posted by: redhatlinux
You still don't understand, when you go OFF CHIP for ANYTHING, it equals sloooow. I know EVERTHING about socket converters, used to build my own, mobo makers (preferred vendors work side by side with AMD !!!, I know I am one !!!!

Liar.

 

[Brian23]

Originally posted by: redhatlinux
Look, there is ALWAYS room for niche market company's, but did anybody here experience Cache on a STICK .... great idea bak in the day. CPU cache memory, looked very similar to main memory of today, "plugged" in to the mobo. Case vendor's even made a 'easliy removable panel', for those "not trained at opening their BOXES ... he he. I built a system for my bro-in-law. OOOPs, tried to upgrade it to larger cache... NOGO, the timing was that critical. HYPER-TRANSPORT is SOOOOOOO SLOW compared to CPU speeds, VERY fast for a BUS, useless for a CPU.

SLOCKETS, socket adapters have been around for ever.... I've been posting in this forum, under a different ID for ever ... find out why certain manuf SLOCKETS only worked with certain mob's .... TIMING, VOLTAGE,Noise EMI component quality, etc etc. No WAY JOSE.

Why are you telling me that Hyper-Transport is slow for a CPU? Did you know that the Athlon 64 exclusively uses Hypertransport to comunicate with the rest of the system? (minus the memory.) My suggestion of putting the core on a seprate plug in board would not effect the system speed because the hyper transport would still run at the same speed. There would be no technical differnces except that the high speed part of the motherboard (memory interface) would be designed, controlled, and manufactured by AMD. The main benefit is increased upgradability options and lower cost for motherboards.

 

[thanatos355]

Originally posted by: Brian23

Originally posted by: thanatos355
"More people would upgrade if you tell them that they can keep their motherboard. "

how many mom and pop dell owners out there do you think even KNOW what a mobo is?

They know the difference when you tell them that they can swap their CPU/memory card for X dollars vs. telling them they need to upgrade their motherboard, cpu, memory, and reinstall their OS and programs because the new motherboard has different drivers. It's cheaper plus less work.

Sure you can change CPU, mem, and mobo without reinstalling, but it's a driver nightmare plus you won't get as good performance since you can't fully remove the files/drivers/registry entrys the old hardware left behind. If you could just replace a CPU card, then all the PCI/AGP/PCIe/IDE drivers can stay exactly where they are.

ahh, you've obviously not heard of a little app called 'sysprep'. go dig around your win2k or xp install disc and see what you find. or just google it. very usefull little app ms gave us.

 

[CheesePoofs]

Originally posted by: Brian23

Originally posted by: redhatlinux
Look, there is ALWAYS room for niche market company's, but did anybody here experience Cache on a STICK .... great idea bak in the day. CPU cache memory, looked very similar to main memory of today, "plugged" in to the mobo. Case vendor's even made a 'easliy removable panel', for those "not trained at opening their BOXES ... he he. I built a system for my bro-in-law. OOOPs, tried to upgrade it to larger cache... NOGO, the timing was that critical. HYPER-TRANSPORT is SOOOOOOO SLOW compared to CPU speeds, VERY fast for a BUS, useless for a CPU.

SLOCKETS, socket adapters have been around for ever.... I've been posting in this forum, under a different ID for ever ... find out why certain manuf SLOCKETS only worked with certain mob's .... TIMING, VOLTAGE,Noise EMI component quality, etc etc. No WAY JOSE.

Why are you telling me that Hyper-Transport is slow for a CPU? Did you know that the Athlon 64 exclusively uses Hypertransport to comunicate with the rest of the system? (minus the memory.) My suggestion of putting the core on a seprate plug in board would not effect the system speed because the hyper transport would still run at the same speed. There would be no technical differnces except that the high speed part of the motherboard (memory interface) would be designed, controlled, and manufactured by AMD. The main benefit is increased upgradability options and lower cost for motherboards.

I think redhatlinux is saying that hypertransport is slow for communication between different parts of the CPU (ie processing units and the cache) or between the CPU and the RAM.

If you're suggesting that the ram be on the card with the CPU, then there are no technical differences except that you have a connector in the middle of the hypertransport lane.

 

[Brian23]

Originally posted by: CheesePoofs

Originally posted by: Brian23

Originally posted by: redhatlinux
Look, there is ALWAYS room for niche market company's, but did anybody here experience Cache on a STICK .... great idea bak in the day. CPU cache memory, looked very similar to main memory of today, "plugged" in to the mobo. Case vendor's even made a 'easliy removable panel', for those "not trained at opening their BOXES ... he he. I built a system for my bro-in-law. OOOPs, tried to upgrade it to larger cache... NOGO, the timing was that critical. HYPER-TRANSPORT is SOOOOOOO SLOW compared to CPU speeds, VERY fast for a BUS, useless for a CPU.

SLOCKETS, socket adapters have been around for ever.... I've been posting in this forum, under a different ID for ever ... find out why certain manuf SLOCKETS only worked with certain mob's .... TIMING, VOLTAGE,Noise EMI component quality, etc etc. No WAY JOSE.

Why are you telling me that Hyper-Transport is slow for a CPU? Did you know that the Athlon 64 exclusively uses Hypertransport to comunicate with the rest of the system? (minus the memory.) My suggestion of putting the core on a seprate plug in board would not effect the system speed because the hyper transport would still run at the same speed. There would be no technical differnces except that the high speed part of the motherboard (memory interface) would be designed, controlled, and manufactured by AMD. The main benefit is increased upgradability options and lower cost for motherboards.

I think redhatlinux is saying that hypertransport is slow for communication between different parts of the CPU (ie processing units and the cache) or between the CPU and the RAM.

If you're suggesting that the ram be on the card with the CPU, then there are no technical differences except that you have a connector in the middle of the hypertransport lane.

I think you're the first person who understands what I've been trying to say.