• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Why wait for Socket 939?

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

hkp0lice

Member
Joined
Mar 1, 2004
Location
Toronto
Ok, I realize that Socket 939 will be a more "mainstream" than Socket 940, therefore it costs less, but you'll still need Registered DIMMs right??

What's all the hype about? Registered Dimms cost a lot, and also adds latency, although I know that the bandwidth provided by the Dual Channel setup will more than makeup for it.

Am I the only person not hyped up about Socket 939?
 
The 64-bit CPU, system and OS are considered as a major new generation in hardwares, softwares, system technology (chipset, memory bus, HT, ...) and silicon technology (90 nm SOI), ....

I prefer the 939 platform for its

- 128-bit memory bus, twice the max memory bandwidth compared to 754, estimated faster than P4 dual channel QDR
- plus on-chip dual channel controller
- 939 allows use of non-registered, non-ECC memory (940 though has 128-bit memory bus and dual channel controller, but requires registered memory)
- future hardware compatibility (130 nm CPU then to 90 nm)
- upgrade ability and reusability of older 939 motherboard/CPU in the future if needed
- 940 motherboards are considered more costly to make than the 939 counterpart due to the pin layout of the 940 CPU

AMD have three CPU socket types for 64-bit CPU, 754, 940 and 939. The difference between them is not simply a matter of pin count, there are implications for each on performance, cost for components (CPU, motherboard, memory, ...), and upgrade path. IMO, I would only get the 939 CPU and motherboard (current plan).

Even cannot wait much further, I would get a first generation 939 CPU (hopefully a reasonable priced 130 nm CPU) and 939 motherboard, since 939 CPU's and motherboards would be reusable with latest hardwares down the road (e.g. next year).


This 939 platform memory bandwidth, as estimated from some test data (so result is preliminary), is impressive. Its efficiency is around 86-90%, which is 15-20% (to be confirmed with more 939 test data) better than the P4 QDR dual channel counterpart.

Its effective bandwidth (not max), running at the same memory bus speed, is about 15-20% higher than that of P4 QDR dual channel and 81-89% higher than that of 754 platform or nforce2 dual channel.

Estimation and importance of 939 platform memory bandwidth (page 19)
 
1.
hkp0lice said:
Ok, I realize that Socket 939 will be a more "mainstream" than Socket 940, therefore it costs less, but you'll still need Registered DIMMs right??


No, Socket 939 allows use of standard DDR memory instead of ECC registered DDR.


2.
hkp0lice said:
Am I the only person not hyped up about Socket 939?

No you're not the only one, others are not hyped up about it because of the expected disproportionate price to performance ratios when compared to Socket 754s.

Socket 939 is expected to be affordable in 2005 but before then, many poeple at the forums post they're waiting for PCI locked VIA K8T800 Pro (not K8T800) chipset or the PCI locked nForce3 250 (not 150) chipset which are due in May so they can use them with cheaper Socket 754 processors.



Note that Ed Stroligo wrote an article today here:
http://www.overclockers.com/articles1004/
which inexplicably ignores that all the leaked info up until now pointed to very few 939 chips being available in 2004 (relatively speaking) which was thought to mean astronomic prices for 939s in 2004 when compared to Socket 754s.

http://www.overclock.net/showthread.php?t=168
 
Re: Re: Why wait for Socket 939?

c627627 said:
1.



No, Socket 939 allows use of standard DDR memory instead of ECC registered DDR.


2.


No you're not the only one, others are not hyped up about it because of the expected disproportionate price to performance ratios when compared to Socket 754s.

Socket 939 is expected to be affordable in 2005 but before then, many poeple at the forums post they're waiting for PCI locked VIA K8T800 Pro (not K8T800) chipset or the PCI locked nForce3 250 (not 150) chipset which are due in May so they can use them with cheaper Socket 754 processors.



Note that Ed Stroligo wrote an article today here:
http://www.overclockers.com/articles1004/
which inexplicably ignores that all the leaked info up until now pointed to very few 939 chips being available in 2004 (relatively speaking) which was thought to mean astronomic prices for 939s in 2004 when compared to Socket 754s.

http://www.overclock.net/showthread.php?t=168

Ahhhh, I was mistaken then... It can use Normal DDR... I thought that would be a major drawback for future A64s.

As for the multiplyer lock issue, I thought the Gigabyte GA-8KNXP has it solved? http://www.neoseeker.com/Articles/Hardware/Reviews/gak8nnxp/

Thing is, when you look at current Athlon FX benches, they're not much faster than same clocked A64.. But I guess some of that could be blamed at the increased latencies Reg ram produces, and probably dual channel itself adds more latencies.
 
Wow, I just looked at that roadmap.. ONLY 3500+ and above are going to be Socket939?!? NO WAY.... AMD can't be that stupid.. How hard is it to make a dual channel chip?!? They're also cutting off 1/2 of the L2 cache?!? THIS IS RIDICULOUS

I'm 100% sure AMD will release some low end NewCastles, they have to bin everythin, can't just throw out chips that can't run 3500+...
 
Firt off, all future dates and PR ratings are speculation. AMD itself changes those things all the time, meaning AMD itself doesn't have an accurate internal roadmap, let alone us.

As for "I thought the Gigabyte GA-8KNXP has it solved..." That's nForce3 150 chipset which is not PCI locked.

This is why we should always post "nForce3 250 (not 150)" instead of just "nForce3 250" when talking about chipsets.
 
Since I have nf7-s rev 2.0 + a mobile barton, a7n8x-dlx + tbred b dlt3c 1700+, I think I can live with them until the dust and price settle for

939 CPU (price and yield to mature)
nforce3 250 gb (or better)
pci-e
64-bit OS
video card
early hardware, software, bios bug fixes
HT devices
raid supports
more features
...

If I get a nforce3 250 754 platform, the only thing it buys me is a more powerful cpu, since new peripheral devices are rare. Within 9-12 months from now, when the above things and features settle, most likely I would have to upgrade again, ....

For people with different hardware situation, sometimes waiting for another 9-12 months may be tough, ....
 
c627627 said:
Firt off, all future dates and PR ratings are speculation. AMD itself changes those things all the time, meaning AMD itself doesn't have an accurate internal roadmap, let alone us.

As for "I thought the Gigabyte GA-8KNXP has it solved..." That's nForce3 150 chipset which is not PCI locked.

This is why we should always post "nForce3 250 (not 150)" instead of just "nForce3 250" when talking about chipsets.

Not 100% true, if you read the anandtech preview of the 250, they stated that the 150 chipset DOES have working locks, but most mobo manufacturers didn't implement it correctly...

I think future games like HL2 and Doom3 may benefit more from Dual channel, but we'll have to wait 'n see... The fact remains that S939 chips are likely to remain expensive compared to P4's, soo.... *shrugs*
 
Not one PCI locked nForce3 150 model has been reported here.

If there is one, it would be very interesting to know what it is.

Also, again:
http://www.overclockers.com/articles1004/

"They'll try to get a premium price on socket 939 chips to start, then introduce a slower model or maybe two at lower prices, maybe one in June/July, another just before 90nm. Prices will probably not go below the $200 level."


$200?! We were talking about over $400 being a problem, I sure would like to know what he based $200 figure on for 939s...

But he did say "not go below $200".

$400 is not below $200.
 
So I assume that AGP locks are working, but PCI is not? I don't see how any AGP card would work in OC'd speed without a lock...
 
hkp0lice said:
Ok, I realize that Socket 939 will be a more "mainstream" than Socket 940, therefore it costs less, but you'll still need Registered DIMMs right??

What's all the hype about? Registered Dimms cost a lot, and also adds latency, although I know that the bandwidth provided by the Dual Channel setup will more than makeup for it.

Am I the only person not hyped up about Socket 939?
I am not hyped at all. Its dual channel memeory is truly pointless, since intel cuts the 1mb l2 cache single channel cpu to 512kb l2 cache dual channel cpu, you dont get a performance jump, they cut the cache to get another channel, plus the cpus at the same MHZ have the same PR rating because of the cache cut.

EDIT MAJOR WHOOOPS replace intel with AMD, I was a little:beer:
 
Last edited:
ok, first of all, the company is called AMD, not Intel. Second, there is not that much performance difference between 1mb and 512kb of L2 cache, tests were made the the increase in performance with more cache is negligble. Also they didn't cut half the cache to get dual channel, they cut half the cache because the FX will have 1mb.
On AMDs chips which have really short pipeline the amount of cache doesn't play a large role in overall performance, specially when it comes to sizes like 512KB and 1MB
 
The Coolest said:
ok, first of all, the company is called AMD, not Intel. Second, there is not that much performance difference between 1mb and 512kb of L2 cache, tests were made the the increase in performance with more cache is negligble. Also they didn't cut half the cache to get dual channel, they cut half the cache because the FX will have 1mb.
On AMDs chips which have really short pipeline the amount of cache doesn't play a large role in overall performance, specially when it comes to sizes like 512KB and 1MB

Sorry, but you're wrong...
Check out these benches:

http://hardocp.com/article.html?art=NTc1LDQ=

The only difference between the 3000+ and 3200+ is 512K L2, and there is a very noticeable difference. 700 3kmarks is noticeable in my book, justifying the extra 200+ in the 3200+ rating....
 
Actually that's only 3.7%. Well I looked at some other benches... And yes in some games, specially UT2003 the amount of cash has an affect on performance, and the 3000+ takes a pretty large performance hit compared to the 3200+.
 
Not with Athlon64s - it appears.

PR Rataing for 64s is not based on Thunderbirds because a 2.2 GHz Athlon 64 is roughly equivalent to a 2.8-2.9 GHz Athlon XP.
 
The Coolest said:
ok, first of all, the company is called AMD, not Intel. Second, there is not that much performance difference between 1mb and 512kb of L2 cache, tests were made the the increase in performance with more cache is negligble. Also they didn't cut half the cache to get dual channel, they cut half the cache because the FX will have 1mb.
On AMDs chips which have really short pipeline the amount of cache doesn't play a large role in overall performance, specially when it comes to sizes like 512KB and 1MB

Why dont they make the only difference between a FX and a S939 A64 the unlocked multiplier?

and it was a major typo, I ment AMD, Intel crossed my mind while I was writing it and I typed in intel and forgot to type in AMD. And I know intel isn't planing to ever have a built in memory controler in their CPUs.
 
...because they want to continue to charge $300-$400 more for the FX.

It would be tough to justify that if the only difference is the multiplier being locked/unlocked, besides almost all of their consumers do not overclock... so they couldn't care less about the multiplier.
 
Last edited:
Back