AMD in 2006 ?

[Viditor]

Socket F is also rumored to possibly have the PCI-E bus integrated on chip

Yup...forgot about that one...cheers!

As to difficulty, I suppose moving the PCIe to the Northbridge could be done, but I'll be damned if I can figure out how they are going to do it...guess that's why I work in TV and not chip design!

🙂

 

[Viditor]

Originally posted by: BlingBlingArsch
i want benchmarks. 🙂 or what are we talking bout here?

OK BBA...just go take a nap, wake up around Xmas 2006, and I swear I'll have them waiting for ya!

😉

 

[Bateluer]

Originally posted by: Bona Fide
M2 = AMD's biggest mistake ever. They're gonna be seeing heavy losses for much of '06 🙁

Yeesh, theres been almost nothing official about M2. No specs, no benchmarks. Nada. Please stop pretending you know more about how to run a multinational corporation than the multinational corporation. 😛

 

[Viditor]

Bona Fide...

1. Latencies are higher on DDR2, that's true. But there is a diminishing return vs. bandwidth that AMD has stated for over a year now is at 667 MHz...this means that even at the higher latencies of DDR2 (and they aren't THAT much higher...), the higher bandwidth overcomes those losses.
2. DDR2 has another advantage...it requires less power to use than DDR. This helps in many different scenarios (mobile, blades, etc...).
3. By next year, DDR2 will be as cheap or cheaper than DDR as well...
4. AMD will continue to manufacture both M2 and S939 all of next year and beyond. They gave developers a 5 year guarantee on the S939 chips...
5. M2 chips will have both Pacifica and Presidio...unfortunately, that also means that (along with Intel) they will have DRM support...sigh. But they will also have virtualization which will rock!

 

[imported_SLIM]

I believe I read somewhere that the DRM part (presidio?) will be a choice in the bios just like CnQ and internal cache. So, don't be afraid of one feature when it is likely you can just turn it off.

 

[Bona Fide]

They're gonna have DRM? :Q

First Windows Vista and now this? Noooooooooo

 

[Sentential]

Originally posted by: BouZouki

DDR2 support : Big mistake. DDR2 latencies are sky-high, and the lowest I've seen are 4-4-4-12. If AMD just stays with DDR, they can garner support from the enthusiast companies such as OCZ, Mushkin, Crucial, and even Corsair, all companies that are continuously improving their DDR series. OCZ just came out with its 2x1GB Gold VX PC-4000 RAM. By moving to DDR2, they will be losing support in the enthusiast and mainstream gamer communities, which account for much of their sales, apart from server-grade computers.

Wow just wow.

DDR2 timmings are well bellow 4-4-4-12 buddy, what if I told you I can get DDR2 running at 3-2-2 fairly easily.

Ya. Some people claim to have the facts when they clearly do not. The best balanced DDR2 available right now is DDR2 800 5-3-3-8 by Mushkin. This RAM can be pushed as high as 1050mhz (which I have done personally). So your rant about DDR2 being garbage is completely bogus.

In addition M2 will infact run MUCH hotter than S939 because the minimum amps given to the socket have shot up significantally. This will push generic M2 cores into the 115W range (as AMD's tech sheet suggests)

In addition the idiocy of AMD's managment is staggering. We will see a repeat of S754 with M2. I dont understand why they would only integrate the SB onto 1207 and not M2. Customers wont buy it and then we will see a bastard opteron then later a 'S1206' like the change from S940 to 939 about a year after M2 is out.

 

[Vegitto]

Originally posted by: Sentential

Originally posted by: BouZouki

DDR2 support : Big mistake. DDR2 latencies are sky-high, and the lowest I've seen are 4-4-4-12. If AMD just stays with DDR, they can garner support from the enthusiast companies such as OCZ, Mushkin, Crucial, and even Corsair, all companies that are continuously improving their DDR series. OCZ just came out with its 2x1GB Gold VX PC-4000 RAM. By moving to DDR2, they will be losing support in the enthusiast and mainstream gamer communities, which account for much of their sales, apart from server-grade computers.

Wow just wow.

DDR2 timmings are well bellow 4-4-4-12 buddy, what if I told you I can get DDR2 running at 3-2-2 fairly easily.

Ya. Some people claim to have the facts when they clearly do not. The best balanced DDR2 available right now is DDR2 800 5-3-3-8 by Mushkin. This RAM can be pushed as high as 1050mhz (which I have done personally). So your rant about DDR2 being garbage is completely bogus.

In addition M2 will infact run MUCH hotter than S939 because the minimum amps given to the socket have shot up significantally. This will push generic M2 cores into the 115W range (as AMD's tech sheet suggests)

In addition the idiocy of AMD's managment is staggering. We will see a repeat of S754 with M2. I dont understand why they would only integrate the SB onto 1207 and not M2. Customers wont buy it and then we will see a bastard opteron then later a 'S1206' like the change from S940 to 939 about a year after M2 is out.

Exactly. That's why I, personally will not buy M2 if this integrated SB works out for S1207.

 

[coomar]

i don't think integrated pci-e bus is coming out on m2, it's for the chipset after it

 

[Brunnis]

Originally posted by: Sentential
In addition M2 will infact run MUCH hotter than S939 because the minimum amps given to the socket have shot up significantally. This will push generic M2 cores into the 115W range (as AMD's tech sheet suggests)

Can't help but notice that some of you guys actually seem to blame the socket for the supposed additional power draw. A chip that output 40-50W on S939 isn't suddenly going to spew out twice as much because of a socket change.

Anyway, you're making the mistake of using TDP to draw your conclusions. Just because the TDP is 115W doesn't mean that there will actually be any CPUs that are this power hungry.

 

[Viditor]

Originally posted by: Sentential
Ya. Some people claim to have the facts when they clearly do not. The best balanced DDR2 available right now is DDR2 800 5-3-3-8 by Mushkin. This RAM can be pushed as high as 1050mhz (which I have done personally). So your rant about DDR2 being garbage is completely bogus.

In addition M2 will infact run MUCH hotter than S939 because the minimum amps given to the socket have shot up significantally. This will push generic M2 cores into the 115W range (as AMD's tech sheet suggests)

In addition the idiocy of AMD's managment is staggering. We will see a repeat of S754 with M2. I dont understand why they would only integrate the SB onto 1207 and not M2. Customers wont buy it and then we will see a bastard opteron then later a 'S1206' like the change from S940 to 939 about a year after M2 is out.

1. At $225/GB for the Mushkin ram, it better be that good! The important point is that on a cost/performance ratio, the DDR2 latencies are still too high for an AMD platform...but (as indicated by the Mushkin ram) that should be changing by sometime next year.

2. M2 itself won't run hotter, the chips will because their clockspeed will be increased (this is the reason for the increase in amps). And it CERTAINLY won't be MUCH hotter (not sure where you get that...). Also, these are TDP numbers you are speculating about...the TDP of the Venice 3000 is 67w, but it actually only runs at 30w under peak load. If that follows on, then even a TDP of 115 will yield a peak load power of ~55w... (as Brunnis mentioned above)
One last on this point, AMD tends to give their entire line a TDP so that OEMs can plan for a longer period of time...

3. The reason that the Southbridge is going into the server line first is that it takes years to qualify a platform for servers...also, remember that they won't just immediately throw away their 90nm lines now that they're running at full efficiency. We should see SB function on the consumer chips (if they keep up their pattern) about a year or 2 after the socket F release. Remember also that AMD is able to increase their marketshare on servers much easier than on desktops because of Intel's marketing tactics...so it's probably not cost effective to release the 1207 for the desktop anytime soon.

 

[Hacp]

Well with the 65 Nm process, I think we can see around 3.2-3.4 GHZ 🙂.

 

[coomar]

probably, 90m, can probably support a 3 ghz, the smaller process should be able to handle those speed bumps

 

[Bona Fide]

Pfft...is AMD gonna play the GHz Game with Intel again? They should move to new technology like onboard PCIe and the like, rather than racing Intel for the 5GHz chip.

 

[Capt Caveman]

Originally posted by: Viditor

Originally posted by: Sentential
Ya. Some people claim to have the facts when they clearly do not. The best balanced DDR2 available right now is DDR2 800 5-3-3-8 by Mushkin. This RAM can be pushed as high as 1050mhz (which I have done personally). So your rant about DDR2 being garbage is completely bogus.

In addition M2 will infact run MUCH hotter than S939 because the minimum amps given to the socket have shot up significantally. This will push generic M2 cores into the 115W range (as AMD's tech sheet suggests)

In addition the idiocy of AMD's managment is staggering. We will see a repeat of S754 with M2. I dont understand why they would only integrate the SB onto 1207 and not M2. Customers wont buy it and then we will see a bastard opteron then later a 'S1206' like the change from S940 to 939 about a year after M2 is out.

1. At $225/GB for the Mushkin ram, it better be that good! The important point is that on a cost/performance ratio, the DDR2 latencies are still too high for an AMD platform...but (as indicated by the Mushkin ram) that should be changing by sometime next year.

2. M2 itself won't run hotter, the chips will because their clockspeed will be increased (this is the reason for the increase in amps). And it CERTAINLY won't be MUCH hotter (not sure where you get that...). Also, these are TDP numbers you are speculating about...the TDP of the Venice 3000 is 67w, but it actually only runs at 30w under peak load. If that follows on, then even a TDP of 115 will yield a peak load power of ~55w... (as Brunnis mentioned above)
One last on this point, AMD tends to give their entire line a TDP so that OEMs can plan for a longer period of time...

3. The reason that the Southbridge is going into the server line first is that it takes years to qualify a platform for servers...also, remember that they won't just immediately throw away their 90nm lines now that they're running at full efficiency. We should see SB function on the consumer chips (if they keep up their pattern) about a year or 2 after the socket F release. Remember also that AMD is able to increase their marketshare on servers much easier than on desktops because of Intel's marketing tactics...so it's probably not cost effective to release the 1207 for the desktop anytime soon.

My 1gb(512x2) of Patriot DDR2 for my Intel 640 could do 3-2-2-4 @ 667 for $180. I think that's pretty good cost/performance.

 

[Furen]

I have one question: is anyone actually SURE that the extra pins on socket F are there because of an integrated PCI-express bus? I mean, I read it on theInq, but that doesnt mean it's absolutely true. Couldn't it be the extra hypertransport links to connect to other CPUs that a regular M2 doesnt need?

If I had to speculate on what the pins on the different sockets actually are for I'd say the following:

Mobile socket S1:

638 pins

Some people (ie. Anandtech) claim it has dual DDR2 support, which I doubt since each DDR2 channel should require around 244 pins (where do I get this number from? The difference between socket 939 and 754 is 185 pins, and all you get is an extra 184pin DDR channel, I didnt feel like reading the socket 939 technical papers, so I'm not absolutely sure about this). Small form-factor is the name of the game for this socket and (it looks to me) it should need about half the surface area as an M2 socket.

Desktop/Workstation socket M2:

940 pins

Assuming that S1 does support dual-channel DDR2, then this socket has an extra 302 pins that are used for who-knows-what. If S1 does not support Dual-channel DDR2, then we still have extra pins, just not as many. I'd say it's hard to know what features will be included on this socket without knowing anything besides the pin count.

Server socket F:

1207 pins

I would guess that it will be just a socket M2 with extra hypertransport links to connect to other CPUs. It could also be, as Viditor said, a CPU with lots of FB ram channels (since the serial FB dimms should have a lower pin-count than regular parallel ram, I would guess a LOT of channels). I would expect this form factor to phase in in about a-year-and-a-half to two years (dont think corporate users will appreciate it if AMD jumps to another socket sooner than that) so FB modules should be in mass production by then.

 

[Viditor]

Originally posted by: Bona Fide
Pfft...is AMD gonna play the GHz Game with Intel again? They should move to new technology like onboard PCIe and the like, rather than racing Intel for the 5GHz chip.

Well, just like with DDR2, clockspeed is still a part of the formula...
I don't see Intel's goal being just clockspeed anymore, and of course AMD is the one that proved efficiency is more important. That said, it won't do any good to ignore a faster clock either...
IMHO, the next big "race" will be to find a production methodolgy with ultra-low leakage (probably nanotubes...).

 

[Viditor]

Originally posted by: Capt Caveman


My 1gb(512x2) of Patriot DDR2 for my Intel 640 could do 3-2-2-4 @ 667 for $180. I think that's pretty good cost/performance.

Was that stock, or did you tweak it? Do you think that the majority of the DDR2 sticks will do that yet? How does that compare to mainstream 1GB DDR 400 at ~$90?

 

[Capt Caveman]

Originally posted by: Viditor

Originally posted by: Capt Caveman


My 1gb(512x2) of Patriot DDR2 for my Intel 640 could do 3-2-2-4 @ 667 for $180. I think that's pretty good cost/performance.

Was that stock, or did you tweak it? Do you think that the majority of the DDR2 sticks will do that yet? How does that compare to mainstream 1GB DDR 400 at ~$90?

I tweaked it. At stock, the timings are 3-2-2-4 @ 533 and 4-3-3-12 @ 700. But if you bump the voltage from 1.8v to 2.0v, you can maintain the timings at higher speeds. The best was 3-2-2-4 @ 667 and 3-3-3-8 @ 700.

The Patriot XBLK uses the same chips as the OCZ Enhanced Bandwidth PC4200 but the OCZ is warranted at voltages up to 2.2v(I believe) and some folks were able to get tighter timings than I at the higher speeds.

At $90, you can find OCZ Value and other 1gb DDR2 kits with latencies at 4-4-4-8 but for a little bit more you get into the 3-3-3-8 territory at 533.

 

[Viditor]

Originally posted by: Capt Caveman

Originally posted by: Viditor

Originally posted by: Capt Caveman


My 1gb(512x2) of Patriot DDR2 for my Intel 640 could do 3-2-2-4 @ 667 for $180. I think that's pretty good cost/performance.

Was that stock, or did you tweak it? Do you think that the majority of the DDR2 sticks will do that yet? How does that compare to mainstream 1GB DDR 400 at ~$90?

I tweaked it. At stock, the timings are 3-2-2-4 @ 533 and 4-3-3-12 @ 700. But if you bump the voltage from 1.8v to 2.0v, you can maintain the timings at higher speeds. The best was 3-2-2-4 @ 667 and 3-3-3-8 @ 700.

The Patriot XBLK uses the same chips as the OCZ Enhanced Bandwidth PC4200 but the OCZ is warranted at voltages up to 2.2v(I believe) and some folks were able to get tighter timings than I at the higher speeds.

At $90, you can find OCZ Value and other 1gb DDR2 kits with latencies at 4-4-4-8 but for a little bit more you get into the 3-3-3-8 territory at 533.

Fair enough...I guess my point is that while we can see that DDR2 timings WILL be acceptable for an AMD platform in the near future, it's not quite there yet for the mainstream. IMHO, AMD has timed this just about perfectly...

Judging by the results you and some others here are getting on DDR2 tweaks, we should see DDR2 at 667 MHz and relatively low latency by Q1 06 in the ~$90/GB range.

 

[Brunnis]

Now that we're talking about DDR2 timings, I have one thing that's bothered me for a while. Wouldn't the higher clock frequency of DDR2 in large part make up for the slower timings? Example:

DDR400 @ CL2 = Two wasted clock cycles = 5 ns

DDR2667 @ CL3 = Three wasted clock cycles = 4.5 ns

Since memory latency is given in number of clock cycles, a higher memory frequency compensates for the slower access.

I'm sure you'll let me know if I've missed something crucial here. 🙂

 

[Shenkoa]

I think M2 is bad news, I have only had my current setup for a month, I am sick of spending loads of money just to have an ok system.

 

[Furen]

Originally posted by: Brunnis
Now that we're talking about DDR2 timings, I have one thing that's bothered me for a while. Wouldn't the higher clock frequency of DDR2 in large part make up for the slower timings? Example:

DDR400 @ CL2 = Two wasted clock cycles = 5 ns

DDR2667 @ CL3 = Three wasted clock cycles = 4.5 ns

Since memory latency is given in number of clock cycles, a higher memory frequency compensates for the slower access.

I'm sure you'll let me know if I've missed something crucial here. 🙂

That would be true if the two modules you mentioned actually operated at 400MHz and 667MHz, but they dont.

DDR 400 = Operates at 200MHz but transfers data on both the rising and falling edge of each clock.

DDR2 667 = Operates at 166MHz and transfers data twice during the rising and twice during the falling edges of each clock (or something of the sort).

So DDR 400 at CAS2T has a CAS latency of 10ns, while DDR2 667 at CAS3T has a CAS latency of 18ns, approximately.

 

[Hard Ball]

Originally posted by: Shenkoa
I think M2 is bad news, I have only had my current setup for a month, I am sick of spending loads of money just to have an ok system.

I don't think you have to worry too much. It seems that the M2 is really just a transitional socket that will be replaced soon after the arrival of K10. The only reason they are even contemplating it is because that K10 will arrive 6 or so months earlier on server workstation chips than on desktop chips; much like the timeline between the original Opteron and A64. DDR-II also seems only to be a transitional solution, with FBDIMM, or possibly DDR-III to be a much more permanent fixture.

M2 is the socket that most will be able to do without. It's going to have a shorter lifespan than that of even S754. And it will be replaced with S1207 or a socket with a similar pin-count less than a year after its inception in mid-06. Don't fret too much; S939 stuff probably will be supported throughout 06, and by the time 07 rolls around, the replacement for M2 would be close at hand anyway. So unless someons is coming from a very old setup (e.g. socketA or S478) there would be very little reason to even consider the M2.

 

[Bona Fide]

Originally posted by: Viditor

Originally posted by: Bona Fide
Pfft...is AMD gonna play the GHz Game with Intel again? They should move to new technology like onboard PCIe and the like, rather than racing Intel for the 5GHz chip.

Well, just like with DDR2, clockspeed is still a part of the formula...
I don't see Intel's goal being just clockspeed anymore, and of course AMD is the one that proved efficiency is more important. That said, it won't do any good to ignore a faster clock either...
IMHO, the next big "race" will be to find a production methodolgy with ultra-low leakage (probably nanotubes...).

Leakage? As in volt leakage? Speed leakage? Inefficiency?