P or A

Sergio

Newcomer
Hi,

I´d like to ask you your opinion on an upgrade. If you could spend a few minutes I would mostly appreciated.

My current system is aging rapidly and is in need of a replacement. At the moment I have a P700 @ 756 MHz with 256 MB SDRAM and a GeForce2 Pro. I am planning on replacing only my motherboard, CPU and the system memory first and wait for the NV30 to appear, to decide what graphics card to buy.

As I see it now, I can go the P4 + 850 or 845 way or the Athlon XP with KT333 / SIS 648 route. If I understand all the articles and benchmarks on the net, the P4 + 850 combination will take the most out of the P4, but at a higher price. But the Athlon XP will give me the best bang for my Euro.
I mainly use my system for playing games and for occasionally compressing music or video files.

What would you recommend and what system seems to be the most future proof?

Thank you in advance!!

P.S. I just saw a thread on the dual DDR. Maybe I should opt for that solution. Do you have any date when it is going to be released?

Sergio
 
As things stand today, if you have the money, DEFINITELY go with P4 + i850E + PC1066.

In the future, not only will there be dual DDR but, I suspect, also quad DRDRAM (dual 32-bit channels) - no, Intel is *not* dropping DRDRAM altogether, Intel is simply promoting their dual DDR support more; their latest roadmaps show DRDRAM still in the 'performance' sector for the future, believe it or not. Keep in mind dual DDR needs a 128-bit FSB (WIDE!!) to achieve equivalent performance to a more feasible 32- or 64-bit DRD FSB...
 
There was another thread a while back that basically asks the same questions.

Basically, if you are interested in spending less money, the athlon will give you better bang for the buck. a 2000+ athlon was something like $100, while a 1600MHz P4 was something like $115. Otoh, the 2.2GHz P4 and 2200+ athlon were about the same price around like $200.

Really, no matter which way you go you'll be getting a good chip. If you decide to go with an athlon, I've personally had really good luck with SIS chipsets. It's what I've got in the system I'm typing this on. They are a bit slower than VIAs, but I've found it to be *very* stable, which is more than I can say for some of the VIA boards at work. I'd assume that any intel chipset for the P4s should be top notch as well. The only mistake you can really make at this point is to sacrifice other components to get a top of the line chip. Instead of buying a 2.8GHz P4 with rdram and a ti4200, get a 2.4GHz P4 with DDR and 9700, or Simply save the money. Paying premium prices for the current top of the line is pretty wasteful unless money is no object.

Nite_Hawk
 
IMO, wait a little longer to see that the landscape it like after Intel launch 'Granite Bay' (Dual Channel DDR).
 
DaveBaumann said:
IMO, wait a little longer to see that the landscape it like after Intel launch 'Granite Bay' (Dual Channel DDR).

Indeed. I consider it a must to wait for the Granite Bay boards. NDA will be lifted in mid-November, but the word on the street is that it will be even faster than a rambus-combo. 8) Anyway what is there not to like about a very fast Intel-chipset made for the workstation crowd? Price? Not a big thing when you can reuse your DDR 2100/2700! Just Yummi, me thinks... 8)
 
LeStoffer said:
NDA will be lifted in mid-November, but the word on the street is that it will be even faster than a rambus-combo. 8)
Well, going by our recent 845E review, unsing DDR266 and our upcoming 845GE review they've already got PC800 RDRAM's number, so I don't think there is going to be much doubt.

But think about it, assuming the dual channel bus is going to be as efficient at 850E's (which you would assume it is from Intel) then even if they limit it to DDR266 then its going to have the same peak bandwidth as 850E and PC1066 but without the latency.
 
The i850 isn't that far behind in latency with PC800 DRDRAM. I always forget why, but DDR's effective latency doesn't linearly scale with the clock, while DRDRAM's does. Furthermore, Ace's hardware did a test and found that PC1066 has 30% lower latency than PC800. What would be interesting is a new Intel DRDRAM chipset with the ability to keep more pages open and so on and PC1066 support.
 
AFAIK, the reason that RDRAM latency scales with frequency better than DDR SDRAM is mostly due to 2 factors:
  • At PC800, the DRAM array within the RDRAM runs at only 100 MHz. This means that row and column access strobes within the RDRAM must be aligned to this slow clock, which slows them down. Apparently, raising the internal clock to 133 MHz (PC1066) does not increase the number of clock cycles needed for these internal strobes, unlike what is the case when pushing DDR RAMS beyond 133 MHz. You probably run into trouble scaling this part of the latency beyond PC1066, though.
  • RDRAM signalling protocol overhead is substantially larger than the overhead for DDR RAM - this overhead scales linearly with cycle time.
 
except for...

the fact that pc 3200 rambus runs damn hot. and isn't known to be stable without cooling. It needs sinks and cooling. And it's expensive. Better bet to go with dual sticks of rambus. A little more expensive than one stick but a minute pct. of performance benchmarks show and it runs cooler. Also easier to find. Some really good pc800 can do 1066 anyways if it can 40ns?
 
arjan de lumens said:
AFAIK, the reason that RDRAM latency scales with frequency better than DDR SDRAM is mostly due to 2 factors:
  • At PC800, the DRAM array within the RDRAM runs at only 100 MHz. This means that row and column access strobes within the RDRAM must be aligned to this slow clock, which slows them down. Apparently, raising the internal clock to 133 MHz (PC1066) does not increase the number of clock cycles needed for these internal strobes, unlike what is the case when pushing DDR RAMS beyond 133 MHz. You probably run into trouble scaling this part of the latency beyond PC1066, though.
  • RDRAM signalling protocol overhead is substantially larger than the overhead for DDR RAM - this overhead scales linearly with cycle time.

Yep! And just FYI, 150 and 166MHz DRDRAM is already sampling. :)

DRDRAM's gonna be killer in the future - progressively higher bandwidth, progressively lower latency, AND a vastly lower trace count VS dual-DDR (hell even single DDR has twice as many traces as dual DRD). AND the termination is much, much simpler for DRDRAM.

Most of DRDRAM's continued presense will probably be due to motherboard makers - they REALLY like how simple its implementation is. Dual-channel DRD only needs 4-layer PCB's... whereas dual-DDR will almost surely need 8 layers or more, considering it needs a mind-bending 128-bit tunnel...
 
Tagrineth said:
Most of DRDRAM's continued presense will probably be due to motherboard makers - they REALLY like how simple its implementation is. Dual-channel DRD only needs 4-layer PCB's... whereas dual-DDR will almost surely need 8 layers or more, considering it needs a mind-bending 128-bit tunnel...

So you have never heard of Nforce? It's Nvidias dual channel DDR SocketA chipset, and it's been out there for quite a while. Nforce motherboards are 4 layer, they where even 4 layer back when dual channel rdram motherboards where all 6 layer. Now there are 4 layer rdram motherboards too, you might have seen them, they have the memory slots in two different places on the MB, because the traces can't pass over each other for electrical reasons.
 
Tagrineth said:
DRDRAM's gonna be killer in the future - progressively higher bandwidth, progressively lower latency, AND a vastly lower trace count VS dual-DDR (hell even single DDR has twice as many traces as dual DRD). AND the termination is much, much simpler for DRDRAM.

Superior architecture yes, but superior price/performance for the mainstream? It still have a long way to go. With Intel betting on dual channel DDR, I don't see SiS and Samsung making the kind of big bucks on Rambus that will hold it through a massive dual channel DDR II era... Did anybody say Betamax?
 
Back
Top