SiS645DX article - discussion?

FatBurger · Mar 11, 2002

The SiS 645DX - neither your sister, nor a 645 year old version of the 486. Discuss.

Quarter Pounder Inside
Don't step in the sarcasm!

pvsurfer · Mar 11, 2002

While the SiS645 appears to be a pretty good chipset, unless I'm misinterpreting the article's test setup specs, it wasn't a fair contest!

If I'm reading the memory spec (in the test setup) correctly, the i850 board had 2 x 128MB of PC800 RDRAM (for a total of 256MB) whereas the SiS645DX had 512MB of PC2700 DDR SDRAM. Why wasn't both platforms tested with the same total amount of memory?

God bless the USA

AMD_Man · Mar 11, 2002

The extra 256MB isn't going to affect those benchmarks.

AMD technology + Intel technology = Intel/AMD Pentathlon IV; the ULTIMATE PC processor

IntelConvert · Mar 11, 2002

That's my take on it as well. Unless there's a misprint - re: the amount of PC800 RDRAM and DDR333 SDRAM - the i850 platform did indeed operate at a disadvantage, especially in Sysmark.

aarona · Mar 11, 2002

They should have used a Northwood and bumped the FSB up to 133Mhz for all systems. I mean the DDR is already asynchronously overclocked at 166 and the RDRAM was just left there sitting at PC800. Not a fair review any way you look at it.

FatBurger · Mar 11, 2002

How is it not a fair review if left at stock speeds? That confuses me.

Quarter Pounder Inside
Don't step in the sarcasm!

aarona · Mar 11, 2002

Because the true potential of RDRAM is not being shown here by not running it faster. The benches are misleading to purchasers when the informed know PC1066 RDRAM leaps far ahead of this. It's an article made to make a new product shine a little brighter than it really does. Fatburger, you should know this after running a TH7II with a Northwood. There's another thing, most processor activity occurs in the cache so why have they used a Willamette here? Because DDR's slight advantage in latency will show up with the smaller cache processor when more activity is sent to ram. Tricks, tricks.

sdausmus · Mar 12, 2002

Perhaps when PC1066 RDRAM is released, we'll see a review and benchmarks for it. Until it is actually released, what is the point of pretending PC800 is PC1066 and reviewing accordingly? Hell, why doesn't someone go get a dual Itanium board and test it against AXPs and P4s, using the argument, "Pentium 5's will be as fast as dual Itaniums"?

camieabz · Mar 12, 2002

As you're obviously posting as the reviews come out, I'd like to know your opinion on these reviews as your first post.

~ Whew! Finished...Now all I need is a Cyrix badge ~

AMD_Man · Mar 12, 2002

lol, that's called overclocking! You can overclock DDR333 too. Comparing DDR333 to PC1066 is unfair because the PC1066 is overclocked, while the DDR333 isn't.

AMD technology + Intel technology = Intel/AMD Pentathlon IV; the ULTIMATE PC processor

FatBurger · Mar 12, 2002

I agree with AMD_Man, it's like saying that your car is the fastest when falling off a cliff, so if you don't push it off a cliff, you're not using all of it's potential.
Ok, it's nothing like that, but you get the idea.

Camie, my first priority is to get the post here, then I go read the article. In other threads, I've posted my thoughts. As far as this review goes...well, frankly I didn't find it very interesting. Don't know why, maybe because I only got 3 hours of sleep last night. I'll post some thoughts tomorrow if you want.

Quarter Pounder Inside
Don't step in the sarcasm!

Crashman · Mar 12, 2002

If anything, it was the SiS645DX at a disadvantage, do to it's slower than FSB asychronous memory clock. And it still won, except in the truely stupid memory-only benchmarks and a couple others. Fact is that the SiS winning REALLY makes Intel look bad.

But the thing that put DDR in the lead in this case was probably the lower cas latency of DDR compared to RDRAM. I am absolutely sure in this case that having the memory run at DDR400 would have made the SiS chipset STOMP the i850, this should send a clear message to Intel and memory manufacturers that it is high time for DDR-II, since no chipset manufacturer has the balls to release a dual-channel DDR solution.

Yes, Dual Channel PC1600 would outperform single channel PC2700 when paired with the P4, assuming the chipset addresses it properly, and such a chipset could easily use PC2100 for 133FSB overclocking, or PC2700 for 166FSB overclocking. But alas, we are stuck with single channel standard DDR, making PC2700 a mearly adequate solution.

What's the frequency, Kenneth?

somerandomguy · Mar 12, 2002

My take on the article was that the DDR and Rambus platforms are so close together it hardly matters. I'd buy based on cost, features and stability rather than performance of those platforms (not that I'm buying right now).

"Ignorance is bliss, but I tend to get screwed over."

Ncogneto · Mar 12, 2002

Did I here VIA just gag? This board may have implecations that will effect the chipsets motherboard makers choose to use for both Intel and AMD platforms. Its only a reference board, it was running its memory async, and was running its memory at CL2.5. would have been interesting to see what happened ( or if it was even possible) to run the cpu at a fsb of 166. I keep hearing how expensive it would be to implement dual channel DDR...Hogwash, nvidia does it with the nforce and its not that much more expensive.

It's not what they tell you, its what they don't tell you!

Crashman · Mar 12, 2002

I don't think you understand...even with these great scores, the memory is the bottleneck. So raising the FSB would do little to help. What is needed is to raise the Memory speed to 400MHz. But since that can't be done with currently available modules, Dual Channel is the only good option. Dual Channel would let you run your memory sychronously (same MHz, FSB at QDR 64bit, memory at DDR 128 bit) while overclocking a 1.6A to 2.66GHz/166FSB.

What's the frequency, Kenneth?

Ncogneto · Mar 12, 2002

Think about it. History has shown us in the past that running memory in async mode has resulted in a drop in memory performance. This phenom is readily apparant in other SIS and VIA chipset boards. I am merly suggesting that it would be interesting to see if the fsb of the p4 was adjusted to a 133 FSB if the memory scores would increase as well thus closing the gap between it and RDRAM.

It's not what they tell you, its what they don't tell you!

Crashman · Mar 12, 2002

I think in this case it's the BANDWIDTH that's the most important, after all, the i850 runs the memory at 8x the clock of the CPU bus, and its only shortcomming seems to be latency!
So the best performance increase would come from increasing memory bandwidth, which with todays memory means Dual Channel.

What's the frequency, Kenneth?

Ncogneto · Mar 12, 2002

the i850 runs the memory at 8x the clock of the CPU,

WOW! REALLY? Dam memory running at 16 gigahertz really is impressive!

It's not what they tell you, its what they don't tell you!

Ncogneto · Mar 12, 2002

Take a good hard look at the memory bandwith tests. Yes you will see the I850 leading, but by a very slim margin. All I am suggesting is that by running the DDR in sync with the processor this small margin might be made up.

It's not what they tell you, its what they don't tell you!

Crashman · Mar 12, 2002

I doubt it would help, because the memory bandwidth would be 1/2 the CPU bus bandwidth.

What's the frequency, Kenneth?

AMD_Man · Mar 12, 2002

History hasn't taught us that about asynchronous memory. You're making the wrong conclusions. PC133 + 200MHz FSB (100MHz DDR) did improve performance over PC100 quite a bit.

AMD technology + Intel technology = Intel/AMD Pentathlon IV; the ULTIMATE PC processor

slvr_phoenix · Mar 12, 2002

I'll post a similar thought that I posted in the similar topic in the CPU section:

Not only was the RDRAM platform at half the memory, but it also was not specified if the RIMMs were single or double sided. All considered on how this article was biased, I'd have to think that they ran double-sided RIMMs, which has been suggested have a slightly worse latency than single-sided RIMMs.

<pre>If you let others think for you, you're the only one to blame when things go wrong.</pre>

FatBurger · Mar 12, 2002

I think the larger amount of RAM would hurt more than the 2-5% reported hit taken by double-sided RDRAM.

Ncogneto, PC800 runs at an effective 800MHz (400MHz DDR), however the P4 bus is a quad-pumped 100MHz, meaning by the true numbers the RAM is 4x as fast, and by the effective numbers, the RAM is 2x as fast. Crash is off on that one (I don't get to say that very often).

Quarter Pounder Inside
Don't step in the sarcasm!

slvr_phoenix · Mar 12, 2002

Oh, I agree. My point wasn't that this would have made much of a difference, but that it could very well be yet another point in the unprofessional manner in which the article was handled. At the very least, they should have specified if the RIMMs were single or double sided. If THG feels that posting the CAS latency is important (which I agree fully on), then so should this be.

<pre>If you let others think for you, you're the
only one to blame when things go wrong.</pre>

FatBurger · Mar 12, 2002

Yes, I agree. I think it would be nice to have a brief readout of specs (just the bare minimum), and then have a link with a popup window for those who want incredibly detailed specs, including driver versions, BIOS settings, etc.

Quarter Pounder Inside
Don't step in the sarcasm!

SiS645DX article - discussion?

Illustrious

Distinguished

Splendid

Distinguished

Distinguished

Illustrious

Distinguished

Distinguished

Splendid

Splendid

Illustrious

Polypheme

Distinguished

Distinguished

Polypheme

Distinguished

Polypheme

Distinguished

Distinguished

Polypheme

Splendid

Splendid

Illustrious

Splendid

Illustrious

Share this page