brisbane dual core 65nm

[Zimon]

Thankyou very much Mr Presley, though I'm not entirely sure how your comments relate to my post......?
- I'm aware of Anand's cache latency findings, but don't believe they are soley responsible for the diifferences that are being reported here.

I'm astounded that AMD haven't tried to explain this to the reviewers they are sending samples to; They can't expect a performance decrease to go unnoticed, surely?
To be fair though, these sites should really be comparing like for like: Interesting as those Firingsquad benches are, they should have dropped the multi on their 4800+ for a direct comparison with the 90nm 4600+, much as Anand did with the 5000+ in their part 2...

[Serge84]

I really can't get to grips with why AMD have started to use these half multipliers....

I suspect they are responsible for most of the performance loss reported by both Anand and Firingsquad with those 4800+ chips:
As far as I understand, they cause a lower RAM speed, and a corresponding ~4% loss in theoretical bandwidth.

Interestingly, the difference in clockspeed between 4600+ and new 4800+ chips is also 4%...
- Maybe we're just seeing differences between MHz and bandwidth sensitive apps and games...?

Seems this would explain how the 4600+ is able to beat the 4800+ in some of the Firingsquad tests?

It fixes memory divider issues is all. We have had half multipliers for years anyways. The only way I can get 354x8.5M is because of the half multiplier. The damn memory controller in mine is limited by the memory dividers and full multipliers and half multipliers break the berrior so its a good thing for some.

[Shintai]

what the duce, that page will not be displayed until after you accept half a dozen tracking cookies

Run from the subject?

[Shintai]

About the half multiplier and the memory controller it would look like this:

GHz - DDR2-800 - DDR2-667 - DDR2-533 - DDR2-400
2.5 - DDR2-714 - DDR2-625 - DDR2-500 - DDR2-400
2.4 - DDR2-800 - DDR2-600 - DDR2-533 - DDR2-400
2.1 - DDR2-700 - DDR2-600 - DDR2-525 - DDR2-400
2.0 - DDR2-800 - DDR2-667 - DDR2-500 - DDR2-400

So worse memory bandwidth, 66% more L2 latency and K8 suddenly drops 10-15% in games.

[savantu]

and where does it say that it will not work in a am2 socket ?

http://www.xbitlabs.com/news/cpu/dis...226153755.html

AMD is projected to introduce new family of desktop microprocessors which are code-named after stars and use the new code-named K8L micro-architecture will be introduced in the third quarter of 2007, according to a recent roadmap of the chipmaker. Those chips are made using 65nm process technology and the family is set to include both dual-core and quad-core microprocessors, though, it is uncertain whether K8L lineup includes single-core central processing units (CPUs) too. It is believed that micro-architectural enhancements of the K8L will allow AMD to more successfully compete against Intel Core 2 family of chips.

The new desktop chips will feature AM2+ form-factor and will not be drop-in compatible with existing infrastructure. By the Q4 of 2007AMD’s product mix will include 20% of AM2+ processors, while by the end of Q1 2008 there will be “above 60%” of AM2+ processors among all chips shipped by AMD, a news-story by DigiTimes web-site claims.

[The Ghost]

http://www.xbitlabs.com/news/cpu/dis...226153755.html

whoopy do

and they got their info from here which states none of that

AM2 to account for 95% of AMD shipments in 1Q

Yen Ting Chen, Taipei; Rodney Chan, DigiTimes.com [Tuesday 26 December 2006]

Socket AM2 will account for 95% of AMD's CPU shipments in the first quarter of 2007, but the chip supplier will begin shifting its focus to AM2+ in the third quarter, according to AMD's product roadmap.

Socket 1207 will still be in the market in the first and second quarter, accounting for 5% of AMD's CPU shipments, while socket 939 will be discontinued by the end of 2006, according to the chipmaker's plan.

In the third quarter of 2006, AMD will introduce its Stars family of 65nm quad-core CPUs that use socket AM2+, whose proportion of the chipmaker's shipments will climb to about 20% by the fourth quarter, and above 60% by the end of the first quarter in 2008, according to the roadmap.


and guess what it says nothing like that where they got their info from


try again ????????

i just love it when writers make stuff up and call it news, and because it fits what some one wants , they call it news

[informal]

a 65nm 4800 is 2.5ghz with 512kb cache, so not its not the same as the 90nm part. from the firing squad review, the memory divider is probably why the 4800 is slower than the 4600 at times, as the 4600 uses a normal multi.

I agree 100% with you.The divider issue is the reason why 65nm is SLIGHTLY slower than 4600+.AMD actually made some changes to the L2 while maintaining the performance within the ~3 or 4 % of the revF which is very good thing.Also we should note that the 65nm part was faster in variaty of tests than 4600+,altough having 20 cycles L2 and sligtly higher clock

[Shintai]

I agree 100% with you.The divider issue is the reason why 65nm is SLIGHTLY slower than 4600+.AMD actually made some changes to the L2 while maintaining the performance within the ~3 or 4 % of the revF which is very good thing.Also we should note that the 65nm part was faster in variaty of tests than 4600+,altough having 20 cycles L2 and sligtly higher clock

Slightly? Did you even check the review? We talk 10-15% in games!

[Shintai]

I think he was speaking clock for clock, not model for model.

(A few FPS numbers as example)
X2 4600+ 90nm, 2.4Ghz
FPS: 108, 104, 115, 114

X2 4800+ 65nm 2.5Ghz
FPS: 100, 95, 101, 100

[zir_blazer]

(A few FPS numbers as example)
X2 4600+ 90nm, 2.4Ghz
FPS: 108, 104, 115, 114

X2 4800+ 65nm 2.5Ghz
FPS: 100, 95, 101, 100

You're not doing a correct comparision, because the non integer Multipliers boast an even slower Memory Frequency. You should compare models that have the same specifications on 90nm as in 65nm, for example, the same 90nm A64X2 4600+ against a 65nm A64X2 4600+, not a 4800+ with 100 MHz more but with slower Memory Frequency.

[Shintai]

You're not doing a correct comparision, because the non integer Multipliers boast an even slower Memory Frequency. You should compare models that have the same specifications on 90nm as in 65nm, for example, the same 90nm A64X2 4600+ against a 65nm A64X2 4600+, not a 4800+ with 100 MHz more but with slower Memory Frequency.

Slower depends on the memory used.
GHz - DDR2-800 - DDR2-667 - DDR2-533 - DDR2-400
2.5 - DDR2-714 - DDR2-625 - DDR2-500 - DDR2-400
2.4 - DDR2-800 - DDR2-600 - DDR2-533 - DDR2-400
2.1 - DDR2-700 - DDR2-600 - DDR2-525 - DDR2-400
2.0 - DDR2-800 - DDR2-667 - DDR2-500 - DDR2-400

But even so, the move from 939 to AM2 showed K8 aint bandwidth limited. The same with the 4x4 losing on all fronts, tho it got more memory bandwidth. Its mainly about the 66% increased cache latency.

[Serge84]

http://badhardware.blogspot.com/2006...70453330892136

Well this debate that another mobo is needed is just silly. AMD doesn't have crappy chipset/memory controller issues like intel. According to this AM2 is comattible with K8L at a limited HTL. If it doesn't get through this time then I give up. BTW its not that hard to OC the HTL like to 1500mhz or 1700mhz. I done it.

http://www.hkepc.com/bbs/news.php?ti...me=0&endtime=0

http://www.hkepc.com/bbs/itnews.php?...me=0&endtime=0

This is also confirmed by dalytech btw.

http://www.google.com/search?hl=en&s...th+K8L&spell=1

By all means look for the proof here all day and night.

http://www.dailytech.com/article.aspx?newsid=3169

How many links does one have to show to prove a point BH's.

[informal]

Shintai,the whole point is that memory clock greatly affects the Brisbane's results in some tests.In others,Brisbane's higher clock(2.5Ghz) manages to compensate for lower clock of the memory.
On the other hand,the increased L2 latency is probably compensated(to some degree,remember the latency rosed 66%!!!) by some minor tweaking in the core itself.
So don't try to say that Brisbane is "worse" cpu.It's great value and adds more by giving us sligtly higher multi.People interested in OCing new 65nm RevGs will probably use higher spec. memory modules capable of 1GHz+(ddr) so the problem with "new" stock half multi's will be "eliminated" with OCing the chips to their limits.
Normal folks care about "noise" and power consumption and yes,performance,all of which Brisbane(and Windsors) posses.The fact that they are not any more the best choice for enthusiasts since intel introduced New Core after years of being embarrassed doesn't mean no one would buy AMD's chips any more.In fact,AMD can't sell enough

[Serge84]

Shintai,the whole point is that memory clock greatly affects the Brisbane's results in some tests.In others,Brisbane's higher clock(2.5Ghz) manages to compensate for lower clock of the memory.
On the other hand,the increased L2 latency is probably compensated(to some degree,remember the latency rosed 66%!!!) by some minor tweaking in the core itself.
So don't try to say that Brisbane is "worse" cpu.It's great value and adds more by giving us sligtly higher multi.People interested in OCing new 65nm RevGs will probably use higher spec. memory modules capable of 1GHz+(ddr) so the problem with "new" stock half multi's will be "eliminated" with OCing the chips to their limits.
Normal folks care about "noise" and power consumption and yes,performance,all of which Brisbane(and Windsors) posses.The fact that they are not any more the best choice for enthusiasts since intel introduced New Core after years of being embarrassed doesn't mean no one would buy AMD's chips any more.In fact,AMD can't sell enough

AMD is 2nd best doesn't mean its a bad choice is right. Its still a very good chip 3rd to none. Atleast they got the power consumption way down.

[Serge84]

Hey, I'll take an Intel chipset over ANY chipset.

There is simply no parallel to them in terms of both stability and performance. How do you think C2D can do so well without an IMC?

Be careful with the blanket statement b/c some of them are just not true.

The best chipsets for C2D in the world are the NF680i and the RD600. Intel isn't even on the list of top 3 chipset maybe the badaxe 2 is good but intel isn't great at everything much less graphics wise (Its laughable). Whats great is C2D however.

[SkunK]

Badaxe is a MOTHERBOARD, not a chipset.

The 965 and 975 chipsets are VERY good. the 680i is a JOKE. It generates an insane amount of heat. And the RD600, while being good, is VERY hard to find.

It's funny though, that as marvellous as those two chipsets are, they always end up being hosted on ty motherboards with flaky BIOS:

PLEASE! HELP ME! I have to set FSB to 397.2753367 to boot in Windows and use a hacked program in Windows itself to raise FSB!!!

PLEASE! HELP ME! I have to clear CMOS to cold boot!!!

PLEASE! HELP ME! I've tried all 36 BIOS and none of them can make my board go above 348 FSB!!!

PLEASE! HELP ME! I need a new strap!

You need a solder iron to make your board boot, noob.

You need liquid nitrogen to cool your chipset, noob.

My ES motherboard owns all of your retail crap, noobs.

My vdroop is 0.2V better than yours, noob.

It makes me wonder if every single board maker sucks or if the problem could be elsewhere.

[SEA]

if almost everything around your processor sucks (MBs, memory PSU and so on) - it's a good time to look at the part in the middle. Yes, the one that is hardly compatible to world around

[nn_step]

I'm voting for "every single board maker sucks".

My DFI-Ultra D was an absolute B!T#H. Hated my Seasonic PSU. Flakiest board I EVER owned.

ASUS A8N-SLI also was a bear. SO PICKY with memory.

I'm of the FIRM belief that board makers only HALF test their products and we get stuck being beta testers for them.

Buy a Via Board, it doesn't matter which one.
It isn't an uberclocker or picky about anything.
But it is extremely stable, totally compatible, and a rock stable base.
you don't get the right to about Mobos since you are going for the points not the stable and reliable board. But then again this is Xtreme Systems and most of the people here aren't even stable anyways so what would be the point

[Serge84]

Badaxe is a MOTHERBOARD, not a chipset.

The 965 and 975 chipsets are VERY good. the 680i is a JOKE. It generates an insane amount of heat. And the RD600, while being good, is VERY hard to find.

The X975 chipset and everything on it is codenamed badaxe for its desine. :l It IS the mobo The chipset makes the mobo, with out it it would be a useless piece of circuits but the circuits are part of badaxe. Every peice of crap that is part of the mobo is part of the badaxe desine it IS the MFing entire board. Chipset and all doesn't matter, every caps placement, chip placement, chipset desine is badaxe.

[Epsilon84]

then again this is Xtreme Systems and most of the people here aren't even stable anyways so what would be the point

I agree, many people here (*ahem* fanboyz *ahem*) are unstable and need to be put into mental institutions to learn to stop fantasizing over chunks of silicon.

[biohead]

The X975 chipset and everything on it is codenamed badaxe for its desine. :l It IS the mobo The chipset makes the mobo, with out it it would be a useless piece of circuits but the circuits are part of badaxe. Every peice of crap that is part of the mobo is part of the badaxe desine it IS the MFing entire board. Chipset and all doesn't matter, every caps placement, chip placement, chipset desine is badaxe.

spelling police!

design

[SkunK]

My DFI-Ultra D was an absolute B!T#H. Hated my Seasonic PSU. Flakiest board I EVER owned.

Mine has been stable for close to a year now and with a Seasonic psu and a Maxtor (more of a nforce4 thing) hd which both allegedly had trouble with the Ultra-D.

How much did I pay for the Ultra-D? 140$ CDN (about 120$ US). Can you find either a 965 or a 975 board around that price that will hold a stable (and significative) overclock day in day out for a year?

[Shintai]

Mine has been stable for close to a year now and with a Seasonic psu and a Maxtor (more of a nforce4 thing) hd which both allegedly had trouble with the Ultra-D.

How much did I pay for the Ultra-D? 140$ CDN (about 120$ US). Can you find either a 965 or a 975 board around that price that will hold a stable (and significative) overclock day in day out for a year?

Any P965 board can basicly do that, from 100$ and up
Even 50$ Via board with DDR1 can do around 20-25% OC. But that pales when you can get over 100% OC on P965.

[SkunK]

Any P965 board can basicly do that, from 100$ and up
Even 50$ Via board with DDR1 can do around 20-25% OC. But that pales when you can get over 100% OC on P965.

I can't of course ask you to show me just one retail 965 board that's been running constantly at any overclock (I'm not talking about 270MHZ FSB here) let alone 100% for one year because those boards have not been out to retail for that long. Show me just ONE that's been doing that for six months.

As for your 50$ Via board. Which one are you talking about? The P5P800 AKA the board that has its NIC broken within two days?

And finally, 20% 25% overclock is reserved to the Asus crowd. As far as I'm concerned, 40% on single core and 30% on dual core without a mV above stock voltage is what I'm talking about.

[SkunK]

Gigabyte DS3.
http://www.newegg.com/Product/Produc...82E16813128012

50-100% OCs are NOT uncommon for this board and it is ROCK solid (even has solid metal caps).

1. That link shows a board on sale at today's price. You wanna compare that to the price (70$, local store) of Ultra-D as of today?
2. Even on sale and today, it's still above 120$ (old prime Ultra-D price)
3. Gigabyte has a reputation up here to blow caps faster than the Moon's orbit around the earth. Many local stores stopped carying Gigabyte motherboards because of that. Personally, I've seen two Gigabyte boards and both of them had blown caps (curiously enough Intel based).
4. 50% overclock not uncommon? Show me just five instances of those with retail (and modified) parts and average joes (like myself) running as long as the board has been out.