My Gigabyte 8INXP arrived today!!

[Kowan]

Ajaykay
Thanks for the tips on possible sites. I'm back and forth on a 2.8 or 3.06 to go in the board.

Doesn't help seeing the Gigabyte ad here on the forums! LoL

 

[Ajaykay]

Mech:

Did a quick compression test for you with a 'warez' copy of UT that I had laying around and I'm happy for you that your score is still safe. Took me 2.5 minutes. I'm still keeping this 8INXP board, however.

I'll give it another go as I ramp up the overclock in the days ahead.

 

[Ajaykay]

Kowan:

I really wanted to pull out all the stops and go with the 3.06 w/HT. But at the show I went to, the best price anyone would give me on the chip was $750. I saw it at Newegg.com for around $650 the day before. If someone at this show would have matched that price, I swear I was crazy enough to buy it then with the 8INXP. I've been putting together the pieces of this system since August and I could wait another week for a mail order 3.06 but sanity finally took hold of me and I cutback and went for the 2.8 for $375. I'm clocking now at 3.16 or so but no HT. Oh well, some times you just gotta bite the bullet.

 

[Flakk]

mechBgon, it looks like you're trying to start a new standard benchmark with your Unreal Tournament Disk 1 and winzip compression idea.. I've seen it in multiple places. I have to put in my doubts about it's effectiveness, and standardness of result. Because you are basing your results on a test of file compression, won't it speed up or slow down depending on hard disk speeds too? And I doubt that is what you are intending to test.. The amount of memory the person has also would probably come into play, as the computer with enough ram during your test wouldnt be using virtual space, while the computer with say 128mb would most likely use it, slowing it down (and that would have nothing to do with the cpu / memory speeds if thats what you are testing).

 

[Flakk]

-- posted twice oops. --

 

[mechBgon]

Originally posted by: Flakk
mechBgon, it looks like you're trying to start a new standard benchmark with your Unreal Tournament Disk 1 and winzip compression idea.. I've seen it in multiple places. I have to put in my doubts about it's effectiveness, and standardness of result. Because you are basing your results on a test of file compression, won't it speed up or slow down depending on hard disk speeds too? And I doubt that is what you are intending to test.. The amount of memory the person has also would probably come into play, as the computer with enough ram during your test wouldnt be using virtual space, while the computer with say 128mb would most likely use it, slowing it down (and that would have nothing to do with the cpu / memory speeds if thats what you are testing).

That is a good and sensible observation. Here are my general findings so far:

• Hard drive speed is not important. I tested with two very different SCSI drives (15000rpm Cheetah X15-36LP versus ancient Quantum Atlas 10k) and the times were essentially the same.
• Hard drive interface is not important. My 15000rpm SCSI-equipped 1700+ on nForce220D gets beat by an overclocked 1600+ at 166/166/1750MHz on KT333, which has a much slower IDE hard drive.
• 256Mb versus 512Mb of RAM is not important. I tested with both 256Mb and 512Mb and the results were essentially the same.
• On KT333, memory frequency is not important by itself. I tested with the PC2100 and PC2700 memory speeds while keeping the CPU at 133MHz, and the results were essentially the same.
• An RDRAM-equipped system destroyed a DDR333-equipped system by a factor of over 2 to 1 with just a 153MHz speed advantage, according to the results by two other people on this forum, and that trend has proven not to be a fluke.
• AthlonXP systems do well for their MHz against DDR-equipped P4's (my 1.4GHz AthlonXP gave faster compression times than a 2.4B P4)
• Duron systems are handicapped on a per-MHz basis, presumably by their lower FSB and smaller L2 cache (no big surprise)
• nForce2 is faster than KT333 (no surprise). SpideyCU has an nForce2 dual-DDR system almost identical to my KT333/1600+OC system and got faster results.

So it looks like it boils down to CPU power, and the ability of the chipset and memory to keep the CPU fed with data and haul away the results. Since WinZip is a real-world application, I think the results are more relevant than abstract PCMark2002 or SiSoft Sandra/Samantha numbers. All the horsepower in the world won't do a car good if it can't get it to the ground (and make the corners too). Eh?

I found out it's possible to make two of the UT folder, and run two separate compression sessions at (almost) the same time, which would make a fascinating HT-versus-real-dualie benchmark. At any rate, I find the results interesting and I appreciate those who've taken the time to try this out.

 

[ArborBarber]

Elvis2...so you're still happy with the mobo? I hope you are, I'm glad you got your's first, this way I have someone I can ask questions about it to...anything other then RAID to look out for or did you solve that problem yet? Hopefully I'll be getting my serial HD soon...

 

[Flakk]

ok sounds good mechBgon.. I guess just make sure you are using the same version of winzip for all the tests.. because a newer version may have an improved or more efficient compression algorythm that would make it appear that things are lots faster when it's just the version of winzip.. but I'm sure you know that. What version of winzip are you using in all your tests?

That really surprised me about the Athlon XP 1.4 being faster than the P4 2.4Bghz........ what's up with that? It makes me think your test doesnt test all the aspects of the chip (a game must run faster on a 2.4 than 1.4) OR that Pentium 4's are much overrrated? What do you think?

True, winzip is a real world application test, but so are other tests, namely the # of frames per second in a game, and that is really what matters to me. So even if a 1.4 Athlon XP could *winzip* faster than a 2.4ghz p4, i would still take that 2.4 if it made my *games* faster....

so many tests.... so much confusion I may run your test when I get my granite bay motherboard. I plan to overclock to 166mhz bus speed, so that would theoretically give me 5.4gb/sec max transfer rate, which SHOULD be much more than a dual RDRAM system.

 

[mechBgon]

Originally posted by: Flakk
ok sounds good mechBgon.. I guess just make sure you are using the same version of winzip for all the tests.. because a newer version may have an improved or more efficient compression algorythm that would make it appear that things are lots faster when it's just the version of winzip.. but I'm sure you know that. What version of winzip are you using in all your tests?

That really surprised me about the Athlon XP 1.4 being faster than the P4 2.4Bghz........ what's up with that? It makes me think your test doesnt test all the aspects of the chip (a game must run faster on a 2.4 than 1.4) OR that Pentium 4's are much overrrated? What do you think?

True, winzip is a real world application test, but so are other tests, namely the # of frames per second in a game, and that is really what matters to me. So even if a 1.4 Athlon XP could *winzip* faster than a 2.4ghz p4, i would still take that 2.4 if it made my *games* faster....

so many tests.... so much confusion I may run your test when I get my granite bay motherboard. I plan to overclock to 166mhz bus speed, so that would theoretically give me 5.4gb/sec max transfer rate, which SHOULD be much more than a dual RDRAM system.

I'm using WinZip 8.1 here.

As for the 2.4B result, step back and look at it again. What should be shocking you is the crazy difference between PC1066 and DDR333 for a similar pair of P4 systems. The AthlonXP result is probably due to its high IPC, pure and simple.

I've also had some people run other intensive real-world animation and rendering benchmarks in Caligari trueSpace 4.3 Demo, using scenes I've made and animated, to see whether the RDRAM-equipped P4 system has a significant edge over an overclocked DDR P4 system (DDR400+ speeds). If you have the patience to wade through it all, you can read the full story here To be brief, in some areas it was beaten by raw MHz from a heavily-oc'ed 1.6A with DDR, in others it came out ahead.

 

[Flakk]

I'm not surprised that a DDR333 system is beaten by a PC1066 RDRAM system, as almost all rambus systems have been using dual channel, whereas all DDR motherboards up until the Granite Bay have been one channel. Here's a little info on the memory I found helpful:

DDR (one channel):
PC3200 (DDR400), 200MHz bus DDR, 3.2 GB/s max transfer rate
PC2700 (DDR333), 167MHz bus DDR, 2.7 GB/s max transfer rate
PC2100 (DDR266), 133MHz bus DDR, 2.1 GB/s max transfer rate
PC1600 (DDR200), 100MHz bus DDR, 1.6 GB/s max transfer rate

RDRAM (one channel):
PC1066, 533MHz bus DDR, 2.1 GB/s max transfer rate
PC800, 400MHz bus DDR, 1.6 GB/s max transfer rate


A system with PC1066 ram would be using it in dual mode (as almost all motherboards that use RDRAM use it in dual mode), therefore:
Dual PC1066 ram = (2.1gb/sec x 2 channels) = 4.1 gb/sec

A single channel Dual DDR333 solution:
Single DDR33 ram = (2.7gb/sec x 1 channel) = 2.7 gb/sec

So you were comparing 4.1gb/sec dual rdram to 2.7gb/sec single ddr, and thus of course the rdram is faster.


Now this takes us to why Granite Bay is so good. Because:
Dual DDR266 ram = (2.1gb/sec x 2 channels) = 4.1 gb/sec
which is equivalent to a dual PC1066 RDRAM solution, and if you overclock the bus on a GB board from 133mhz to 167mhz, you get the benefits of DDR333 memory, assuming your memory can handle that speed:
Dual DDR333 ram = (2.7gb/sec x 2 channels) = 5.4 gb/sec

Overclocked bus speeds on a Granite Bay system should OUTPERFORM a RDRAM solution, and *rougly* equal the speeds when not overclocked. Now you could overclock the RDRAM system too, but at least Granite Bay is the first to match it.

Hopefully that helps show it all.. Im not trying to be a know it all or have a big ego, so hopefully it didnt come across like that. Finally we have a solution that makes RDRAM outdated. Now, if only granite bay motherboards werent so darn expensive themselves If that's your concern then you can wait for Springdale or one of the other dual ddr solutions coming in the next few months, which should be both faster than Granite Bay, and less expensive.

 

[mechBgon]

Originally posted by: Flakk
I'm not surprised that a DDR333 system is beaten by a PC1066 RDRAM system, as almost all rambus systems have been using dual channel, whereas all DDR motherboards up until the Granite Bay have been one channel. Here's a little info on the memory I found helpful:

DDR (one channel):
PC3200 (DDR400), 200MHz bus DDR, 3.2 GB/s max transfer rate
PC2700 (DDR333), 167MHz bus DDR, 2.7 GB/s max transfer rate
PC2100 (DDR266), 133MHz bus DDR, 2.1 GB/s max transfer rate
PC1600 (DDR200), 100MHz bus DDR, 1.6 GB/s max transfer rate

RDRAM (one channel):
PC1066, 533MHz bus DDR, 2.1 GB/s max transfer rate
PC800, 400MHz bus DDR, 1.6 GB/s max transfer rate


A system with PC1066 ram would be using it in dual mode (as almost all motherboards that use RDRAM use it in dual mode), therefore:
Dual PC1066 ram = (2.1gb/sec x 2 channels) = 4.1 gb/sec

A single channel Dual DDR333 solution:
Single DDR33 ram = (2.7gb/sec x 1 channel) = 2.7 gb/sec

So you were comparing 4.1gb/sec dual rdram to 2.7gb/sec single ddr, and thus of course the rdram is faster.


Now this takes us to why Granite Bay is so good. Because:
Dual DDR266 ram = (2.1gb/sec x 2 channels) = 4.1 gb/sec
which is equivalent to a dual PC1066 RDRAM solution, and if you overclock the bus on a GB board from 133mhz to 167mhz, you get the benefits of DDR333 memory, assuming your memory can handle that speed:
Dual DDR333 ram = (2.7gb/sec x 2 channels) = 5.4 gb/sec

Overclocked bus speeds on a Granite Bay system should OUTPERFORM a RDRAM solution, and *rougly* equal the speeds when not overclocked. Now you could overclock the RDRAM system too, but at least Granite Bay is the first to match it.

Hopefully that helps show it all.. Im not trying to be a know it all or have a big ego, so hopefully it didnt come across like that. Finally we have a solution that makes RDRAM outdated. Now, if only granite bay motherboards werent so darn expensive themselves If that's your concern then you can wait for Springdale or one of the other dual ddr solutions coming in the next few months, which should be both faster than Granite Bay, and less expensive.

By "dual-channel" for the RDRAM, are you referring to the combination of the two 16-bit modules into a 32-bit data path? Because there IS a genuine dual-channel RDRAM chipset: Intel 860. i860 is real dual-channel RDRAM, and it only supports Xeons. Intel 850 and 850E are not dual-channel.

As for all the bandwidth data, I haven't checked your math but I assume it's all correct. Regardless, if you look up a few posts to Ajaykay's post, his 2.8GHz + Granite Bay compression time is about 1 minute behind that of a 2.53GHz P4 equipped with (single-channel) PC1066 RDRAM. It's no skin off my back one way or the other, since I'm not a P4 owner, but if they're going to hype this thing then I want to see if it lives up to it or not WinZip looks to be THE killer app for RDRAM.

 

[Flakk]

Hmm.. you bring up a good point about the 16 and 32 bit. I dont know how that affects the score, cause i dont know that much more about RDRAM, but the numbers i gave above were 16bit ones. I wonder how much faster 32-bit RDRAM is. I'll take a guess and say that single channel 32bit rdram is the same speed as dual channel 16bit rdram?

If Granite Bay does prove to be that much slower than rambus, that will be interesting to see... it definitely has been hyped to be a near equal performer. I suppose I will find out for myself when I get my Granite Bay board in a week or so I'll let u know for sure

 

[mechBgon]

16-bit RDRAM modules are used in matched pairs to form the 32-bit data path. 32-bit RDRAM modules combine both halves of the path onto one module. Not many boards have been designed around "RIMM3200/RIMM4200," with the Asus P4T533 being one of the first to come to mind. Even this is still single-channel RDRAM.

I agree that for most enthusiasts, Granite Bay would probably be a no-brainer choice if this prediction had come true:

Users looking at purchasing an all-in-one solution will love the P4G8X; Serial ATA RAID, Gigabit LAN, sound, FireWire, USB 2.0, and a dual channel chipset are excellent features for users that need as much as possible on a motherboard. VGA is just about the only thing missing from the P4G8X that all-in-one minded users might need. Though the price will initially be steep ($190), this is to be expected with a dual channel chipset, especially one from Intel. Prices will fall though, and we expect motherboards like the P4G8X to get as low as $140 in the next couple of months.

from Evan's E7205 (Granite Bay) article

As it is, the price is rather high.

I also admit that the WinZip benchmark is sort of an anomoly, but a fascinating one... what the heck makes it work SO much faster with RDRAM, even when dual-channel DDR *should* be faster based on the bandwidth numbers? :Q

 

[Flakk]

Even this is still single-channel RDRAM.

but at 32bit it is acting as the dual 16bit would, so in that respect the figures i gave above still hold. Now, what would be pretty interesting is if they made DUAL 32bit RDRAM....... where that would take us, would be quite nice (something like 8.4gb/sec?!? maybe that will happen when system busses can handle that much data without being saturated, but 4.2gb is the current limit with 533mhz bus systems.... and by the time 'dual 32bit rdram' could be used it would be probably outdated.... just lots of jumbled thoughts for u )

Unless you plan on doing a LOT of winzipping (some kind of a professional winzipper *laugh*) then you probably will want to include also more sensible, and more variety of real life benchmarks. Perhaps there is some explanation for that "anomoly".. which i'd be quite interested to hear... It really does come down to how it performs, so no benchmark can take that "truth" away

and yes, sadly, it appears the cost of Granite Bay is all that is holding it back for most users.

i had fun discussing this whole topic with you... it gave me a chance to practice my memory knowledge (of which i have developed only in the last 1-2 weeks... so i feel a bit new at it all still). Hope u didnt find it too taxing, and perhaps fun as well

 

[mechBgon]

Originally posted by: Flakk

Even this is still single-channel RDRAM.

but at 32bit it is acting as the dual 16bit would, so in that respect the figures i gave above still hold. Now, what would be pretty interesting is if they made DUAL 32bit RDRAM....... where that would take us, would be quite nice (something like 8.4gb/sec?!? maybe that will happen when system busses can handle that much data without being saturated, but 4.2gb is the current limit with 533mhz bus systems.... and by the time 'dual 32bit rdram' could be used it would be probably outdated.... just lots of jumbled thoughts for u )

Unless you plan on doing a LOT of winzipping (some kind of a professional winzipper *laugh*) then you probably will want to start looking at more sensible, and more of a variety of real life benchmarks. Perhaps there is some explanation for that "anomoly".. which i'd be quite interested to hear... It really does come down to how it performs, so no benchmark can take that "truth" away

Yeah as far as I can tell, the cost of Granite Bay is all that is holding it back for most users.

Ahem! A DUAL-CHANNEL 32-bit RDRAM PLATFORM EXISTS NOW. It is called Intel 860. Two channels of 32-bit RDRAM (no, not on one module, you use matching pairs of 16-bit modules on each of the two channels... same difference).

As for "sensible" real-life benchmarks, look at the results here and here. I don't see anything really spectacular... just two P4 platforms that perform about the same. Look through the rest of the benchmarks Evan did too. Only in an overclocked condition does GB start to show a big performance lead, and if that's your cup of tea, then go for it. The question is, how far do you have to OC it before you've got a faster system than if you bought a P4T533-C and used 16-bit PC1066 on it, and spent the leftover money on the next-faster CPU (or video card)? *shrug*

So there is my philisophical contribution for the night. Happy new year to all you AnandTechers!

 

[Flakk]

As for "sensible" real-life benchmarks, look at the results here and here.

So hopefully GB will perform pretty darn close to 850e solutions, if not QUITE as good. (thats why im going overclock *evil laugh*)

As for that intel 860, that sounds cool... are there any motherbaords that use it yet? or is it one of those things that we'll see in many months from now. I'm not that up to date on that chipset.. obviously, this is the first *i've* heard of it.

happy new years... im off to celebrate

 

[mechBgon]

Originally posted by: Flakk

As for "sensible" real-life benchmarks, look at the results here and here.

So hopefully GB will perform pretty darn close to 850e solutions, if not QUITE as good. (thats why im going overclock *evil laugh*)

As for that intel 860, that sounds cool... are there any motherbaords that use it yet? or is it one of those things that we'll see in many months from now. I'm not that up to date on that chipset.. obviously, this is the first *i've* heard of it.

happy new years... im off to celebrate

Yep, there are four of them by SuperMicro here, one from Tyan here, and you can dredge up even more of them from here. Fun! Hardcore stuff though, I doubt any of those boards are under $400.

 

[rhinohide]

Noticed knowledge micro has the MSI GNB on egay for $249.

 

[Jgtdragon]

Elvis2,
What is the DPS? Does it do anything? I remeber from a review on andantech that it doesn't help overclocking.

 

[Flakk]

i like it cuz it has pretty blue led lights on the DPS and I have a clear window on my case Perhaps liking it for the wrong reasons, but i dont mind

 

[caboob]

i think with the already excellent voltage regulation on the gigabyte boards, the dps card may be overkill but you cant have enough voltage stability when overclocked. I believe that user feedback on these forums will indicate that this board will have the best OC'd stability. Its too early right now to tell with reviewers getting hand-picked boards.

 

[MrEric]

While searching on ebay stores, I stumble on a store that has the regular Asus P4G8X for sale for $211.97. I don't know how to make a link but here is the address:

http://cgi.ebay.com/ws/eBayISAPI.dll?ViewItem&item=2074829707&category=1244

Edit: Nevermind, I just got an email back from him. He doesn't have them in stock.

 

[TheCoz]

Originally posted by: mechBgon
By "dual-channel" for the RDRAM, are you referring to the combination of the two 16-bit modules into a 32-bit data path? Because there IS a genuine dual-channel RDRAM chipset: Intel 860. i860 is real dual-channel RDRAM, and it only supports Xeons. Intel 850 and 850E are not dual-channel.

The 82850/82850E MCH and 82860 MCH both use a dual-channel design. They both have two 16-bit RDRAM channels. The 82850E MCH can use 32-bit RIMMs (that have two 16-bit channels on the same RIMM) or two 16-bit RIMMs (that have a single 16-bit channel on each RIMM) together, either way it's using two 16-bit channels in a dual-channel mode of operation. The i860 uses an MRH-R (Memory Repeater Hub) to split each RDRAM channel into two channels to double the maximum memory capacity - not the bandwidth. So, even though i860 has 4 memory channels they still operate off of two 16-bit RDRAM channels in the same way i850/850E does. So, effective bandwidth is the same (2 x 1.6GB/s = 3.2GB/s) even though memory capacity is doubled. The 82860 MCH does NOT use a quad 16-bit channel design, nor is it a dual 32-bit channel design either.

 

[caboob]

While searching on ebay stores, I stumble on a store that has the regular Asus P4G8X for sale for $211.97. I don't know how to make a link but here is the address:

http://cgi.ebay.com/ws/eBayISAPI.dll?ViewItem&item=2074829707&category=1244

Edit: Nevermind, I just got an email back from him. He doesn't have them in stock.

now the ah heck's jacked the price up to $265!!

 

[jpetermann]

Yeah, I tried ordering from him as well. Told me he had none in stock. I had purchased for the 2.11.xx and he told me AFTER I bought the board that he did not have them. Then, he called his "supplier" and emailed me back and said they would not be at the 2.11xx price, but it would be 25x.xx. He still had no idea when he would have them. Oh well, it was a nice try.

PJ