• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

4x4 Next Month

That may end up being true. :D

The thing is though, 4x4, and dual Woodcrests, are aimed at very different markets. One for the enthusiast community, and the other for Servers and Workstations. It has no place in any kind of comparison. The only thing I can think a useful comparison can be made is if you were treating it theoretically.

Besides, I'm still not convinced that Core2 platform will scale all that well. We will have to wait and see. FSB versus HT is what it comes down to, and I dont think the GTL+ FSB will have what it takes to scale. It may still take the lead, but I dont think it will be by a significant margin if at all. And that is entirely thanks to an ungodly amount of cache. And maybe even quad FSB's :eek: How much do you suppose that MB would cost? I mean WoW!

But this is speculation on my part. Maybe even a bit of hype. We'll have to wait and see how the ball rolls down the hill. ;)
 
duby229 said:
And maybe even quad FSB's
Click to expand...

How does the whole Quad FSB apply to this discssion?

Tigerton is going to use it, but not Woodcrest/Cloverton or Conroe/Kentsfield, so I don't see the relation

Thanks
 
duby229 said:
Laid the smackdown on what? 4x4 hasnt been released yet, and as far as I know nobody has seen benches yet. So what did the Woodcrest lay the smackdown on, and how does this have anything to do with 4x4? Who has the advantage in video subsystems?

I'm only asking becouse your not being quite clear. Please help me understand. ;)
Click to expand...

Two Socket Dual Core Opteron vs. Two Socket Dual Core Woody have been done already. Do you think AMD made improvements architectural wise to FX-7x? The DP and Two FSBs do work well together at 1333MHz.

AMD has always sold their smaller cache Processors for less, why change now? So we're talking about one Kentsfield or two Woodys with 8MB of L2 vs. Two FX-7x with 2MB of L2 each. That's why the price came up.

Link

GamePC said:
If you made it through the last several pages of benchmarks, you can clearly see that Intel’s new Xeon 5150 and 5160 processors are absolutely top of the line in terms of workstation application performance. Both of these models were able to best AMD’s top of the line Opteron 285 processors in nearly every benchmark, which previous generation Xeon processors simply were not capable of doing. Server performance is still a close call, but Intel’s new Xeon processors again perform fantastically well here. While topping the performance charts is a fantastic achievement for Intel, making things even sweeter here is that Intel is delivering better performance at lower prices. AMD’s Opteron 285 processor still sells for $1100-$1200, whereas the new Xeon 5160 sells for $900-$1000, whereas the Xeon 5150 (which still bests the Opteron 285 in most tests) sells for $700-$800. Thus, Intel is winning not only in raw performance, but also price/performance as well.
Click to expand...

2CPU, Tech Report, Anand and others have written the same or similar.

GamePC, like other boutique shops have higher prices but they are higher for Intel and AMD, since they sell both. Price fell since that review. That Box/Kit with two FX-74's will sell for $1500 because some sites have 3GHz Woodys for almost that price. Remains to be seen how much VARs will mark up FX-7x or their 4 X 4 platforms. I estimate at least $400 for the 4 X 4 motherboard.

Page Computers

eWiz

Many folks' favorite
Newegg

Went looking for a motherboard and found this.
Tyan S5380G2NR

If you like to see how DDR2 667 FBDIMMs vs Regular DDR2 read this review?

Now, I ask you, is that clear enough for you?
 
duby229 said:
That may end up being true. :D

The thing is though, 4x4, and dual Woodcrests, are aimed at very different markets. One for the enthusiast community, and the other for Servers and Workstations. It has no place in any kind of comparison. The only thing I can think a useful comparison can be made is if you were treating it theoretically.

Besides, I'm still not convinced that Core2 platform will scale all that well. We will have to wait and see. FSB versus HT is what it comes down to, and I dont think the GTL+ FSB will have what it takes to scale. It may still take the lead, but I dont think it will be by a significant margin if at all. And that is entirely thanks to an ungodly amount of cache. And maybe even quad FSB's :eek: How much do you suppose that MB would cost? I mean WoW!

But this is speculation on my part. Maybe even a bit of hype. We'll have to wait and see how the ball rolls down the hill. ;)
Click to expand...

They're aimed at any market the user wants.
4 X 4 is a converted server platform.
All the new processors are converted Opterons.
It's a quick fix to catch back up for some kind of misplaced E-Penos* contest. Might be meant more for F@n$ than anything else LOL!
It's AMD using two processors to battle one.

After that's said though, I hope it works and AMD makes a wad of Cash. No Sarcasm meant. I'm dead serious about that. Just as I said last year, neither company needs to wipe the floor with the other for too long, that hurts ALL of our pockets.
 
i think Core architecture will scale beautifully...

by the time we get past quad core, CSI will be there to carry the arch and future Intel ones into the future :)

so yea, Core architecture will scale excellently
 
brucedeluxe169 said:
i think Core architecture will scale beautifully...
Click to expand...
then how come a dual DC opteron system at 2.6ghz is isn't too far behind a dual woodcrest system at 3ghz? ;)
http://techreport.com/etc/2006q2/woodcrest/index.x?pg=1


there's also the anandtech test, but they did some really weird stuff.
http://www.anandtech.com/IT/showdoc.aspx?i=2772&p=1

check out page 8.. note how for the opteron platform, the dual core is faster than the 2x dual core :confused:
discrephancies like this arise everywhere throughout the article, so i have generally disregarded it as trash
 
(cf)Eclipse said:
then how come a dual DC opteron system at 2.6ghz is isn't too far behind a dual woodcrest system at 3ghz? ;)
Click to expand...

As far as I understand it FB-DIMM.
 
well, in the techreport article, the woodcrest system is faster by alot depending on the app (well, most of the apps they tested). Sure there were some instances where the opteron was close, but it didn't win anything... so I fail to see how this is a valid counterpoint.

Besides, even the Prescotts were able to get close to and beat the K8s in a few apps... but obviously, largely lost in most
 
enumae said:
As far as I understand it FB-DIMM.
Click to expand...
fb-dimms aren't THAT slow. even if they were, memory subsystem has a pretty small effect on overall system performance. look at the latency test, it's about where you would expect it to be, with opterons having an IMC and the woodcrest not. ;)
 
brucedeluxe169 said:
well, in the techreport article, the woodcrest system is faster by alot depending on the app (well, most of the apps they tested). Sure there were some instances where the opteron was close, but it didn't win anything... so I fail to see how this is a valid counterpoint.

Besides, even the Prescotts were able to get close to and beat the K8s in a few apps... but obviously, largely lost in most
Click to expand...
true, i won't deny that, but it appears that the programs that the opteron sysem was closer in were the ones that really very parallel in nature, like pov-ray and the audio conversion. with the same basic subsystem as the netburst line, there is no reason why core will scale a lot better. once intel finishes up CSI, things will definitely perk up a lot when it comes to massive multithreading :D
 
(cf)Eclipse said:
fb-dimms aren't THAT slow. even if they were, memory subsystem has a pretty small effect on overall system performance. look at the latency test, it's about where you would expect it to be, with opterons having an IMC and the woodcrest not. ;)
Click to expand...

Actually , from what I understand the way in which memory test programs work ( Sandra ) will never show the full potential of FBDIMMs.They were designed from the start for server like tasks and thus are optimized for such ( huge number of concurrent acceses ).Desktop benchmarking programs aren't able to put a heavy load on FBDIMMs and incorrect conclusions will be drawn like "FBDIMMs suck".
 
Ok, here is why I thought it was the FB-DIMM's...

This is a quote from CPU magazine and was written by Anand (founder of Anandtech).

FBD uses normal DDR2 (and eventually DDR3) devices, so costs won’t be that much higher than those of normal DDR2 modules. What separates FBD from regular DDR2 is a chip on each module called the AMB (Advanced Memory Buffer), which buffers all data/address/command requests to the DRAM devices—hence the name “Fully Buffered DIMM.” Buffering them lets you have up to eight FB-DIMMs per channel.

The AMB also acts as a translator; one side communicates with parallel DDR2 devices, while the other has a high-speed serial interface that communicates between AMBs and the memory controller. This aspect of the AMB’s role allows Intel to have its serial interface while still interfacing to commodity DDR2 DRAM, effectively giving it the best of both worlds.

Unfortunately, the AMB is a drain on performance. As you can guess, high-speed translation always comes with a penalty, and thus latency is higher with FBD than conventional DDR2 memory. The latency penalty will keep FBD from becoming a desktop standard, and it may end up killing it in the enterprise market down the road. The best overall solution is still going to be a native serial memory architecture, as the performance loss at the AMB is significant.
Click to expand...

Hope this explains why I believe or understand that FB-DIMM is the reason for Opteron to be close to Woodcrest.
 
enumae said:
Ok, here is why I thought it was the FB-DIMM's...

This is a quote from CPU magazine and was written by Anand (founder of Anandtech).

Hope this explains why I believe or understand that FB-DIMM is the reason for Opteron to be close to Woodcrest.
Click to expand...

That's why I linked to the GamePC review of the X6800 2.93 vs. 3GHz Woody. You see which apps takes a hit and ones that don't. There will be apps that will perform faster on One Dual Core Opteron than on two FX-7x at the same speed.

Software can be optimized for Intel and or AMD processors. Still wondering why AMD hired one of the writers of ScienceMark? That's why even Netburst can squeese out a win every once-in-a-while. There might also be apps that are bottlenecked by Memory, I/O, platform latency and etc.. just as Games are bottlenecked by GPUs.
 
Arvidas said:
LOL

All you say is so reliable like this? :eek: :p
Click to expand...

Each core has 128K L1 and a 1MB L2. 4 256K L2s Equal 1MB of L2 total. Just like a lot of stuff posted I'll believe it when I see it.

This link says otherwise.

http://www.xbitlabs.com/news/cpu/display/20061017130006.html

The new AMD Athlon 64 FX-70 (2.60GHz, 2MB L2 cache [1MB per core]), FX-72 (2.80GHz, 2MB L2 cache [1MB per core]) and FX-74 (3.00GHz, 2MB L2 cache [1MB per core]) microprocessors designed to work in pairs will be compatible with socket F infrastructure, a 1207-pin land grid array (LGA) form-factor. It is unclear whether the new AMD Athlon 64 FX processors need registered DDR2 memory modules, which cost more than typical DDR2 components and which are somewhat slower, but it is highly-likely that this will be a requirement.

As a result of transition to workstation-class technologies, users of next-generation Athlon 64 FX processors will be able to install two of such chips. However, besides of the fact that they will need to acquire two microprocessors instead of one, they will need new – specially designed – cooling systems, mainboards, memory modules and, perhaps, power supply units, as new FX-series models will consume up to 125W of power.
Click to expand...

The 256K L2 figure came from Quad Core K8L shipping with that amount. 4MB is still half as much as 8MB. I honestly thought it would be 512K for each core.

So here's one of the places I saw the temps.

http://www.gamers-underground.com/forum/showthread.php?t=23437
 
Quad K8L has 512kb L2 per core and 2mb L3 shared. Dual K8L, 512kb L2 per core, and 2mb (perhaps 1mb) L3 shared.

Where have you read FX-7x 256kb L2?
And Quad K8L 256kb L2?
 
Donnie27 said:
Each core has 128K L1 and a 1MB L2. 4 256K L2s Equal 1MB of L2 total. Just like a lot of stuff posted I'll believe it when I see it.
Click to expand...
it says explicitly.. 1MB per core. i don't understand how you can take it any differently. there's two cores on each chip, this 2MB per chip.. as stated. you're just posting this to piss people off, as always:


Donnie27 said:
I must admit, I posted crap wrong for a reaction.
Click to expand...
:rolleyes:
 
Donnie27 said:
Each core has 128K L1 and a 1MB L2. 4 256K L2s Equal 1MB of L2 total. Just like a lot of stuff posted I'll believe it when I see it.
Click to expand...

thats just illogical, does that mean theres 32k of L1 cache per core too?? your illogical.

you said there is 1mb of L2 cache for EACH core, that means 4 CORES in a 4x4 setup, meaning 4mb of L2 cache. you make minimal sense. and its getting on my nerves.
 
4x4 gets its name from 4 cores and 4 nvidia gpus in Sli right??? If this is the case then I am glad AMD and Nvidia are getting along even though AMD bought ATI. phew......and really this thread seems to be getting ridiculously out of hand.
 
As I Lay Dying said:
4x4 gets its name from 4 cores and 4 nvidia gpus in Sli right??? If this is the case then I am glad AMD and Nvidia are getting along even though AMD bought ATI. phew......and really this thread seems to be getting ridiculously out of hand.
Click to expand...
it does indeed! nvidia basically threw two 590 sli's together to make "680" :D
 
any new features in this 680 chipset over the current chipsets besides quad sli?

Also (cf)Eclipse would you know when the torrenza boards is gonna be available?? Also I know IBM said the roadrunner super computer is gonna be Opteron and Cell cpus, any chance IBM is gonna make cell co-processors available to the general market. I know the linux kernel already has cell support, if this pans out I dunno how intel will counter an AMD K8L cpu with an IBM cell co-processor, of course speculation.
 
(cf)Eclipse said:
quad ethernet, like.. 12 sata ports, etc. let me find the "specs" again
http://www.theinquirer.net/default.aspx?article=35334

and i know NOTHING about torrenza :(
Click to expand...

hahahaha holy crap maybe 4x4 doesn't look that bad, if you get 2 quad core Barcelona's to my knowledge are the first K8L's (right?), and probably one High end Nvidia GPU, (no I am not an Nvidia Fan boi, I just get nvidia video cards because of the better linux driver support). Hell I guess you don't need one, but for a killer desktop server combo.
 
first quad core chips will be K8L, yes ;)

conbine two of those with two dual G80's... heh :D
 
hahahaha, yeah I could go for 2 K8L's no problems, but, just one G80 I don't game enough to justify two G80's. Also are Barcelona's gonna be HT3?
 
(cf)Eclipse said:
first quad core chips will be K8L, yes ;)

conbine two of those with two dual G80's... heh :D
Click to expand...


Exactly what I plan on doing! Zomg! Two 8950gx2's (if they come out) and 2 quad cores and I will have a godly system!


-Brandon
 
As I Lay Dying said:
hahahaha, yeah I could go for 2 K8L's no problems, but, just one G80 I don't game enough to justify two G80's. Also are Barcelona's gonna be HT3?
Click to expand...
yes.. which will probably help multithreaded scaling a bit for 4x4 and in the multi-socket opteron rigs. i don't see immediate benefits for ht3 outside of that area though
 
(cf)Eclipse said:
yes.. which will probably help multithreaded scaling a bit for 4x4 and in the multi-socket opteron rigs. i don't see immediate benefits for ht3 outside of that area though
Click to expand...

thanks for the info.
 
sorry my understanding is that the 7950 GX2 recommends a 400W PSU, I assuming the 8x50GX2 needs more, 4 GX2 would kill your 1.5KW PSU.
 
As I Lay Dying said:
any new features in this 680 chipset over the current chipsets besides quad sli?

Also (cf)Eclipse would you know when the torrenza boards is gonna be available?? Also I know IBM said the roadrunner super computer is gonna be Opteron and Cell cpus, any chance IBM is gonna make cell co-processors available to the general market. I know the linux kernel already has cell support, if this pans out I dunno how intel will counter an AMD K8L cpu with an IBM cell co-processor, of course speculation.
Click to expand...

As far as Cell co-processor is concerned, I dont see it hitting the desktop market anytime soon. It may never. Yould have to write software to support it. In an X86 world that is likely to never happen.

However, I DO believe that GP-GPU's will be the next big thing. With Fusion integrating the GPU on die, and AMD developing a new ISA extension...... Well let's just say that at this point it is going to happen. :cool:
 
An 8800GTX is requiring 600 if I remember correctly. Maybe its 800.. so 1.5kw's would be too little no matter what.. You guys should be discussing this in my new thread, though, not clogging up this one.


-Brandon
 
(cf)Eclipse said:
it says explicitly.. 1MB per core. i don't understand how you can take it any differently. there's two cores on each chip, this 2MB per chip.. as stated. you're just posting this to piss people off, as always:

:rolleyes:
Click to expand...

You got it!
 
duby229 said:
As far as Cell co-processor is concerned, I dont see it hitting the desktop market anytime soon. It may never. Yould have to write software to support it. In an X86 world that is likely to never happen.

However, I DO believe that GP-GPU's will be the next big thing. With Fusion integrating the GPU on die, and AMD developing a new ISA extension...... Well let's just say that at this point it is going to happen. :cool:
Click to expand...

well ummm if IBM and others are really backing torrenza, then I wouldn't be surprised if they add support to the linux kernel for co-processing. I mean they were the ones who added the cell processor support to the linux kernel. But I do agree with you this could never ever happen.
 
CCUABIDExORxDIE said:
thats just illogical, does that mean theres 32k of L1 cache per core too?? your illogical.

you said there is 1mb of L2 cache for EACH core, that means 4 CORES in a 4x4 setup, meaning 4mb of L2 cache. you make minimal sense. and its getting on my nerves.
Click to expand...

Look, I know you can't read the whole thread but Eclipse has.

http://www.theinquirer.net/default.aspx?article=34918

AMD QuadFather matches Kentsfield

Two FX-70 CPU pack matches pricing, clocks of Core 2 QC6700

Inq said:
All of the CPUs will feature 256KB of L1 and 2MB of L2 cache memory (each core has 64+64KB L1 and 1MB L2 cache), and be placed in three speed grades.
Click to expand...

See the misprint? Both instead of All. I've seen a lot of these! I saw folks saying 256K, 512K and yes 1MB L2's per. Maybe cheaper models with less cache will follow, hell I don't know. I do know;
They re 1MB per core versions.
Socket F or 1207, not AM2 or 939.
Price might range from 1000 to 1500.
They're 90nm 2xx Opterons on a new package.
125W TDP PER PROCESSOR!!
Lanches in less than 3 weeks.
Now way in hell is a 500W PSU going to power the monster.

Yes, I posted some wild stuff and even admitted to doing so. Stay on your toes? I've posted a lot of Damned good links as well to prove I was BSing and to make my points!
 
Donnie27 said:
See the misprint? Both instead of All. I've seen a lot of these! I saw folks saying 256K, 512K and yes 1MB L2's per. Maybe cheaper models with less cache will follow, hell I don't know. I do know;
They re 1MB per core versions.
Socket F or 1207, not AM2 or 939.
Price might range from 1000 to 1500.
Click to expand...
they mean all as in the fx-70, 72 and 74.

so let's quote it and break it down for you

All of the CPUs will feature 256KB of L1 and 2MB of L2 cache memory (each core has 64+64KB L1 and 1MB L2 cache), and be placed in three speed grades.
Click to expand...
so, obviously, as you know, it says 64+64KB L1 and 1MB L2 per core. thus, each cpu will have 256KB L1 and 2MB L2 total. NOT 256KB L2 as you were saying before

i see that price comment again. i'm not gonna correct you as there are no definite prices, so it would be hypocritical of me to say "it'll be this", but too bad. the fx-74's will match whatever the top end kentsfield is. by saying that the high end is gonna be 1500, you're implying that the fastest kentsfield is gonna be 1500 as well ;)
 
You must log in or register to reply here.
Back
Top