Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Link Speed?

hmm - so any ideas how to get the link speed up to 5.0 GT/s for a 1 GB Sapphire and Nehalem Mac Pro?
You will not feel difference, so dont bother :)

This "link speed" thingy is driving me nuts.
How important is it?
Is a 4870 512MB with a "link speed" of 5 GT/s faster than a 4870 512MB card with a "link speed" of only 2.5 GT/s?
Why can't the flashed video cards get their "link speeds" up to 5 GT/s?
 
This "link speed" thingy is driving me nuts.
How important is it?
Is a 4870 512MB with a "link speed" of 5 GT/s faster than a 4870 512MB card with a "link speed" of only 2.5 GT/s?
Why can't the flashed video cards get their "link speeds" up to 5 GT/s?
OK, here's the deal with 5 GT/s vs. 2.5 GT/s:

2.5 GT/s is the maximum bus speed of PCI-e 1.0 slot (as opposed to PCI-e 2.0, which is 5 GT/s). So this card is essentially providing the same performance it would plugged into a PCI-e 1.0 slot. The card will still get 16 lanes, just at PCI-e 1.0 speeds.

The current argument is that the 4870 will not (under any normal circumstance) saturate the bus at 2.5 GT/s, let alone 5.0 GT/s, so it really shouldn't matter. Maybe when running "software" SLI/Crossfire (where you don't have a physical bridge between the cards) it would make a difference because the cards have to talk to each other a lot via the PCI-e bus so they can cooperate, but in a non SLI/Crossfire situation it will make little/no difference.

If you're still worried, getting a card with more memory (1GB) will help alleviate this issue because there will be less data flowing back and forth between system memory and video card memory. But unless you're loading and unloading assets like crazy (enough to totally fill the card's video memory AND saturate the bus in doing so), you'll see NO difference.

Related thread, non-Mac: http://forum.beyond3d.com/showthread.php?t=45229
 
OK, here's the deal with 5 GT/s vs. 2.5 GT/s:

2.5 GT/s is the maximum bus speed of PCI-e 1.0 slot (as opposed to PCI-e 2.0, which is 5 GT/s). So this card is essentially providing the same performance it would plugged into a PCI-e 1.0 slot. The card will still get 16 lanes, just at PCI-e 1.0 speeds.

The current argument is that the 4870 will not (under any normal circumstance) saturate the bus at 2.5 GT/s, let alone 5.0 GT/s, so it really shouldn't matter. Maybe when running "software" SLI/Crossfire (where you don't have a physical bridge between the cards) it would make a difference because the cards have to talk to each other a lot via the PCI-e bus so they can cooperate, but in a non SLI/Crossfire situation it will make little/no difference.

If you're still worried, getting a card with more memory (1GB) will help alleviate this issue because there will be less data flowing back and forth between system memory and video card memory. But unless you're loading and unloading assets like crazy (enough to totally fill the card's video memory AND saturate the bus in doing so), you'll see NO difference.

Related thread, non-Mac: http://forum.beyond3d.com/showthread.php?t=45229

According to BEIGE in post 1232 of this thread, the difference between the 2.5 GT/s and 5 GT/s in his non-Crossfire setup translates to over 20% drop in performance in the app he uses.
 
According to BEIGE in post 1232 of this thread, the difference between the 2.5 GT/s and 5 GT/s in his non-Crossfire setup translates to over 20% drop in performance in the app he uses.

Mudbox is one of the rare programs that is heavily bandwidth bound. Most applications shouldn't really be too affected if at all, particularly if you get a 1GB card. If games are important to you then its even less of an issue.
 
According to BEIGE in post 1232 of this thread, the difference between the 2.5 GT/s and 5 GT/s in his non-Crossfire setup translates to over 20% drop in performance in the app he uses.
Yes, in Mudbox, sculpting on the GPU (tons of i/o traffic), with a 12-million polygon model. I guess I was looking at this more from a gaming perspective.
 
Yes, in Mudbox, sculpting on the GPU (tons of i/o traffic), with a 12-million polygon model. I guess I was looking at this more from a gaming perspective.

From a gaming perspective, a 1GB 4870 card running on 2.5 GT/s will be better than a 512MB 4870 card running on 5 GT/s. Some games (i.e. Crysis) can saturate 512MB of memory but I don't think any games can currently saturate 1GB.
 
The current argument is that the 4870 will not (under any normal circumstance) saturate the bus at 2.5 GT/s, let alone 5.0 GT/s, so it really shouldn't matter. Maybe when running "software" SLI/Crossfire (where you don't have a physical bridge between the cards) it would make a difference because the cards have to talk to each other a lot via the PCI-e bus so they can cooperate, but in a non SLI/Crossfire situation it will make little/no difference.

If you're still worried, getting a card with more memory (1GB) will help alleviate this issue because there will be less data flowing back and forth between system memory and video card memory. But unless you're loading and unloading assets like crazy (enough to totally fill the card's video memory AND saturate the bus in doing so), you'll see NO difference.

Related thread, non-Mac: http://forum.beyond3d.com/showthread.php?t=45229

That link was somewhat interesting.
One problem I have with it is the discussion is 1 1/2 years old and things change.
One person brought up "gpgpu." Maybe Snow Leopard's OpenCL will provide a bottleneck for the 2.5 GT/s video card?

From a gaming perspective, a 1GB 4870 card running on 2.5 GT/s will be better than a 512MB 4870 card running on 5 GT/s. Some games (i.e. Crysis) can saturate 512MB of memory but I don't think any games can currently saturate 1GB.

That would be good if true because I definitely want a 1GB card.
 
About aty_init

I have read alot about it, but not how it works

From my understanding, it is a kext+driver you download and install

The kext somehow initializes a PC radeon on the Mac?

Do I need a Mac card in slot 1?

Or will a single, PC, say 4890, inserted in slot 1, boot up fine in OSX, with original performance?
 
Getting 2.5GT/s like everyone else and still have the wake from sleep color snow issue which requires me to turn on/off the monitor to get ride of. :(
 
I have the Sapphire card that many others have, flashed with the same method, so here are my experiences since upgrading to 10.5.7:

Plugging in a second monitor still causes a crash. Interestingly, it actually gets going on both monitors, then crashes about 5 seconds thereafter. I got a crash log from the last one, and I will paste it here, in case it's at all helpful:

Wed May 13 18:31:00 2009
panic(cpu 7 caller 0x00194B15): "pmap_flush_tlbs() timeout: " "cpu(s) failing to respond to interrupts, pmap=0xf0b2ae0 cpus_to_respond=0x1"@/SourceCache/xnu/xnu-1228.12.14/osfmk/i386/pmap.c:4582
Backtrace (CPU 7), Frame : Return Address (4 potential args on stack)
0x85a8fba8 : 0x12b4c6 (0x45ec20 0x85a8fbdc 0x13355c 0x0)
0x85a8fbf8 : 0x194b15 (0x465018 0xf0b2ae0 0x1 0x0)
0x85a8fc68 : 0x194e3b (0x85a8ff3c 0x1 0x23d3e 0x0)
0x85a8fd18 : 0x1972db (0xf0b2ae0 0x23d3e000 0x0 0x85a459f0)
0x85a8fd78 : 0x16abc5 (0xf0b2ae0 0x23d3e000 0x0 0x23d5e000)
0x85a8fe98 : 0x16b18d (0x23d5e000 0x0 0x0 0x0)
0x85a8fed8 : 0x1836b5 (0xf08da2c 0x23d3e000 0x0 0x23d5e000)
0x85a8ff38 : 0x379ea7 (0xf08da2c 0x23d3e000 0x0 0x20000)
0x85a8ff78 : 0x3e2e3b (0xf1a3c88 0xe85f1e0 0xe85f224 0x0)
0x85a8ffc8 : 0x1a1bfa (0xe6f44c0 0x0 0x1a40b5 0xe6f44c0)
No mapping exists for frame pointer
Backtrace terminated-invalid frame pointer 0xbfffee18

BSD process name corresponding to current thread: WindowServer

Other thoughts:

I checked system profiler after the update, and I was interested to note a new section that I don't remember being there before, though I could be wrong, and does not appear for my other card: "Television," (see attached picture) which has a "yes" for the 4870. Thinking that analog support might be present through the S-video, I got the adapter plugged in to send composite out to a small TV, but had no result when I used detect displays or tried monitoring out to it via Final Cut - I may not be remembering how to activate it properly, so if anybody knows, or tried something similar with their 3870, please let me know.
 

Attachments

  • Picture 1.png
    Picture 1.png
    90.3 KB · Views: 209
I have the Sapphire card that many others have, flashed with the same method, so here are my experiences since upgrading to 10.5.7:

Plugging in a second monitor still causes a crash. Interestingly, it actually gets going on both monitors, then crashes about 5 seconds thereafter. I got a crash log from the last one, and I will paste it here, in case it's at all helpful:



Other thoughts:

I checked system profiler after the update, and I was interested to note a new section that I don't remember being there before, though I could be wrong, and does not appear for my other card: "Television," (see attached picture) which has a "yes" for the 4870. Thinking that analog support might be present through the S-video, I got the adapter plugged in to send composite out to a small TV, but had no result when I used detect displays or tried monitoring out to it via Final Cut - I may not be remembering how to activate it properly, so if anybody knows, or tried something similar with their 3870, please let me know.
Judging from older screenshots, "television" was listed in the properties before the update.
"ATi display utilites" used to be provided by ATi for its OEM cards. It offers various options, like antialiasing on games that don't support it natively, managing TV out... You could see if the utility manages the card, but I am not sure that it will, since ATi does not sell the 4870 for Macs.

Now regarding the consol log, it seems that the problem is that "No mapping exists for frame pointer" which causes an "invalid frame pointer", terminates "Backtrace" and makes one CPU panic. Not sure what it means, but I guess that the frame buffer (?) does not manages an additional display
I am not even sure if they updated the utility for the 3870 they shipped.
 
Link Speed

I still don't like the idea of having a 2.5 GT/s "link speed."
It is like taking a step backwards, back to PCIe 1.1.
PCIe 1.1 has a bandwidth of 2.5 GT/s, PCIe 2.0 has a bandwidth of 5.0 GT/s.
Many, many video cards advertise the advantages of having their 5.0 GT/s PCIe 2.0 support.

Here is a typical video card advertisement: "PCI Express 2.0 Support - a future-proofing bridge to tomorrow's most bandwidth-hungry games and 3D applications by maximizing the 5 GT/s PCI Express 2.0 bandwidth."

Here's another: "PCI Express Base 2.0 specification doubles the interconnect bit rate from 2.5 GT/s to 5 GT/s in a seamless and compatible manner. The performance boost to 5 GT/s is by far the most important feature of the PCI Express 2.0 specifications. It effectively increases the aggregate bandwidth of a 16-lane link to approximately 16 GB/s. The higher bandwidth will allow product designers to implement narrower interconnect links to achieve high performance."

I don't understand why the flashed video cards are PCIe 1.1 instead of PCIe 2.0.

I am not going to spend money on a new, powerful video card and go backwards in technology.
Unless I can get an HD 4890 or a HD 4870 1GB card that has a 5.0 GT/s "link speed", my next video card will be the GTX 285.
 
As an Amazon Associate, MacRumors earns a commission from qualifying purchases made through links in this post.
You buy a 2,5k Mac and then trying to save 200 bucks at most on the graphics card... me no understand :eek:
Fiddling with cables, soldering your own Molex connectors, dumping, flashing, gimped link speed, no warranty etc...

Simply buy the 4870 from Apple, plug it in and relax like I did ;)
 
You buy a 2,5k Mac and then trying to save 200 bucks at most on the graphics card... me no understand :eek:
Fiddling with cables, soldering your own Molex connectors, dumping, flashing, gimped link speed, no warranty etc...

Simply buy the 4870 from Apple, plug it in and relax like I did ;)

Well if I bought a Mac Pro in 2006 (no more warranty) with 2.5 max link speed anyway, bought the cables straight from ATI (no rolling), bought my XFX card for $175, flashed/installed it in an hour, total cost of $208 saving me $142 and gaining me a second DVI instead of Display Port, plus 1 GB instead of 512 MB memory and the bragging rights that I was able to do it my way instead of Apple's way...well that's why I did it. I'd do it again.

Oh yeah and my productivity with my pro apps has dramatically increased over my 2600XT card.

I'm happy with my choice.
 
You buy a 2,5k Mac and then trying to save 200 bucks at most on the graphics card... me no understand :eek:
Fiddling with cables, soldering your own Molex connectors, dumping, flashing, gimped link speed, no warranty etc...

Simply buy the 4870 from Apple, plug it in and relax like I did ;)
If you have an extra $172 to spend, go for it. To me that's the "easy tax". Like changing your own oil vs. having a shop do it.

But honestly, for a games, the 1GB card even at 2.5 GT/s is better than the retail Apple one. If the apple retail card was 1GB I'd consider getting it, but the 512MB is the real problem.
 
You buy a 2,5k Mac and then trying to save 200 bucks at most on the graphics card... me no understand :eek:
Fiddling with cables, soldering your own Molex connectors, dumping, flashing, gimped link speed, no warranty etc...

Simply buy the 4870 from Apple, plug it in and relax like I did ;)

It really wasn't the money for me... I wanted 1GB.
 
I still don't like the idea of having a 2.5 GT/s "link speed."
It is like taking a step backwards, back to PCIe 1.1.
PCIe 1.1 has a bandwidth of 2.5 GT/s, PCIe 2.0 has a bandwidth of 5.0 GT/s.
Many, many video cards advertise the advantages of having their 5.0 GT/s PCIe 2.0 support.

Here is a typical video card advertisement: "PCI Express 2.0 Support - a future-proofing bridge to tomorrow's most bandwidth-hungry games and 3D applications by maximizing the 5 GT/s PCI Express 2.0 bandwidth."

Here's another: "PCI Express Base 2.0 specification doubles the interconnect bit rate from 2.5 GT/s to 5 GT/s in a seamless and compatible manner. The performance boost to 5 GT/s is by far the most important feature of the PCI Express 2.0 specifications. It effectively increases the aggregate bandwidth of a 16-lane link to approximately 16 GB/s. The higher bandwidth will allow product designers to implement narrower interconnect links to achieve high performance."

I don't understand why the flashed video cards are PCIe 1.1 instead of PCIe 2.0.

I am not going to spend money on a new, powerful video card and go backwards in technology.
Unless I can get an HD 4890 or a HD 4870 1GB card that has a 5.0 GT/s "link speed", my next video card will be the GTX 285.

I have similar feelings and will probably pick up a GTX 285 when it becomes available, too.
 
As an Amazon Associate, MacRumors earns a commission from qualifying purchases made through links in this post.
"PCI Express 2.0 Support - a future-proofing bridge to tomorrow's most bandwidth-hungry games and 3D applications by maximizing the 5 GT/s PCI Express 2.0 bandwidth."

"PCI Express Base 2.0 specification doubles the interconnect bit rate from 2.5 GT/s to 5 GT/s in a seamless and compatible manner. The performance boost to 5 GT/s is by far the most important feature of the PCI Express 2.0 specifications. It effectively increases the aggregate bandwidth of a 16-lane link to approximately 16 GB/s. The higher bandwidth will allow product designers to implement narrower interconnect links to achieve high performance."
Not to argue, but PCI-e 2.0's only important feature is the increased link speed...which, at the moment, is not utilized by any but a tiny, tiny fraction of applications. By the time "tomorrow's most bandwidth-hungry games" come out, I'll probably be ready for a new video card anyway (processing power-wise).

Is there any way to monitor the bandwidth use on a PCI-e slot in real time? That would settle most of these arguments.
 
You buy a 2,5k Mac and then trying to save 200 bucks at most on the graphics card... me no understand
Fiddling with cables, soldering your own Molex connectors, dumping, flashing, gimped link speed, no warranty etc...

well I have both the 512 MB retail card and a 1 GB sapphire that I was hoping to replace it with since I need the extra texture memory. But seeing as it's slower with Mudbox (where my heavy GPU lifting is done), I can't use the 1 GB. I think I'll sell this one and get the Nvidia GTX 285 if it's faster and has more memory.
 
Maybe BearFeats should do a review into the 2.5GT/s vs 5GT/s. If I had a Mac Pro with PCie 2.0 and I had a genuine Apple card then I'd look into it myself. Honestly, for most people, a 1GB 4870 fully clocked card running in a PCIe 1.1 slot will be faster than a 512MB 4870 under clocked card running in a PCIe 2.0 slot.
 
well Cinebench was also slower so YMMV

Do you remember how much slower?

BTW, Warning for anyone using the 4870, official or flashed: Be careful about the quality of DVI cables you use. I bought a dual link cable off eBay for about £10, didn't seem that cheaply built and digital's digital and all that. However, after using the cable for about ten minutes I started getting speckling and noise on my screen. I got worried that it might be the flashed cards but its not. Plugging my ACD in without the extension cable works just fine with no noise. Try the other cable again, noise after a few minutes. The connector gets pretty hot and so I'm led to believe that the heat these 4870s pump out via the fan slot is sufficient to cause poor quality DVI cables to error.

Of course, I only found this out after I spent all afternoon sorting my office out and moving my Mac Pro to the other side of my desk. Its only a 1m extension cable but its now zip tied to several other cables and hidden away with zip ties under my desk. :(

Now I have to wonder, where on Earth can I get a good quality 1m long dual link DVI extension lead in the UK tomorrow without having to resort to waiting until Monday to have one delivered. :( PC World, Currys and Maplin's all seem to do male to male leads rather than male to female. :( Sitting with my Mac Pro now on its side taking up half of my desk is not the result I was hoping for after all this moving and tidying and zip-tieing!! :p
 
it was about 200 OpenGL whatevers slower - basically making it the same as the 8800 GT.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.