Understanding the Granite Bay Performance

[apoppin]

(Today's) Good Review at Aces

RDRAM still "wins" most of the time in the benchs and some likely 'reasons' are given.

Which one is the best? Granite Bay, the i845 PE, SIS648 or the i850E? Most people like a simple answer. Unfortunately the truth is - in our humble opinion - rather complex. Each chipset has some some advantage but also some serious disadvantages.

 

[apoppin]

Oh, come on . . . you guys really understand GB's oddly 'mixed' performance and shortcomings compared with RDRAM chipsets?

 

[Adul]

heh springdale should be better😛

🙂


and all can say to rdram i bye bye 😛

 

[CrazySaint]

Originally posted by: Adul
heh springdale should be better😛

🙂


and all can say to rdram i bye bye 😛

*waves goodbye to RDRAM* 😎

One thing that this review fails to mention is that when it comes to overclocking, GB smacks around RDRAM and makes it cry uncle 😉

 

[sharkeeper]

GB is good for those of us that use >3GB RAM in our systems! That would cost a fortune with the RAMbust. Next year people will be giving them (RDRAM) to trick or treaters--the bad ones that throw eggs at your car! 😛

Cheers!

 

[apoppin]

Originally posted by: sharkeeper
GB is good for those of us that use >3GB RAM in our systems! That would cost a fortune with the RAMbust. Next year people will be giving them (RDRAM) to trick or treaters--the bad ones that throw eggs at your car! 😛

Cheers!

You have extra RDRAM? :Q

Can I egg your car? 😀

The fact of the matter is, RDRAM still gets the best (desktop) performance . . . well, at least until GB gets the bug ironed out (or Springdale).

 

[Athlon4all]

One thing that this review fails to mention is that when it comes to overclocking, GB smacks around RDRAM and makes it cry uncle

True, but still

Oh, come on . . . you guys really understand GB's oddly 'mixed' performance and shortcomings compared with RDRAM chipsets?

I certainly didn't. This certainly does clear up some things.

I think that Ace's made very good points. I personally think tho that 845PE is close enough to E7205/850e to make the cost of 850e/E7205 not worth it for most. I would find it very difficult if I were buying a new P4 rig to go 850e/E7205 over 845PE. If I did tho spend money on 850e or E7205, and were having only 512MB, I would go for 850e. But if I were getting like 1GB+, it would be E7205 for sure

 

[MrGrim]

When I upgraded from my 486 to a PII I bought PC66 SDRAM.
When I upgraded from my PII to a PIII I bought PC100 SDRAM.
When I upgraded from my PIII to a Thunderbird I bought PC133 SDRAM.
When I upgraded to my P4 I bought RDRAM.

I am currently enjoying the fastest chipset/memory combo on the market. Do I like RDRAM? No it's just a type of memory. Do I hate RDRAM? No it's just a type of memory.

 

[andreasl]

I wonder where all those people are who used to claim that GB wouth outperform the 850E with a margin of 5-10% due to lower latency? Hasn't anyone's pride been hurt even just a little over this?

BTW, I'm not a Rambus supporter, I'm just curious why the reaction has been totally left out. Seems no one is even a bit interested in learning why it didn't happen..

 

[apoppin]

Originally posted by: andreasl
I wonder where all those people are who used to claim that GB wouth outperform the 850E with a margin of 5-10% due to lower latency? Hasn't anyone's pride been hurt even just a little over this?

BTW, I'm not a Rambus supporter, I'm just curious why the reaction has been totally left out. Seems no one is even a bit interested in learning why it didn't happen..

It's always been this way at General. That's partly why I posted this unpopular thread. 😉

🙂
(the other reason is that the article is really interesting and really gets into the nitty-gritty of RDRAM vs. dual DDR performance (in a very understandable manner and even manages to hint at the "whys" that Ace's does so well)

 

[andreasl]

Originally posted by: apoppin

Originally posted by: andreasl
I wonder where all those people are who used to claim that GB wouth outperform the 850E with a margin of 5-10% due to lower latency? Hasn't anyone's pride been hurt even just a little over this?

BTW, I'm not a Rambus supporter, I'm just curious why the reaction has been totally left out. Seems no one is even a bit interested in learning why it didn't happen..

It's always been this way at General. That's partly why I posted this unpopular thread. 😉

🙂
(the other reason is that the article is really interesting and really gets into the nitty-gritty of RDRAM vs. dual DDR performance (in a very understandable manner and even manages to hint at the "whys" that Ace's does so well)

Yeah I have been waiting for this review from Aces for a long time, since no other review that I read even attempted to explain the 'strange' performance of GB. I guess this separates the professional hardware reviewers from the rest wannabe sites.

It also seems to be that current memory latency tests cannot be trusted to give accurate information of what is really going on. Not that I'm saying these tests give inaccurate information. Just that there seems to be a whole host of unknown factors at work here that no one seemed to be aware of and these programs won't display. Maybe things aren't so cut and dry as some sites like us to believe (AT included).

I know I'll look differenty at reviews from sites like www.hardwarerox0rs.com from now on.. 🙂 (don't bother look it up as it doesn't exist)

 

[apoppin]

Originally posted by: andreasl

Yeah I have been waiting for this review from Aces for a long time, since no other review that I read even attempted to explain the 'strange' performance of GB. I guess this separates the professional hardware reviewers from the rest wannabe sites.

It also seems to be that current memory latency tests cannot be trusted to give accurate information of what is really going on. Not that I'm saying these tests give inaccurate information. Just that there seems to be a whole host of unknown factors at work here that no one seemed to be aware of and these programs won't display. Maybe things aren't so cut and dry as some sites like us to believe (AT included).

I know I'll look differenty at reviews from sites like www.hardwarerox0rs.com from now on.. 🙂 (don't bother look it up as it doesn't exist)

www.hardwarerox0rs.com








































😀 OK . . . j/k'ing
Anyway, Aces has some very in depth analysis not found on a lot of other popular tech sites - part of this they have credited to their forum (think we're "geeks" here?) 😉

 

[MadRat]

Originally posted by: andreasl
I wonder where all those people are who used to claim that GB wouth outperform the 850E with a margin of 5-10% due to lower latency? Hasn't anyone's pride been hurt even just a little over this?

BTW, I'm not a Rambus supporter, I'm just curious why the reaction has been totally left out. Seems no one is even a bit interested in learning why it didn't happen..

Bufferage.

 

[dexvx]

Originally posted by: andreasl
I wonder where all those people are who used to claim that GB wouth outperform the 850E with a margin of 5-10% due to lower latency? Hasn't anyone's pride been hurt even just a little over this?

BTW, I'm not a Rambus supporter, I'm just curious why the reaction has been totally left out. Seems no one is even a bit interested in learning why it didn't happen..

Its because people just dislike Rambus as a company and want to see them lose the performance crown. PC800/i820, i840 RDRam was barely faster than PC133/i815e and people expected DDR to blow RDRam out of the water. However, PC1066/i850 smacks DDR/i845 around, even when overclocked, and then people were waiting for DC DDR to smack PC1066. Even though i expected DC PC2700 to beat PC1066/i850e solidly. looks like that hasnt happened either.

People just underestimated RDRam, and they underestimated it by a very very large margin.

 

[oldfart]

The appeal for Granite Bay is for the overclocker. With the sky high FSB overclocks many are getting with P4s, the DC DDR is the way to go. With RDRAM, you can hit the limit of available memory speed well before you hit the limit of the CPUs FSB potential. With GB, you can run as high as you want without that limitation. Even if you overclocked to an insane 200 MHz FSB, all you would need is commonly available PC3200 to do it.

 

[formulav8]

What you all have to understand is that the P4 was created and designed around Rambus Ram. So therefore that should be the optimum combo.




Jason

 

[JackBurton]

SWEET! I just checked the price on a 512MB stick of 32bit PC1066 and it was $248! Not bad at all. If I was going to buy DDR memory, it was going to be Corsair XMS PC3500 CL2 (Platinum) which runs about $219 (512MB stick). Now given the price of GB motherboards (~$200+) it looks like the RDRAM solution actually comes out cheaper (Asus P4T533 = $172). The article also mentions DDR memory having problems running at very low timings with multiple DDR sticks, even with PC3500 sticks. Also, and correct me if I misunderstood, the P4 they were using was unlocked so I'm assuming they were trying to achieve the highest FSB possible with the GB boards. In that test the Gigabyte board reached the highest FSB with 168MHz. Not too exciting if you ask me. The BIG selling point of GB is the ability to use up to 4GB of memory, but when you start adding multiple sticks, you'll have to be less aggressive with your memory timing and consenqently loosing performance. With the 32bit PC1066 RDRAM you should be good up to 150MHz FSB (PC1200), and the performance keeps coming! With that said, I think I'll take an Asus P4T533 please...well until Springdale. 🙂

 

[andreasl]

Originally posted by: MadRat

Originally posted by: andreasl
I wonder where all those people are who used to claim that GB wouth outperform the 850E with a margin of 5-10% due to lower latency? Hasn't anyone's pride been hurt even just a little over this?

BTW, I'm not a Rambus supporter, I'm just curious why the reaction has been totally left out. Seems no one is even a bit interested in learning why it didn't happen..

Bufferage.

Sorry? What does that mean?

What you all have to understand is that the P4 was created and designed around Rambus Ram. So therefore that should be the optimum combo.




Jason

Yeah, can you back that up with a technical argument?

 

[MadRat]

The Granite Bay should have better performance based on the theoretical latency of DDR RAM versus RDRAM. The fact that it does not leads me to believe that Intel does not adequately buffer information coming and going from the memory to the chipset. What must be happening is that the memory is getting backlogged with requests as the current information in the chipset buffer makes its way to the processor. Since Granite Bay is more of a proof of concept product at this time then end all design expect Springdale to not suffer these same penalties.

 

[andreasl]

Originally posted by: MadRat
The Granite Bay should have better performance based on the theoretical latency of DDR RAM versus RDRAM. The fact that it does not leads me to believe that Intel does not adequately buffer information coming and going from the memory to the chipset. What must be happening is that the memory is getting backlogged with requests as the current information in the chipset buffer makes its way to the processor. Since Granite Bay is more of a proof of concept product at this time then end all design expect Springdale to not suffer these same penalties.

Do you have any proof to support this claim? AFAIK GB supports a 12 deep in-order issue que just like the 845E (and derivates) does. That's more than the 850E which only supports 8. It also supports more open pages than the 850E.

Maybe it's the whole concept of "DDR has lower latency than RDRAM" that has to be revised instead. I'm sure it's true in some sense but it seems to depend on the workload. If there are lots of combined read and writes going on at the same time then RDRAM seems to have a distinct edge over DDR. And seeing how modern CPUs will always have more than 1 outstanding read combined with writeback from the L2 cache all at the same time, maybe that is more of a typical case than what these latency benchmarks attempt to display?

Also keep in mind that the 850 chipset is 2 years old by now, and it's using the same memory controller essentially as the old 840 chipset which is even a year older. I wonder how a modern RDRAM chipset, that the designers went all-out to get max performance, would perform. Some of the engineers who worked on the EV7 didn't have much high regard for the 850E in terms of taking advantage of RDRAM features.