Official AM2 Thread!

[LoneWolf121188]

Hey Dragon, when new news comes around, mind adding a date beside the article? It'de be nice to keep track of things as they occur.

*Agrees*

Why 333MHz HTT? Why not 400? I mean, a lot of us will be hitting 400 anyway, why doesn't AMD just do that from the start?

 

[Gautam]

There's not enough actual DDR2 that's capable of that. As of right now, maybe a couple of IC's, all relatively pricey. Doesn't make much sense to spec em at DDR2-800.

 

[c627627]

http://www.c627627.com/AMD/Athlon64/

I think Ed is right that there will not be intro level CPUs in June.

But I don't see dual core Semprons any time soon.

AMD talks about stock speeds and has asked for DDR II 800 to be tested on AM2s. Not all AM2s will run off of 800 so there may be both 667 & 800.

800 would be limited to top end CPUs for reasons OC Detective stated. So right now, the guess is that 800 would be FX only.

That fits nicely with 7 x 400 = 2800 MHz FX 62, so there's that.

Will there be .5 multipliers? Will other forms of DDR II be officially used in addition to 667 & 800?


If there are no .5 multis, then the FX follow up would be at 3.2 GHz and then what? .5s would fit 2.8 then 3.0 then maybe 3.2 FX over the next couple of years before DDR III arrives 2008ish.


So back to DDR II, with 667, 5000+ would be 2.667 GHz with half FXs cache maybe?


Keep an eye on those eng samples even though they were not what the final product turned out to be with 939s...

http://www.c627627.com/RAM/

 

[Sliver]

exactly how much faster are the DDR2 Athlons going to be than their equivilant DDR1 Athlons? The CAS latency and timings on DDR2 are utter crap compared to DDR1, so exactly how much faster does the DDR2 need to be to compensate for the crappy timings?

 

[c627627]

If you're looking at DDR I, at default speeds, PC2700 (166 FSB) even with "crap timings" is better than PC2100 (133 FSB) with better CAS latency and timings, right?

 

[Gautam]

There is DDR2 capable of some excellent timings. eg, this is what mine does in an Intel.

 

[inkfx]

I cant wait for the release. I hope I can save up enough money to get one when it comes out.

 

[Velocity]

AM2 will be my next big upgrade. But how long do you guys think it will take for a crossfire board to appear? Or Should I get SLI?

 

[Sliver]

If you're looking at DDR I, at default speeds, PC2700 (166 FSB) even with "crap timings" is better than PC2100 (133 FSB) with better CAS latency and timings, right?

At first I didn't know, so then I had a brilliant idea to find out for myself. So I went to change the timings of my memory (It's a PC2100 256 stick I bought for 30 bucks. It can do DDR 400 speeds at 2,2,2,5 timings and still have plenty of headroom.) and run some benchmarks to see. The first set at 133 with the timings tightened. Then run them at 166 at standard DDR2 timings. (5,5,5,15) And I discovered that my motherboard is an even worse piece of crap than even I'd expected. And doesn't let me do anything with the timings or anything remotely involving overclocking. So then, I still didn't know. And then I went to make a post in this thread and as I write these words I still don't know. But it's driving me nuts and I want to find out. So can someone do a quick test to find out the exact difference by testing their RAM at PC2100 speeds with DDR1 timings (2,2,2,5 and a CAS of 2 or 2.5) and then PC2700 at DDR2 (or as close as you can get) timings? I'd actually like to see the difference.

 

[RedDragonXXX]

Hey Dragon, when new news comes around, mind adding a date beside the article? It'de be nice to keep track of things as they occur.
In regards to clockspeeds...is what AMD has on Socket939 the most efficient A64 can do?
I kind of expect equivalent speed ratings like 4800+, for example, would have lower clock speeds than Socket939 (Say 2.1GHz instead of 2.4GHz).

All do my best. So far the articles above are all taken from within this week and the last pic was from awhile ago and just lists different cores.

I would have added more from the past but recently most of that stuff turned out to be false so there is no point in doing it.

In regards to clockspeds it's to early to tell cause we have yet to see a AM2 sample so we have no idea what to expect, but I'm sure that AM2 will be very similar to s939 in terms of clock speeds and OC potential if not better.

 

[oxid]

timings are a clock cycle delay, are they not?
so if the clock speed is higher, the delay gets smaller.
ie: with a delay of two clock cycles on 200 mhz, it would be the same as a delay of four on 400 mhz. so ddr2-800 can have double the timings as ddr400 and the delay in ns(nanoseconds) would be exactly thesame, and thus also the performance.
mhz is king

 

[Gautam]

Sliver, using DDR I & old Socket A, Front Side Bus (speed of communication between the microprocessor and the chipset) of 166 MHz will beat 133 MHz, regardless of timings.

166 FSB is better than 133 FSB, timings are just tweaking for often meaningless benchmark scores. At similar FSB speeds, timings may make a difference, although I've often wondered how much of a real life difference to tell you the truth...

Operating speeds: http://www.c627627.com/RAM/

In general, 2-2-2 to 3-3-2.5 is worth about 50MHz across the board. This is why in the early A64 days, I recommended nearly everyone to buy low latency memory. Not to get every last bit of performance, but because any decent amount of memory hit a rock wall around 220MHz. At least if you're running tight timings, you get around the same performance as running ~270 3-3-2.5. Though I can tell you folks this, low latency DDR2 will be next to impossible to find. Higher speed, higher latency parts are going to be the norm.

Once again I wouldn't expect DDRII-800 for the FX62. A handful out there like myself have high end low latency, low speed memory. I don't think AMD would screw that group as its the same group that would buy the FX.

 

[Sentential]

*Agrees*

Why 333MHz HTT? Why not 400? I mean, a lot of us will be hitting 400 anyway, why doesn't AMD just do that from the start?

That was the fastest ram that was approved by JEDEC when they devolped the controller.

http://www.c627627.com/AMD/Athlon64/

I think Ed is right that there will not be intro level CPUs in June.

But I don't see dual core Semprons any time soon.

AMD talks about stock speeds and has asked for DDR II 800 to be tested on AM2s. Not all AM2s will run off of 800 so there may be both 667 & 800.

800 would be limited to top end CPUs for reasons OC Detective stated. So right now, the guess is that 800 would be FX only.

That fits nicely with 7 x 400 = 2800 MHz FX 62, so there's that.

Will there be .5 multipliers? Will other forms of DDR II be officially used in addition to 667 & 800?


If there are no .5 multis, then the FX follow up would be at 3.2 GHz and then what? .5s would fit 2.8 then 3.0 then maybe 3.2 FX over the next couple of years before DDR III arrives 2008ish.


So back to DDR II, with 667, 5000+ would be 2.667 GHz with half FXs cache maybe?


Keep an eye on those eng samples even though they were not what the final product turned out to be with 939s...

http://www.c627627.com/RAM/

Thats fine but the ESs available right now are 333 x 8 = 2.66 with 2x1MB of cache.

There will be a 3200+ and 3500+ on release just as a side note

 

[oxid]

like I said, DDR2-800 with 4-4-4-10 timings is EXACTLY thesame delay in nanoseconds as DDR-400 at 2-2-2-5.

 

[Gautam]

Thats fine but the ESs available right now are 333 x 8 = 2.66 with 2x1MB of cache.

Now just how much are ya allowed to spill?

 

[Gautam]

like I said, DDR2-800 with 4-4-4-10 timings is EXACTLY thesame delay in nanoseconds as DDR-400 at 2-2-2-5.

No.

 

[Sentential]

like I said, DDR2-800 with 4-4-4-10 timings is EXACTLY thesame delay in nanoseconds as DDR-400 at 2-2-2-5.

I wholly disagree:

 

[oxid]

No.

then tell me why

 

[Sentential]

then tell me why

See my post for explination

 

[Gautam]

You made a very, very large simplification there. There are literally hundreds of latency parameters in communication between the CPU and memory. Several we have control over, tons of them we do not. "5-2-2-2" or "10-4-4-4" are just your typical standard four timings offered in most BIOSes. Just check out the latest A64 boards themselves, they offer control over around 50 parameters in themselves. Theoretically, if every single latency parameter were doubled, you'd be correct, but fact of the matter is, majority of them probably stay the same. User control is granted over some significant ones.