FX Athlons to be socket F?

[nrb]

According to this alleged AMD roadmap (the pedigree of which is, it must be admitted, open to question):

http://tweakers.net/nieuws/44504/AMD-4x4-platform-gebaseerd-op-Socket-1207.html

the FX-70, FX-72 and FX-74 Athlon processors will be socket F rather than socket AM2. The reason this is interesting is that AMD have been saying for a long time that their forthcoming 4x4 platform would only work with FX-series processors - but no one could think how AMD was going to enforce that, and therefore many people assumed that it probably would actually work just as well with non-FX chips (or could be hacked to do so).

But if this turns out to be true, then the reason why 4x4 requires FX processors is very clear: it requires socket F chips, so only FX-series processors (or perhaps Opterons, designed for server use) will actually fit into 4x4 motherboards. Since 4x4 is presumably just a slightly adjusted version of existing dual-socket Opteron technology, using socket F would mean that 4x4 is actually even more similar to forthcoming dual-socket F boards (which will be socket F) - so doing this would make sense.

If this is right, then I think it makes it less likely that there will be any chance of an "affordable" 4x4 system.

 

[hans007]

more or less you'd be buying a server board with dual pci-e slots. most dual socket boards do not have dual pci-e graphics slots as the boards would have to be huge. so more or less a cad workstation board.

it will probably require registered ddr2 ram. this is another lame, last minute hacked together gaming platform just like the original FX51/fx53 boards with socket 940. I think anyone who buys this would have to be a real amd loyalist especially since even a 3ghz dual dual core, woudl probably just barely be better than a core2duo quadro 6600 or 6700 based system which would be less expensive with just 1 cpu, and 1 socket board and non registered ram.

 

[myocardia]

Umm, I thought everyone knew that AMD's 4x4 was just B.S., that was really just to try and steal thunder from Kentsfield. Was I wrong?

 

[lopri]

From my observation, the 4x4 is neither a server/workstation derivative nor a stop-gap solution. AMD wants to establish a new open standard which connects CPUs and co-processors via HyperTransport, the so-called Torrenza, and the 4x4 is the first step towards that directioin.

However the decision to go with Socket F is, to say the least, very disappointing. If anything this shows AMD was once again incapable of presenting an appropriate loadmap, thus creating unnecessary confusion among potential early adopters. I'd guess the Socket AM3 will be nothing but a Socket F variance.

 

[nrb]

Originally posted by: lopri
From my observation, the 4x4 is neither a server/workstation derivative nor a stop-gap solution. AMD wants to establish a new open standard which connects CPUs and co-processors via HyperTransport, the so-called Torrenza, and the 4x4 is the first step towards that directioin.

However the decision to go with Socket F is, to say the least, very disappointing. If anything this shows AMD was once again incapable of presenting an appropriate loadmap, thus creating unnecessary confusion among potential early adopters. I'd guess the Socket AM3 will be nothing but a Socket F variance.

Torrenza will be quite cool - but I'm not sure 4x4 is in any way a step towards Torrenza. I think it is basically just an attempt to move dual-socket systems out of the server space and into the high-end desktop space.

We already know that socket AM3 processors will be compatible with socket AM2 motherboards, so I doubt they will be anything like socket F.

 

[Polish3d]

Originally posted by: hans007
more or less you'd be buying a server board with dual pci-e slots. most dual socket boards do not have dual pci-e graphics slots as the boards would have to be huge. so more or less a cad workstation board.

it will probably require registered ddr2 ram. this is another lame, last minute hacked together gaming platform just like the original FX51/fx53 boards with socket 940. I think anyone who buys this would have to be a real amd loyalist especially since even a 3ghz dual dual core, woudl probably just barely be better than a core2duo quadro 6600 or 6700 based system which would be less expensive with just 1 cpu, and 1 socket board and non registered ram.

3.2 ghz to be better or equal to e6600

 

[lopri]

I forgot that AM3 is supposed to be compatible with AM2. (Although I'm still not totally convinced with the fact the AM3 CPUs will have IMCs for both DDR2 and DDR3) I stand corrected on that part.

Originally posted by: Frackal
3.2 ghz to be better or equal to e6600

I wouldn't comment on 3.2GHz vs 2.4GHz but I'd like to comment that we have plenty of flaming threads on this forum as it is and there is no need for another one where the thread focus isn't performance/clockspeed comparison.

 

[Furen]

I'll repost what I posted on TR because, frankly, I like the sound of my own voice =)
Frankly, I'm disappointed that we'll have to deal with yet another socket change. I was kind of hoping AMD had actually planned the move onto 4x4 while designing AM2 but this is obviously not the case. I think the FXs get 10 more points because you get 2 CPUs in the same box, so the FX-70 will be two 2.6GHz FXs. This, of course, means that AMD was BSing when it said that people could buy a single chip and then upgrade later on. It also means that you won't be able to drop in accelerators into 4x4 systems, since you won't have an open socket.

That said, there's a bit of silver lining in these news. Hopefully, because the CPUs are socket F, you'll be able to use Opteron 2xxx parts instead of FXs, so buying a single CPU at a time could be doable. Then there's the fact that Accelerators for socket F could work in 4x4, which makes it easier on 3rd parties that make these. If there were dual-socket AM2 systems then these manufacturers would be forced to support two sockets or ignore the desktop completely. Another nice thing, in my opinion, is the fact that AMD seems to be about to kill off AM2 FX parts, so there will be significant differences between regular X2s and FX parts, the FX series being dual-socket "Desktop" parts.

 

[RichUK]

Meh, they'll still be fabbed on 90nm for the first part, and at a 125W a piece too.

 

[DeathReborn]

Socket F has more HT Links available if i'm remembering correctly which might explain the Socket F question.

As for the motherboards I suspect they'll be eATX format, which is fine for me as my case is capable of taking them.

Originally posted by: RichUK
Meh, they'll still be fabbed on 90nm for the first part, and at a 125W a piece too.

Sounds like the new Prescott. They really should have used 65nm for these to keep the TDP to maybe 89W or lower.

 

[aka1nas]

Originally posted by: DeathReborn
Socket F has more HT Links available if i'm remembering correctly which might explain the Socket F question.

As for the motherboards I suspect they'll be eATX format, which is fine for me as my case is capable of taking them.

Originally posted by: RichUK
Meh, they'll still be fabbed on 90nm for the first part, and at a 125W a piece too.

Sounds like the new Prescott. They really should have used 65nm for these to keep the TDP to maybe 89W or lower.

Yep, socket AM2 chips are just like socket 939 ones AFAIK, in that they do not have an extra coherent HT link needed for interprocessor communication. They would have had to either use modified Opteron 2xx class chips or have some extra glue logic in place, which would likely negate all the advantages of the K8 architecture.

 

[hans007]

apparently their 65nm yields are only up to 2.6 ghz at the top end . i'm sure there will be tons of 2.0-2.6 ghz chips on the market soon. it will probably be similar to when they first released the winchester 90nm chips and the clawhammers were still clocked higher.

we'll have to wait for a "venice" of the 65nm process to see if theyc an really squeak out more mhz. and probably only 3.0-3.2 ghz anyway.

 

[jiffylube1024]

Originally posted by: myocardia
Umm, I thought everyone knew that AMD's 4x4 was just B.S., that was really just to try and steal thunder from Kentsfield. Was I wrong?

Pretty much what I was thinking. 4X4 is novel but will be price prohibitive and really not for anyone outside the uber hardcore/uber rich...

 

[DeathReborn]

Originally posted by: jiffylube1024

Originally posted by: myocardia
Umm, I thought everyone knew that AMD's 4x4 was just B.S., that was really just to try and steal thunder from Kentsfield. Was I wrong?

Pretty much what I was thinking. 4X4 is novel but will be price prohibitive and really not for anyone outside the uber hardcore/uber rich...

Not only that but it might hinder sales of Dual Socket Workstations (depending on price & availability). It's nothing but a stopgap between Dual & Quad Core CPU's though. I think that 1 equally clocked Quad Core K8 would perform the same if not better & produce a cooler system in the process.

 

[Furen]

Originally posted by: DeathReborn

Originally posted by: jiffylube1024

Originally posted by: myocardia
Umm, I thought everyone knew that AMD's 4x4 was just B.S., that was really just to try and steal thunder from Kentsfield. Was I wrong?

Pretty much what I was thinking. 4X4 is novel but will be price prohibitive and really not for anyone outside the uber hardcore/uber rich...

Not only that but it might hinder sales of Dual Socket Workstations (depending on price & availability). It's nothing but a stopgap between Dual & Quad Core CPU's though. I think that 1 equally clocked Quad Core K8 would perform the same if not better & produce a cooler system in the process.

AMD's dual-socket workstation sales are negligible. In truth, this may actually increase sales since it looks like the entry-price should be lower. Currently a 2.8GHz Opteron 2xxx is $1.1k and the 2.6GHz part is almost $900. Since AMD says the slowest FX pack will be "well below" a grand I'd assume that the middle speed grade will be the $1000 bundle. Servers benefit from registered RAM (so people don't mind paying through the nose for "server" chips), workstations don't, or at least not to the same extent. Yes, a quad-core k8(L) will be faster because of the many improvements and the single memory controller BUT you could always drop in two quad-cores into 4x4...


I'd be hesitant to call it a stop-gap solution, it seems to me like it's AMD's response to Intel's MCP products. Until Intel releases its native quad-core parts and MCPs them (nice verb, huh?) AMD will have twice as many cores on its high-end "desktop" parts.

 

[coldpower27]

Originally posted by: Furen
I'd be hesitant to call it a stop-gap solution, it seems to me like it's AMD's response to Intel's MCP products. Until Intel releases its native quad-core parts and MCPs them (nice verb, huh?) AMD will have twice as many cores on its high-end "desktop" parts.

What do you refer to when you call for MCP products?

 

[Furen]

MCP=Multi-chip package. Intel can do that easily because it tends to have a die-size advantage (which lets 2 dice actually fit on the package) and has an external northbridge.