VOGONS


First post, by squiggly

User metadata
Rank Member
Rank
Member

I am planning a 440BX build soon enough...and of course I want to run it at 133fsb. I have read lots of articles talking about the stability of GPUs at ~88Mhz AGP bus - but not many talk about the actual GPU performance increase as a result.

Does the actual GPU core frequency increase with the AGP bus speed? And VRAM? Is there a difference between overclocking a GPU using powerstrip to the same increase that would come from overclocking the AGP bus? I guess you get faster transfer rates with a higher bus speed as well.

I am planning on using a Quadro2 Pro which from what I have read will be very stable at a higher frequency, however I was also thinking of overclocking it, but not sure how that interacts with the overclocked bus.

Reply 1 of 2, by The Serpent Rider

User metadata
Rank l33t++
Rank
l33t++

but not many talk about the actual GPU performance increase as a result.

Because it can't be separated from FSB increase and major performance boost that comes with it.

I must be some kind of standard: the anonymous gangbanger of the 21st century.

Reply 2 of 2, by dionb

User metadata
Rank l33t++
Rank
l33t++
squiggly wrote:

I am planning a 440BX build soon enough...and of course I want to run it at 133fsb. I have read lots of articles talking about the stability of GPUs at ~88Mhz AGP bus - but not many talk about the actual GPU performance increase as a result.

On a BX you can't separate these, but you could easily test this on a board that can do both 2/3 and 1/2 AGP dividers (i.e. most Via ApolloPro133 or indeed i815 based ones)

Does the actual GPU core frequency increase with the AGP bus speed? And VRAM? Is there a difference between overclocking a GPU using powerstrip to the same increase that would come from overclocking the AGP bus? I guess you get faster transfer rates with a higher bus speed as well.

Faster AGP bus doesn't affect clocks on the VGA card as the card has its own clock crystal and derives both core and RAM clock from that. VRAM isn's a special case, it should behave the same in that sense as plain DRAM, SGRAM, SDRAM or indeed any other mem tech (MDRAM, RDRAM...)

I am planning on using a Quadro2 Pro which from what I have read will be very stable at a higher frequency, however I was also thinking of overclocking it, but not sure how that interacts with the overclocked bus.

Basically there are three potential effects:

1) if the bus speed is too high for the card to handle, things fail. Pretty obvious one. Assuming the card can handle the bus speed not an issue, at which point the other two come into play.
2) altered latencies. Generally faster clocks lead to lower latencies, but if the core/mem clocks are clean multiples of 66MHz and not 88MHz it's possible that average latency might actually increase. Unlikely though, and probably insignificant IRL either way.
3) higher bus bandwidth. Running the bus faster lets more data be pumped through it. This is great, but will only affect overall speeds if the 66MHz AGP 1.0 (2x) bus was the bottleneck in the first place. Generally, AGP (and indeed PCIe) buses are almost never bottlenecking for period card when gaming. Only in corner cases with cards much newer than the bus (or SLI, but that's not relevant for AGP) will bus bandwidth be relevant. On workstation loads (rendering) this might be different- so if you intend to use that Quadro Pro 2 for what it was designed for it might have an impact.