AMD Rumors anyone else hear these.

[JerseyDevil]

2 Rumors I hear floating around the office.

1- AMD will be increasing the FSB on the 4200+ and 4400+ They will be 250FSB Chips.
After seeing Anands review this seems legitimate.

2-Now this one is out there. AMD will do a manufacturing run of the Athlon 64 on thier 110nm process currently being used for memory production. Its thought the 110nm will provide the perfect balance they cannot achieve with 90nm and enough speed increase over 130nm to produce higher speed chips that that wont experience the current leakage problems of 90nm.

The chips are expected to reach speeds of 3.2Ghz. Given the 110nm process is approx 20% smaller than 130nm and 130nm chips run at 2.6Ghz that puts it around 3.12ghz. This doesnt sound too far off. After all Intel can achieve 3.2 on 130nm so electrically it is possible? Especially with SOI and IBM behind it.

Anyone confirm or knowledgable enough to know if this is technically probable?

If so 110nm might be the holy grail AMD has been looking for. If Intel does the same for Prescott that may solve their strained silicon problems as well for a while.

 

[dannybin1742]

quite interesting, maybe plausable

 

[JerseyDevil]

Anand recently interviewed with someone at AMD maybe he can confirm?

 

[Gamingphreek]

Hmm well doesn't HT take the place of FSB, so that probably wouldn't work.

The second one seems perfectely feasible, probably for lower end like 754 though.

-Kevin

 

[imported_Bleh]

I thought a smaller nm would increase the efficency of the mhz cause the 939 90nm 3000+ runs at 1.8 ghz while the 754 130nm runs at 2 ghz and the performance is the same.

 

[Avalon]

Originally posted by: Bleh
I thought a smaller nm would increase the efficency of the mhz cause the 939 90nm 3000+ runs at 1.8 ghz while the 754 130nm runs at 2 ghz and the performance is the same.

I think most benchmarks show that a 3000+ 754 is faster than a 3000+ 939. Clock for clock the 90nm chip is faster, but it can't make up for a lack in 200mhz.

 

[SonicIce]

Originally posted by: Avalon

Originally posted by: Bleh
I thought a smaller nm would increase the efficency of the mhz cause the 939 90nm 3000+ runs at 1.8 ghz while the 754 130nm runs at 2 ghz and the performance is the same.

I think most benchmarks show that a 3000+ 754 is faster than a 3000+ 939. Clock for clock the 90nm chip is faster, but it can't make up for a lack in 200mhz.

But the 3000+ 939 has amazing overclocking potential that the 754's cant touch without water.

 

[imported_michaelpatrick33]

AMD will not have 110nm processor chips. They will have 130nm FX-55 at 2600 and the 130nm 4000+ at 2400 with 1meg L2cache like the FX-53 but with the multiplier locked upward (but not downward for Cool n Quiet obviously) The 4200+ and 4400+ will be 90nm parts and the FX-57 will be a 90nm part. The FAB making flash at 110nm is not capable of making processors and it never will be. The equipment and tools are memory only (converted from a processor FAB I believe).

If AMD raises the HTT and memory link on the processor to 250 (250 x 4 presumably as opposed to 200 x 5 now on 939) then AMD will be running a non JDEC standard memory (JDEC only supports memory on DDR1 up to 200DDR) and I doubt they would have a non industry standard memory running for servers or workstations. Also, a lot of motherboards may have problems running at 250 memory 1:1 without higher memory voltages. AMD will be moving to DDR2 on their processors late next year at DDR2 400 (800 in market speak) when DDR2 latencies and performance actually allow it to become a feasible alternative to DDR1.

It would be fascinating and an Intel destroyer if AMD did go to a 250 memory HTT for it would kill DDR2 533 and probably match or beat DDR2 667 on the Intel platform.

 

[miketheidiot]

Originally posted by: JerseyDevil
2 Rumors I hear floating around the office.

1- AMD will be increasing the FSB on the 4200+ and 4400+ They will be 250FSB Chips.
After seeing Anands review this seems legitimate.

2-Now this one is out there. AMD will do a manufacturing run of the Athlon 64 on thier 110nm process currently being used for memory production. Its thought the 110nm will provide the perfect balance they cannot achieve with 90nm and enough speed increase over 130nm to produce higher speed chips that that wont experience the current leakage problems of 90nm.

The chips are expected to reach speeds of 3.2Ghz. Given the 110nm process is approx 20% smaller than 130nm and 130nm chips run at 2.6Ghz that puts it around 3.12ghz. This doesnt sound too far off. After all Intel can achieve 3.2 on 130nm so electrically it is possible? Especially with SOI and IBM behind it.

Anyone confirm or knowledgable enough to know if this is technically probable?

If so 110nm might be the holy grail AMD has been looking for. If Intel does the same for Prescott that may solve their strained silicon problems as well for a while.

THis sounds like bullshit to me. First, im not sure how easy it would be to convert memory production to processor production. Second im not sure why they would do this when their 90nm is working fine. The 250 fsb sounds alot like 1000mhz HT divided by 4.

 

[miketheidiot]

Originally posted by: michaelpatrick33
If AMD raises the HTT and memory link on the processor to 250 (250 x 4 presumably as opposed to 200 x 5 now on 939) then AMD will be running a non JDEC standard memory (JDEC only supports memory on DDR1 up to 200DDR) and I doubt they would have a non industry standard memory running for servers or workstations. Also, a lot of motherboards may have problems running at 250 memory 1:1 without higher memory voltages. AMD will be moving to DDR2 on their processors late next year at DDR2 400 (800 in market speak) when DDR2 latencies and performance actually allow it to become a feasible alternative to DDR1.

It would be fascinating and an Intel destroyer if AMD did go to a 250 memory HTT for it would kill DDR2 533 and probably match or beat DDR2 667 on the Intel platform.

If amd went to DDR2 they would need a new socket(ddr2 has more pins). Not going to happen. DDR2 of anyspeed cannot compare to HT.

 

[imported_michaelpatrick33]

Originally posted by: miketheidiot

Originally posted by: michaelpatrick33
If AMD raises the HTT and memory link on the processor to 250 (250 x 4 presumably as opposed to 200 x 5 now on 939) then AMD will be running a non JDEC standard memory (JDEC only supports memory on DDR1 up to 200DDR) and I doubt they would have a non industry standard memory running for servers or workstations. Also, a lot of motherboards may have problems running at 250 memory 1:1 without higher memory voltages. AMD will be moving to DDR2 on their processors late next year at DDR2 400 (800 in market speak) when DDR2 latencies and performance actually allow it to become a feasible alternative to DDR1.

It would be fascinating and an Intel destroyer if AMD did go to a 250 memory HTT for it would kill DDR2 533 and probably match or beat DDR2 667 on the Intel platform.

If amd went to DDR2 they would need a new socket(ddr2 has more pins). Not going to happen. DDR2 of anyspeed cannot compare to HT.

They will move to DD2 but it will be awhile. They have to or Intel will eventually leave them in the dust. AMD has said before they will move to DD2 when it makes sense performance wise to warrant such a change. Probably the K10 will have a DDR2 400 memory on-die controller, or higher, depending on market conditions. AMD won't be able to let Intel have a DDR2 400 memory (by which I mean 800 in market speak) while they still sit at DDR1 200. Even with the on-die controller it will be outperformed.

Remember I always use the actual frequency of the memory when speaking about it

 

[ALIEN3001]

Remember I always use the actual frequency of the memory when speaking about it

Yep, that's smart. DDR2 800 has an actual frequency of 200MHz, DDR1 400 has an actual frequency of 200MHz, just don't forget that. DDR2's internal frequency is lower than DDR1's.

Secondly, DDR2 is not just about pure speed. It's latencies are higher, plus Intel's FSB is still 200MHz QDR (800MHz), so DDR2 800 wouldn't make much of a difference from current DDR2 133 (533).

Thirdly, it would be wise to accept a DDR1 466MHz (or 533MHz) standard, since meny suppliers are already selling those for quite some time. It would make sense, and with AMD's 1GHz HTT (8GB/s dedicated to RAM access ONLY) it would be much better than DDR2 200MHz (800) with Intel's 800FSB (or even 1066MHz, but those won't be out in any other form than Pentium4 EE for some time), it would be a lower latency, lower priced, less overheating solution.

 

[JerseyDevil]

We had an AMD rep at the office plugging away AMD. IBM tagged along trying to get us to use IBM/AMD servers. Were mostly a HP Cough Cough Crap Service shop. Lousy presentation IBM should update their charts they quickly passed by many outdated and old Opteron graphics. IBM really needs to do more homework and quit relying on their name to get them in the door. One thing is for sure HP should be kicked out the door. They are so incompetent and have such lousy support now I could get better answers from akid working at CompUSA or even worse BestBuy than these guys.

So one of the questions was how is AMD getting the word around and its mainly word of mouth and presentations however superbowl sunday they may plunk down a few million to spread awareness of AMD processors. Also since the Superbowl ads sit on websites after the superbowl for more people to check out. When the AMD rep said the SUN word IBM eyes quickly perked up and nothing more was said.

As for the 250FSB comment I have flying around one tech said something about intel going to a 1066fsb which he felt would increase Intel performance. The AMD rep made a comment that they arent the only ones raising the FSB. Its been done in the past with AMD too. So we are indeed assuming a 250FSB is comming. This also makes sense that AMD will prevent overclockers in a way because they could lock own the multiplier lower and this would allow for less FSB overclocking room.

Now for the 110nm. Someone commented on speed of the CPU not being able to clock very high on 90nm having the same problems as intel and how they thought they could stay competitive in a speed war. The AMD rep made the comment that sometimes you need to take smaller steps instead of taking big steps in order to find the perfect performance ratio. Moving a process thats 40% 90nm or even 65nm is the future but sometimes going with a small die shrink is a better solition that taking such a big step. Obviously no one is ready for such a big step. When one tech said 110nm the AMD rep just smiled and winked. So he was either gay or hinting that Opterons might get a run on 110nm.

Sorry I should have combined these threads.

 

[imported_michaelpatrick33]

There will not be a 110nm processor FAB. Not going to happen. Look at AMD roadmaps. Look at Intel Roadmaps. It will cost A LOT of money to have a 130nm, a 110nm, and a 90nm FAB process running all at once just for processors. It is illogical and makes no sense. Not going to happen. There are already 90nm processors and AMD has publicly stated that 50% of their A64's will be 90nm FROM 130NM within 3-4 months. Their Fujitsu/AMD Flash foundry in Austin (Spanion?) has just finished a full conversion to 110NM FABRICATION. Maybe you are confusing flash foundry production and processor foundry production.

AMD cannot lock the lower multiplier and still use Cool n Quiet without changing their microcode and processors and why would they introduce Cool n Quiet and tout it and then take it away. Makes no sense.

As I said before if AMD raising the HTT (there is no FSB on an integrated memory controller) to 233 or 250 with 1:1 memory frequencies than they will not run JDEC spec'd memory and I doubt they would do that but you never know with AMD

 

[Avalon]

Originally posted by: SonicIce

Originally posted by: Avalon

Originally posted by: Bleh
I thought a smaller nm would increase the efficency of the mhz cause the 939 90nm 3000+ runs at 1.8 ghz while the 754 130nm runs at 2 ghz and the performance is the same.

I think most benchmarks show that a 3000+ 754 is faster than a 3000+ 939. Clock for clock the 90nm chip is faster, but it can't make up for a lack in 200mhz.

But the 3000+ 939 has amazing overclocking potential that the 754's cant touch without water.

We weren't talking about overclocking potential. He claimed that a 1.8ghz 90nm chip had the same performance as a 2.0ghz 130nm 754 chip. The 3000+ 754 is faster, although not clock for clock.

 

[InlineFive]

The 250Mhz HT-Link has been known to be coming for a while. As for the 110nm fabbing I haven't heard anything.

 

[JerseyDevil]

The 250Mhz HT-Link has been known to be coming for a while

Ok so what kind of benefit could we see with this? Im thinking this will only benfit the server world but I am not technical like I used to be and just dont have time to do so. Im thinking Multi CPU performance could be increased for servers. Please more info.

 

[Sahakiel]

DDR-II differs from DDR in one primary aspect: The output bus is clocked twice as fast as the memory cells. Therefore, DDR running at 200MHz is the equivalent of DDR-II running at 400MHz. However, thanks to marketing, that becomes DDR400 vs DDR-II 800. The latency numbers reflect the change, that's why they're higher.
Also, DDR-II runs at lower voltages than DDR, thus reducing power consumption.

Originally posted by: Avalon
We weren't talking about overclocking potential. He claimed that a 1.8ghz 90nm chip had the same performance as a 2.0ghz 130nm 754 chip. The 3000+ 754 is faster, although not clock for clock.

Just a guess, but might want to check the cache sizes.
BTW, process technology does nothing for the "effective work per clock." A 2GHz design on 130nm and 90nm will perform the same for all intents and purposes. About the only area that would be affected is memory access latency, which could be either the same or 1 cycle more for the 130nm part. If there were a difference in latencies, the 90nm part would have a slightly better performance characteristic at the same clock speed compared to the 130nm part.

HyperTransport is not used to connect the memory controller to the memory cells. K8, no matter what the marketing reps imply, still uses a FSB to interface with the on-die memory controller. It just happens to be entirely on-die.
HyperTransport is designed primarily as a chip-to-chip interconnect. It's not just for CPU's. That means everything on your motherboard I/O-wise is eventually connected to HyperTransport links. A faster and/or wider HyperTransport link means you can send more data at once; your I/O devices will be less prone to starvation. Your PCI-E video cards, gigabit NICs, SATA HDDs, SATA DVD-R's, USB2 and Firewire 800 devices won't saturate anytime soon. However, if it is simply a clock increase, then latencies may decrease, which will do more than pure bandwidth at this point.