Ratdude's main rig V3, maybe

[RJARRRPCGP]

Re: Ratdude's main rig V3, maybe

What the f*$k so you think 9600gt was when it came out? Do you think it was top of the line of the video cards? G94 chip was a castrated G92, with half the amount of processing units, just like radeon 6670 is the same chip used in 6850 or more powerful cards but with processing units disabled.

Sounds just like the Radeon 9000 Pro back before 2004, which was pretty much a crippled 8500!

I was wondering if these type of tactics were why ATI got bought out?

[RJARRRPCGP]

Re: Ratdude's main rig V3, maybe

I just bought a new 1GB Radeon HD 6570 for my HTPC. Really impressed. Will it SLI nope. But it's better to have a good single card rather than two crsppy SLI cards...

"crispy" SLI cards? LOL.

[ratdude747]

Re: Ratdude's main rig V3, maybe

Where did I say you should put an nVidia AND an AMD/ATI video card?

Is 24A on the 3.3V rail enough to power all of those high voltage RAM sticks?

I believe ati cards are out of the question for his rig, something about linux

in that case, he should consider a low GT 600 card. Like a GT 640 or so

A few quick points:

-SN, you never suggested that. Somebody else did.
-It's not really a linux issue (well, it could be), it's rather a preference... To me, I'd rather run nvidia cards on nvidia chipsets (intel and old amd I can run either and still feel ok).

-I don't know about the 3.3V, I have no idea

Now this post may take some analysis:

It's really frustrating to see what goes in your head... it's so full of some misconceptions or assumptions...

No, I have different goals and preferences.

On one hand, you say that since you have two pcie x16 you want to use both to maximize stuff, but when someone tells you that they're actually like x8 slots due to low bandwidth, it's suddenly no longer an issue.
Then you go on about agp and not using all bandwidth and make it seem like bandwidth is not important, but then it's suddenly very important to use both pcie x16 again, for bandwidth and other reasons. It's ridiculous. Make up your mind.

You misunderstood me.

1. PCIe 16x 1.1 is PCIe 16x 1.1 pretty much. It's not 8x (see below). In fact it's in the chip specs (see the article i quoted a few posts ago)

2. Regarding the "make up my mind". You're comparing apples to pairs of apples. Let me explain:

"I'm not going to use up the 16x"- that's for a single card.

"I have double the bandwidth with two cards": that's for two cards.

The general idea is that by running two cards via SLI I'm best using my available bandwidth.

As for the hypertransport link speeds (Somebody will bring it up, I know):

(from tyan's spec page on the K8WE)

"Three HyperTransport™ links support up to 8.0GB/s data transfer rate each link"

That's the speed that each link connects to the CPU with. Yeah, the RAM speed is slower (direct to CPU via internal controller, not hypertransport at all), but that's why the GPUs have their own RAM in the first place.

Also, each PCIe 16x 1.1 slot can do 4.0GB/s. (250 MB/s x16).

That link is shared by some other stuff:



(also note that there appears to be 3 links per CPU (one unused), for a total of 5 links).

First, each slot is controlled by a separate link and chip... One is on the main 2200 chip, the second is on an auxillary 2050 chip (2200 with redundant connections removed, otherwise the same chip)

Second, the links may be shared y other devices on the chips, but it's mostly low load devices, the fastest being the SATA ports. However, the high load SATA drives are on a RAID card via the PCI-X controller, which on a separate link/chip (AMD 5131, never had an issue with that one since I don't hot swap)... So there you go, full PCIe speed to the cpu's (minus any converter chip losses).

And by the way, weren't you the guy who told me not to use a newer card? Let me see:

The issue is how these pci express ports are implemented.
On consumer hardware, the lanes to the video card pci express come from the cpu and the less important lanes are from the chipset...
on these workstation boards each pci express goes to a chip which then talks to the cpu: https://cdn.badcaps-static.com/pdfs/...2b07aca8a6.pdf

Modern video card drivers don't take this in consideration and they're "tuned" to assume direct connection to processors, so you can get into instability issues.

If you say it's gonna cause problems, then don't suggest it!

I can only determine that you don't really care about "optimum" usage of hardware, you just don't want to see slots unused, which is just dumb. If you had 4 pci express slots, you would probably have asked how to make quad sli with some similar boards.

If I had a Quad SLI board, I'd still double it (or look to down the road), not quad it. Why? PSU draw. Two is OK. 3/4, a little too much. Maybe double it eventually If I got a deal on two more cards.

Then you say

"I'll say it, it just doesn't feel right putting crap level new stuff on such a well built high end (in it's day) computer. Like having a restored muscle car and putting cheap tires on it"

What the f*$k so you think 9600gt was when it came out? Do you think it was top of the line of the video cards? G94 chip was a castrated G92, with half the amount of processing units, just like radeon 6670 is the same chip used in 6850 or more powerful cards but with processing units disabled.
Seems like everything that's cheap these days is crap, everything that was expensive back then or is expensive now is premium to you... that's a ridiculous assumption. nVidia cards were always more expensive, always more lousy, with more problems everywhere except drivers (in linux)... doesn't make them better or more high end, it's still crap.

9600gt was never racing tires. To keep up with your analogy, it's like you buying rethreaded high end racing tires on your muscle car instead of buying new middle of the road racing tires, just because you want to see the brand name on the tires.
It doesn't change the fact that the rethreaded tires have less performance than the new, more modern, better designed middle budget racing tires.
If you don't have the money for high end racing tires, you shouldn't buy obsoleted rethreaded, unsafe tires just because of a logo on them or because they were good years ago.

You've got it bass ackwards.

9600 was the low end of the high segment. The highest of the "single slot" cards. They ran cooler and as a result draw less power and last longer (less BGA stress) than the 9800 variants. They also cool better since there is an open slot between them (I wouldn't dare fill that unless I had to, and it would be a small card). You want cheap, that's the 9200 and lower 9400's... higher 9400's and 9500's are the mid grades, with 9600 being the transition to the high cards.


it's also a cut-off point, the next card up in that "segment" is the 260, which is a hot high-draw 2 slot card. The card below it, is a cheap looking 240, which has no SLI capability. 9600 is the best SLI capable single slot lower draw card.


And the tire analogy doesn't fit as well... not retreads, those would be reflow cards (like my dead 8800GTS, no thanks). Rather new quality tires that are on the street safe edge of racy that are of a vintage design... it's what the job calls for. After all, if you're running a restored car (or computer), you are supposed to use period (or near period) looking parts. It would be like putting a fart can on a corvette. It's just not right.

[ratdude747]

Re: Ratdude's main rig V3, maybe

Where did I say you should put an nVidia AND an AMD/ATI video card?

Is 24A on the 3.3V rail enough to power all of those high voltage RAM sticks?

I believe ati cards are out of the question for his rig, something about linux

in that case, he should consider a low GT 600 card. Like a GT 640 or so

A few quick points:

-SN, you never suggested that. I meants using an AMD/ATI card on an Nvidia chipset.
-It's not really a linux issue (well, it could be), it's rather a preference... To me, I'd rather run nvidia cards on nvidia chipsets (intel and old amd I can run either and still feel ok). IMHO it's less likely to have quirks that way (since Nvidia probably tweaked the chip design primarily using Nvidia cards)

-I don't know about the 3.3V, I have no idea

Now this post may take some analysis:

It's really frustrating to see what goes in your head... it's so full of some misconceptions or assumptions...

No, I have different goals and preferences.

On one hand, you say that since you have two pcie x16 you want to use both to maximize stuff, but when someone tells you that they're actually like x8 slots due to low bandwidth, it's suddenly no longer an issue.
Then you go on about agp and not using all bandwidth and make it seem like bandwidth is not important, but then it's suddenly very important to use both pcie x16 again, for bandwidth and other reasons. It's ridiculous. Make up your mind.

You misunderstood me.

1. PCIe 16x 1.1 is PCIe 16x 1.1 pretty much. It's not 8x (see below). In fact it's in the chip specs (see the article i quoted a few posts ago)

2. Regarding the "make up my mind". You're comparing apples to pairs of apples. Let me explain:

"I'm not going to use up the 16x"- that's for a single card.

"I have double the bandwidth with two cards": that's for two cards.

The general idea is that by running two cards via SLI I'm best using my available bandwidth.

As for the hypertransport link speeds (Somebody will bring it up, I know):

(from tyan's spec page on the K8WE)

"Three HyperTransport™ links support up to 8.0GB/s data transfer rate each link"

That's the speed that each link connects to the CPU with. Yeah, the RAM speed is slower, but that's why the GPUs have their own RAM in the first place.

Also, each PCIe 16x 1.1 slot can do 4.0GB/s. (250 MB/s x16).

That link is shared by some other stuff:



(also note that there appears to be 3 links per CPU, for a total of 6 links)

First, each slot is controlled by a separate link...

Second, the links may be shared, but it's mostly low load devices, the fastest being the SATA ports. However, the high load SATA drives are on a RAID card via the PCI-X controller, which on a separate link... So there you go, full PCIe speed to the cpu (minus any converter chip losses).

And by the way, weren't you the guy who told me not to use a newer card? Let me see:

The issue is how these pci express ports are implemented.
On consumer hardware, the lanes to the video card pci express come from the cpu and the less important lanes are from the chipset...
on these workstation boards each pci express goes to a chip which then talks to the cpu: https://cdn.badcaps-static.com/pdfs/...2b07aca8a6.pdf

Modern video card drivers don't take this in consideration and they're "tuned" to assume direct connection to processors, so you can get into instability issues.

If you say it's gonna cause problems, then don't suggest it!

I can only determine that you don't really care about "optimum" usage of hardware, you just don't want to see slots unused, which is just dumb. If you had 4 pci express slots, you would probably have asked how to make quad sli with some similar boards.

If I had a Quad SLI board, I'd still double it (or look to down the road), not quad it. Why? PSU draw. Two is OK. 3/4, a little too much. Maybe double it eventually If I got a deal on two more cards.

Then you say

"I'll say it, it just doesn't feel right putting crap level new stuff on such a well built high end (in it's day) computer. Like having a restored muscle car and putting cheap tires on it"

What the f*$k so you think 9600gt was when it came out? Do you think it was top of the line of the video cards? G94 chip was a castrated G92, with half the amount of processing units, just like radeon 6670 is the same chip used in 6850 or more powerful cards but with processing units disabled.
Seems like everything that's cheap these days is crap, everything that was expensive back then or is expensive now is premium to you... that's a ridiculous assumption. nVidia cards were always more expensive, always more lousy, with more problems everywhere except drivers (in linux)... doesn't make them better or more high end, it's still crap.

9600gt was never racing tires. To keep up with your analogy, it's like you buying rethreaded high end racing tires on your muscle car instead of buying new middle of the road racing tires, just because you want to see the brand name on the tires.
It doesn't change the fact that the rethreaded tires have less performance than the new, more modern, better designed middle budget racing tires.
If you don't have the money for high end racing tires, you shouldn't buy obsoleted rethreaded, unsafe tires just because of a logo on them or because they were good years ago.

You've got it bass ackwards.

9600 was the low end of the high segment. The highest of the "single slot" cards. They ran cooler and as a result draw less power and last longer (less BGA stress) than the 9800 variants. They also cool better since there is an open slot between them (I wouldn't dare fill that unless I had to, and it would be a small card). You want cheap, that's the 9200 and lower 9400's... higher 9400's and 9500's are the mid grades, with 9600 being the transition to the high cards.


it's also a cut-off point, the next card up in that "segment" is the 260, which is a hot high-draw 2 slot card. The card below it, is a cheap looking 240, which has no SLI capability. 9600 is the best SLI capable single slot lower draw card.


And the tire analogy doesn't fit as well... not retreads, those would be reflow cards (like my dead 8800GTS, no thanks). Rather new quality tires that are on the street safe edge of racy that are of a vintage design... it's what the job calls for. After all, if you're running a restored car (or computer), you are supposed to use period (or near period) looking parts. It would be like putting a fart can on a corvette. It's just not right.

[Uranium-235]

Re: Ratdude's main rig V3, maybe

I believe ati cards are out of the question for his rig, something about linux

in that case, he should consider a low GT 600 card. Like a GT 640 or so

[Pentium4]

Re: Ratdude's main rig V3, maybe

Is 24A on the 3.3V rail enough to power all of those high voltage RAM sticks?

[shovenose]

Re: Ratdude's main rig V3, maybe

I don't like mix matching nvidia and AMD/ATI..

Where did I say you should put an nVidia AND an AMD/ATI video card?

[mariushm]

Re: Ratdude's main rig V3, maybe

It's really frustrating to see what goes in your head... it's so full of some misconceptions or assumptions...

On one hand, you say that since you have two pcie x16 you want to use both to maximize stuff, but when someone tells you that they're actually like x8 slots due to low bandwidth, it's suddenly no longer an issue.
Then you go on about agp and not using all bandwidth and make it seem like bandwidth is not important, but then it's suddenly very important to use both pcie x16 again, for bandwidth and other reasons. It's ridiculous. Make up your mind.
I can only determine that you don't really care about "optimum" usage of hardware, you just don't want to see slots unused, which is just dumb. If you had 4 pci express slots, you would probably have asked how to make quad sli with some similar boards.


Then you say

"I'll say it, it just doesn't feel right putting crap level new stuff on such a well built high end (in it's day) computer. Like having a restored muscle car and putting cheap tires on it"

What the f*$k so you think 9600gt was when it came out? Do you think it was top of the line of the video cards? G94 chip was a castrated G92, with half the amount of processing units, just like radeon 6670 is the same chip used in 6850 or more powerful cards but with processing units disabled.
Seems like everything that's cheap these days is crap, everything that was expensive back then or is expensive now is premium to you... that's a ridiculous assumption. nVidia cards were always more expensive, always more lousy, with more problems everywhere except drivers (in linux)... doesn't make them better or more high end, it's still crap.

9600gt was never racing tires. To keep up with your analogy, it's like you buying rethreaded high end racing tires on your muscle car instead of buying new middle of the road racing tires, just because you want to see the brand name on the tires.
It doesn't change the fact that the rethreaded tires have less performance than the new, more modern, better designed middle budget racing tires.
If you don't have the money for high end racing tires, you shouldn't buy obsoleted rethreaded, unsafe tires just because of a logo on them or because they were good years ago.

[ratdude747]

Re: Ratdude's main rig V3, maybe

No on the newer card... Maybe it's a peeve, but I don't like mix matching nvidia and AMD/ATI. As for thr 3 monitors, you could say the same about the HD4670 AGP (Same layout), it's Bullshit. To me, multiple monitors = mutiple idenical ports. In that case you either get 2 VGA (via DVI adapter) or two DVI (via HDMI adapter)...

Also No on an 8800GT (I know it was a reference GPU) ... I have an 8800GTS in the pile, it spent more time on the reflow tray than on the motherboard. Junk.

Reason I said the 9600 was it had all the 3d codec support i needed (I don't need DX11 for any games I play or the like) but runs cool enough to not need reflowed every 3 months. Since I have two PCIe x16 channels, I might as well maximise the bandwidth and use both.

Plus, if it's like AGP, just because you are using say PCIe 16x, doesn't mean that all of that speed will be used... so for an early PCIe 2.0 card like the 9600, I don't think the fact I'm (AFAIK) on PCIe 1.0 is gonna affect things much. If we were talking a new card, then yeah, it might, hence why I think performance wise 2 9600's is the way to go.

And power draw wise, the K8WE setup with two SLI'd 9600's is only maybe 40W difference, at the edge of what my PSU can do (recommended PS 620W, minimum 570W) but still within specs (granted, I was using a "quite shit" calculator). As for power bills, We're still under 700W, which is where I draw the "draws enough to care about" line.

Remind me not to ask these kind of questions down the road... since all I get is a bunch of "buy new" crap... I'll say it, it just doesn't feel right putting crap level new stuff on such a well built high end (in it's day) computer. Like having a restored muscle car and putting cheap tires on it... it will work, but you just don't do that. If this were a new build for somebody on a budget, it would make sense... however, that's not my hobby or goal here.

[shovenose]

Re: Ratdude's main rig V3, maybe

I just bought a new 1GB Radeon HD 6570 for my HTPC. Really impressed. Will it SLI nope. But it's better to have a good single card rather than two crsppy SLI cards...
Would recommend the 7770 or 7790 as well

[mariushm]

Re: Ratdude's main rig V3, maybe

SLI won't bring you 90% performance increase. It will vary from game to game, but usually you may have anything between 40% and about 80% performance increase.

The PCI Express bandwidth won't matter much, since the synchronization and important sli stuff would be done through the flex cable.

And otherwise, the difference between pci express x8 and x16 is generally about 1-3% of performance, you lose about 1 frame per second out of 60. Hardly anything to worry about. That ancient gpu is not even capable of saturating a x8 slot bandwidth.

You might also want to check what pci express VERSION is that nForce processional 2200 chipset capable of handling - i have a strong suspicion it can only do pci express 1.1, in which case the bandwidth is only about 8 GB/s compared to pci express 2.0 which can do 16 GB/s, so then your pci express x16 is sort of like a pci express x8. But anyway, it's pointless discussion, the DDR1 can do about 12 GB/s maximum bandwidth and the fact that you have 16 modules doesn't help, afaik it's still only dual channel (so 12gb with dual channel) so pci express x8, x16 ... makes no difference in your case.

Generally, especially with older cards, it's not worth the effort.

9600 GT is based on an ancient gpu. It's a 2008 chip that's basically a re-spin/"upgrade" (minor improvements) of the 8600 GTS that was launched in 2007. The more current "GT 240" is also basically a renamed 9600 GT.. nVidia loves just renaming products and charging the same from crap.

It's also probably supports a maximum of DirectX 10 and OpenGL 2.0, which means bye bye to newer games.

If you're up for paying 60$ +shipping for two shitty cards, you're better off paying about 65-70$ (50-60$ after mail in rebate card but don't count on it) for a card like Radeon 6670:

http://www.newegg.com/Product/Produc...82E16814102988
http://www.newegg.com/Product/Produc...82E16814125430

Maybe about 30-50% faster than a single 9600gt but supports DirectX 11, OpenGL 4.1, dvi+vga+hdmi, about 10w @idle, under 75w at load (compared to about 95w for each 9600gt), support for 3 monitors and eyefinity etc etc

You can see here a comparison between 6670 and 8800GT, which should be actually a bit faster than 9600GT: http://www.anandtech.com/bench/Product/521?vs=613
The 6670 still beats it anywhere it counts.

[Uranium-235]

Re: Ratdude's main rig V3, maybe

you don't really needs SLI do you? Win7 should run on a single 9600 with no problem, linux too. Only if you're gaming would this matter, and even at that some games don't like sli, especially older (pre CUDA) SLI

[ratdude747]

Re: Ratdude's main rig V3, maybe

Here's the situation:

I don't really have the funds to comfortably do this (buying a vehicle on monday, I'll only have $1000 in the account until I get my van sold). Yeah, the money is there, but until I sell that van I don't think $170 is safe to spend in one shot. (Or I might)

But if it was, here's what I'd buy:

K8WE -wasn't able to negotiate, best deal I saw, not even the denutted (one LAN port) HP version could be found cheaper. If you have one and want to sell it, PM me.

2 9600's- they run a lot cooler than the 9800 and also draw less power (this setup draws as much as the current setup actually). Since SLI = 90% boost over a single GPU, that ought to be the kind of power I need to run things well (newer cards either draw more juice or can't SLI).

Flexible SLI bridge - since the above was missing the SLI brige (and rigid ones can't be found less than 70mm), a flexible one will have to do.

That would do it... and down the road, I could upgrade it to SATA opticals (I already have a DVD burner, I could add a blue ray drive if I wanted to). But otherwise my true dream rig would be complete.

Oh, and one more thing: any off topic comments regarding ivy bridge celerons or the like will be deleted from here on out.

^ said, Should I just spend the $170 and be done with it? Or should I wait and pray the current deals don't go bye bye (with no better deal to replace it)? Either way, the current setup is going to go bye bye.. the GPU will be returned to V2, and I'll swap sound cards (and try to undo the AC97 to HD conversion).

edit- FYI, the k8WE's SLI runs at full 16x :

http://www.legitreviews.com/article/152/2/

None of that 8x SLI bulcrap. Should actually run decently. Now I need to convince myself $170 is OK to spend right now (or to find a K8WE for a lot cheaper, I'd settle for maybe $75ish)

[severach]

Re: Ratdude's main rig V3, maybe

Actually, I'm worried about later chip tech. The later the tech, the more points of failure, especially with the die shrinking!

Heat === failure. Little else matters. VIA has the lowest motherboard failure rate and it's wasn't because their tech was better. They ran the coolest. Seems VIA was more interested in low heat than nick'n a few more percents off the benchmarks. It shows.

Cool chips, long life. nVidia wouldn't have gone up in flames nearly so fast had their chips been engineered to run cool.

[370forlife]

Re: Ratdude's main rig V3, maybe

Actually, I'm worried about later chip tech. The later the tech, the more points of failure, especially with the die shrinking!

And it didn't help that Intel decided on LGA socket tech!

At least since Core i7 and i5, there's been reports coming in of fried socket pins!
Apparently from poor contact!

I don't want to say this, consumer hardware is more prone to crap components!

Most server processors use LGA now. In fact, AMD's normal processors still use PGA (with some exceptions), where as their newer Opteron's use LGA.

[RJARRRPCGP]

Re: Ratdude's main rig V3, maybe

Consumer hardware simply has fewer points of failure.

Actually, I'm worried about later chip tech. The later the tech, the more points of failure, especially with the die shrinking!

And it didn't help that Intel decided on LGA socket tech!

At least since Core i7 and i5, there's been reports coming in of fried socket pins!
Apparently from poor contact!

I don't want to say this, consumer hardware is more prone to crap components!

[370forlife]

Re: Ratdude's main rig V3, maybe

I was speaking hypothetically. No way I'd buy one with my own money.

Board came in, no go. The seller is being classy and emailing me a shipping label to return it...

No luck on finding a mobo under budget... the one guy will only sell it as a lot (and it's missing the SLI bridge, which I don't think can be replaced). I may see on the replaceabilty of the SLI bridge... if it's not like i think it is, then I'm going back to V2, since V3 is utterly useless ( I can't get half of my stuff to run).

Lesson: If you want compatibilty, go intel (I've yet to have an issue with true intel non-GPU stuff).

edit- looks like all SLI bridges are too long, as IIRC 70mm is too long. I could use a flexible cable though (they're cheap on ebay).

My board came with the SLI bridge the first time the seller sent it to me. I sent it back to him, and he sent me another without the sli bridge. Figures. I just bought a flexible Asus one off amazon and all is well.

[ratdude747]

Re: Ratdude's main rig V3, maybe

Guys, if thinking about getting a 3770K or 3570K, I would just wait for the "Haswell" platform.

The i7 3820 is a waste of money with the "Haswell" platform around the corner!

I was speaking hypothetically. No way I'd buy one with my own money.

Board came in, no go. The seller is being classy and emailing me a shipping label to return it...

No luck on finding a mobo under budget... the one guy will only sell it as a lot (and it's missing the SLI bridge, which I don't think can be replaced). I may see on the replaceabilty of the SLI bridge... if it's not like i think it is, then I'm going back to V2, since V3 is utterly useless ( I can't get half of my stuff to run).

Lesson: If you want compatibilty, go intel (I've yet to have an issue with true intel non-GPU stuff).

edit- looks like all SLI bridges are too long, as IIRC 70mm is too long. I could use a flexible cable though (they're cheap on ebay).

[RJARRRPCGP]

Re: Ratdude's main rig V3, maybe

Guys, if thinking about getting a 3770K or 3570K, I would just wait for the "Haswell" platform.

The i7 3820 is a waste of money with the "Haswell" platform around the corner!

[Uranium-235]

Re: Ratdude's main rig V3, maybe

these complications come up when it comes to poorly supported hardware. The AMD AGP chip was not something AMD should of done. But it makes sense they would not support it for Win7 cause of its age

you can get a workstation build, but those usually require some moolah

^^and for the person up there who said atom is the new celeron they're wrong. Intel released a low wattage ivy-bridge based celeron BGA chip on many new Mini ITX combo boards. Its the celeron 800 series

rat, there is only so much anyone can do till software makes things impossible