A64: 3200+ 90nm > 3500+ 130nm?

Kolath · Oct 11, 2004

I saw a link to this article on HardOCP today. It benchmarks a 90nm 939 3200+, 939 3500+, 757 3400+, and FX-53 and finds tat the 3200+ beats out the 3500+ and rivals the FX-53 when all are set at 2.5GHz. Since I've never heard of madshrimps.be before, is this stuff reliable? Would an OCed 3200+ really perform as well as this article claims?

Also, how much of an OC beyond 3.5GHz can the 3500+ be expected to get?

sleepyzman · Oct 11, 2004

yes Mad Shrimps is a reputable site

michaelkahl · Oct 11, 2004

i think the main question here on everyone's mind is, is this chip really beating the 3500+..that just seems silly?

d]g[ts · Oct 11, 2004

I would believe it is just showing that raising the speed of the CPU to 2.5GHZ on all CPU's Shows what gains you achieve from raising the memory speeds. the 3200 90NM runs 10x250 while the 3500 runs 11x227. (would like to confirm this) but would show why the 3200 does better when oc'ed

Kolath · Oct 11, 2004

So if it is just a matter of the different memory speeds, then I would assuem that the 3500+ should beat it if the FSB is raised to the same 250. My main question is, how much more of an overclock is the 3500+ capable of? Also, is there going to be a 90nm 3500+ anytime soon?

iNterBeiNg · Oct 11, 2004

I would assume that with memory and clock speed being equal the 3500+ and the 3200+ should have almost identical performance since they have the same amount of cache, of course the fx-53 would still beat all since it has twice as much.

I think what their really trying to get at with the article is the fact that right now the 90 nm 3200+ is a great deal for the amount of performance you can suck out of it, and its only going to get better as amd improves their 90 nm process.

cujo · Oct 11, 2004

d]g[ts said:
I would believe it is just showing that raising the speed of the CPU to 2.5GHZ on all CPU's Shows what gains you achieve from raising the memory speeds. the 3200 90NM runs 10x250 while the 3500 runs 11x227. (would like to confirm this) but would show why the 3200 does better when oc'ed

if you actually read it he states that multi for all cpus is 10. therefore the 3500 is not at 227 but at 250 like the rest of the cpus in the test.

d]g[ts · Oct 11, 2004

Thanks, did not see it when i read it. and i did state i would like to confirm that too.

Alacritan · Oct 11, 2004

At the same clockspeed they're almost the same chip, albeit with different cache sizes. However, the 3200+ at 90nm has had some unspecified architecture improvements made to it that the FX-53 doesn't have, which may be why the cache size difference doesn't play as big a role as you would think.

hitechjb1 · Oct 11, 2004

The wait for a good and cost-effective A64 CPU should be mostly over (except for dual-core which may be 939-upgradable, ...)

hitechjb1 said:
Low PR 90 nm 939 (Sept 2004)
90 nm 939 Winchester 3000+/3200+/3500+ 512 KB L2
These 90 nm 939 CPU with new revision D0 (with some core enhancement ?) should perform better than a 130 nm NewCastle at same clock frequency.

With a 939 motherboard, IMO, for new built, the 939 combo should be a better choice than a 754 system with a NewCastle, and even a 754 ClawHammer, especially taking into account for future compatibility and memory intensive applications. Pricewise, a 90 nm 939 system is also as cost effective as a then 754 system. A 90 nm 939 3200+ Winchester should be a good choice for a cost effective, high performance, high bandwidth, overclocking A64 system with AGP or PCI-e.

New motherboards with chipset (such as Nforce4 from Nvidia, K8T890 from VIA) for PCI-express support are planned towards end of 2004.

hitechjb1 said:
939 Price performance system (added Sept 2004)

- A64 939 3000+ (x9) or 3200+ (x10) 90 nm 512 KB L2
- 939 motherboard w/ PCI-e
- PCI-e video card Nvidia 6800/6600, or ATI X800/X700
- 2x512 MB DDR500 dual channel or overclock equivalent
- SLK-948U or XP-90 or XP-120
e.g. 3200+ Winchester, x10, HTT >= 250 MHz, memory (1:1) >= 250 MHz 1T, CPU >= 2.5 GHz

A64 CPUs, chipsets, motherboards

Petr · Oct 12, 2004

Alacritan said:
At the same clockspeed they're almost the same chip, albeit with different cache sizes. However, the 3200+ at 90nm has had some unspecified architecture improvements made to it that the FX-53 doesn't have, which may be why the cache size difference doesn't play as big a role as you would think.

90nm A64 (DH8-D0) has this improvements over 130nm (DH7-CG):
- improved DRAM page closing policy
- improved memory addressing with graphics cards using main memory (eg. integrated cards) as frame buffer
- memory controller power reductions (DDR receivers go off in default)
- memory power consumption reductions (CKE pins disconnect)
- second write combining buffer
- SAHF and LAHF instructions are now supported in 64bit mode

And it also has 8 errata fixed and has only 67W TDP for 3000+, 3200+ and 3500+ ratings (older chips have 89W TDP). All this make 90nm DH8-D0 about 1 percent faster clock for clock compared to 130nm DH7-CG.

Link: My review with benchmarks comparing clock for clock performance

BeerHunter · Oct 12, 2004

How much is it?

From what i've seen nothing can touch a 2800+ price to performance wise when overclocked to 2500Mhz. Until I know price I won't say but also 754 boards seem superior.

@/2ct!< · Oct 12, 2004

The semron 754 @ 2500mhz seems to be another good price performance

Petr · Oct 13, 2004

@/2ct!< said:
The semron 754 @ 2500mhz seems to be another good price performance

Sempron has too small cache what results in poor performance in some applications. Barton XP-M is better IMO.

BeerHunter · Oct 13, 2004

Petr said:
Sempron has too small cache what results in poor performance in some applications. Barton XP-M is better IMO.

Like what? This site has 3100 sempron tied with 2800 a64 plus it's $40 cheaper according to pricewatch.com. 30% less money for same performance? Sounds like an even greater bargain Mr, @/2ct!< is suggesting.

Petr · Oct 13, 2004

BeerHunter said:
Like what? This site has 3100 sempron tied with 2800 a64 plus it's $40 cheaper according to pricewatch.com. 30% less money for same performance? Sounds like an even greater bargain Mr, @/2ct!< is suggesting.

Average framerates (average performance) is not the whole story. For me lowest framerates matter. Here Sempron 3100+ is many times worse than Barton XP-M because of small cache. In many applications (mainly games) this is clearly visible by eye - like in Commanche 4 or Aquamark 3. Today, I would buy two chips - for price/performance XP-M 2600+ / 47W and for performance A64 3200+ s939 90nm.

hitechjb1 · Oct 13, 2004

I do not like the sempron due to its small L2 cache and non-64 bit for building my general purpose system. Even if its 64-bit was/could be enabled (I doubt), its equivalent cache size would be even smaller to run 64-bit OS/applications. Though it may look cheaper, but %-wise is small for the cost of the system, and it may be more costly in the long run, ....

In the last two years, the Tbred B 1700+ DLT3C JIUHB was a great price-perfromance and overclocking CPU, and the Barton XP-M 2600+ IQYFA/IQYHA has also been great and performed even better than the 1700+. The newly debut 90 nm 939 3200+ Winchester looks real good, especially when production ramps up and yield matures, I intend to get one, ....

BeerHunter · Oct 13, 2004

Wow. you're compareing a $255 chip to ones that cost $70-99 in thier prime. Hardly a good deal. I think those two cheap chips arn't but 20-25% slower even now everybody overclocked. I'm sorry but I disagree. The massive price premium, almost 300%, does'nt justify the performace even if you get it to 2600Mhz.

hitechjb1 · Oct 13, 2004

In the context of A64, many members have been paying $180 - $280 for a 754 2800+ to 3200+ to build an A64 system.

So a 90 nm 939 3200+ falls into that range, comparing to a 754 A64 and from some initial reviews, the 90 nm 939 Winchester seems to be a good chip (cooler, newer revision for slightly better performance) for building a 939 system. The eventual price premium, about $70-120 over an AXP, is for the 64-bit feature, low latency L2 to main memory (L3) which form a superior class of its own for a general purpose, multi-tasking system. $100 is not a huge % for the total cost of a system.

As for price concern, I would assume and wait a little bit after the initial price premium/hike due to supply, the 3200+ should be around $200. If price is really an issue, one could get a 90 nm 939 3000+ which is also doable, but not as "flexible" as a 3200+ in general terms.

A64: 3200+ 90nm > 3500+ 130nm?

Kolath

Registered

sleepyzman

Member

michaelkahl

Member

d]g[ts

Member

Kolath

Registered

iNterBeiNg

Member

cujo

Member

d]g[ts

Member

Alacritan

Member

hitechjb1

Senior Member

Petr

Member

BeerHunter

Member

@/2ct!<

Member

Petr

Member

BeerHunter

Member

Petr

Member

hitechjb1

Senior Member

BeerHunter

Member

hitechjb1

Senior Member

Similar threads