Inquirer claims to have 8600gt/gts 3d mark scores

[PingSpike]

Link

8600GTS/GT benchmarked

Not CeBIT 007 Nice scores, one big problem


By Charlie Demerjian in Hanover: Tuesday 20 March 2007, 07:48

I WAS LUCKY get some quality time with a 8600GTS and 8600GT recently, and happened upon a few numbers you might be interested in. All numbers were measured on an Intel X6800 machine running the latest 3DMark.
The raw scores are 5500 3DMark06 for the GTS and 4800 for the GT, blurred slightly to protect the guilty. That is the good part. The bad part is when you start using heavy textures, performance drops off notably, and I mean notably. Think cliff.

This is most likely because of two things, drivers or bus width. NV drivers are still pretty badly broken, and upcoming cards are probably less of a priority than getting the ones they released almost 6 months ago functional.

The other thing it could be is the narrow memory bus simply choking on all the data. If this is the case, don't look for improvements, this card will always be a benchmark special.

There could be a host of other things, from pre-production boards (doubtful in this case), to sunspots, but the performance drop is quite real. Keep an eye out for this when the boards are finally released on April 17. µ

Hard to say...he thinks they take a hit from the memory bandwidth, how much ram were these cards suppose to have?

 

[Acanthus]

The cards are crippled by a 128bit bus.

Im shocked (i predicted this when i heard the specs about a month ago)

 

[Sniper82]

Looks like I am gonna hold off till June if this is the case. I had a x1800xt 512mb and I want something that out performs it by quite a bit. I might save for a 320mb 8800GTS or wait for some 8900 series.

 

[agathodaimon]

dang. I was waiting for these 8600 cards.
Maybe I'll just wait a bit longer for ATI's offering.

 

[hectorsm]

I guess I don't understand why people are disapointed. A 5500 score is about half what the 8800GTX scores for less than half of the price. For a midrange card it looks good to me.

Am I missing something?

 

[Matt2]

Originally posted by: hectorsm
I guess I don't understand why people are disapointed. A 5500 score is about half what the 8800GTX scores for less than half of the price. For a midrange card it looks good to me.

Am I missing something?

Yeah, I really dont get why people had such a hard on for these cards. When you pay mid-range price, dont expect anything more than mid-range performance. Simple as that.

 

[chizow]

I think we saw a similar performance hit with larger textures with the 9600pro vs. 9700/9800pro, so it makes sense. I still think the 384-bit and 320-bit buses on the GTX/GTS are overkill for single core/card configurations, but looks like 128-bit is definitely a bottleneck.

Still, at $149-$199, a lot of people are going to be happy with this part in the mid-range. Depending on how well they OC, they could still be very competitive parts in the low high-end compared to parts like the 320MB GTS.

 

[aka1nas]

Thats around the same territory as my X1900XT. Not bad at all for that pricepoint and size/power footprint.

 

[Acanthus]

Originally posted by: Matt2

Originally posted by: hectorsm
I guess I don't understand why people are disapointed. A 5500 score is about half what the 8800GTX scores for less than half of the price. For a midrange card it looks good to me.

Am I missing something?

Yeah, I really dont get why people had such a hard on for these cards. When you pay mid-range price, dont expect anything more than mid-range performance. Simple as that.

Because i want something that i can play games on without AA at decent peroformance for $200ish.

Its the only setting im willing to turn down to save money.

That and the card is crippled, had they made it 256 bit they could have used cheaper ram and got better results.

 

[Matt2]

It would have also increased performance and stepped on the 8800GTS 320mb's toes.

Not to mention that creating a card with a 256bit bus as opposed to 128bit bus would have increased the cost far more than cheaper RAM would have offset. Your $200 card quickly turns to $280 and then you might as well get a 8800GTS.

 

[Acanthus]

Originally posted by: Matt2
It would have also increased performance and stepped on the 8800GTS 320mb's toes.

Not to mention that creating a card with a 256bit bus as opposed to 128bit bus would have increased the cost far more than cheaper RAM would have offset. Your $200 card quickly turns to $280 and then you might as well get a 8800GTS.

If you think adding layers to the card adds $80 in costs you are severely mistaken

 

[Cookie Monster]

Link

8600GTS: 7000
8600 GT: 5800

And for the love of god, they got that info from this site.

Link

Notice that those were scored using an A64 single core CPU?

So yea, i believe the 8600GTS will be a quite a match for the 7900GTX. And finally, stop with the 256bit nonsense. YOU DONT NEED 256bit if you can compensate it with using higher clocked memory. The only reason it wasnt possible back then is because DDR2 was hot for a memory chip, DDR1 couldnt be clocked high due to limitations and lastly they were expensive at that time. But nowadays where GDDR3 is mature as you get, dont even need cooling on the memory, and reaching 2000mhz (1.0ns) theres no reason why you should increase memory interface because the performance/price ratio for such an addition isnt worth it overall. (die size increases, transistor budget, more complex PCB, no of memory chips etc)

Show me the situations of why you think the card is being bandwidth starved. Current gen hardware seems to indicate that there arent many bandwidth starved scenarios. A clear example is seen by the X1950XTX compared to the X1900XTX.

I dont see why your stressing that the mid range card has to be 256bit. How about 7600GT vs the X1800GTO. Memory interface/bandwidth may matter, but its the architectural power thats the most influential in performance. Just look at the 7600GT vs the 6800ultra or how about X1650XT vs the X850XTPE.

 

[Acanthus]

Lets find someone with an 8800GTX, and have them cut the memory clock in half to simulate having a 192bit bus instead of a 384.

Or do the same with a 8800GTS, or even a 7800GTX.

A heavy performance dive with high textures indicates a memory bottleneck, either in size or bandwidth.

 

[SexyK]

Seriously, I don't get what you people are complaining about. If you want more memory bandwidth then pay for a higher-end card. It's really not that complicated. Of course this card is "crippled" compared to the 8800's, it's priced accordingly. If it wasn't "crippled" then wouldn't it be a high-end card? As it stands, it seems like these will be great cards for the target audience (i.e. someone willing to shell out some cash to play the latest games but not willing to break the bank for ultra-high resolutions and extreme levels of AA).

 

[chizow]

Originally posted by: Acanthus
Lets find someone with an 8800GTX, and have them cut the memory clock in half to simulate having a 192bit bus instead of a 384.

Or do the same with a 8800GTS, or even a 7800GTX.

A heavy performance dive with high textures indicates a memory bottleneck, either in size or bandwidth.

Ya exactly. I do think the 384/320-bit bus on the GTX/GTS is overkill in most situations, but 128-bit looks to definitely choke the 8600s. Cookie Monster brings up some good points with the 7600GT, but the 7600GT also had much higher memory clock speeds then comparable 256-bit parts (1400-1600 effective vs. 800-1200 effective).

The memory clocks on the G80 and G84 parts look to be the same with GDDR3 running @2000. Also, increasing memory clocks on the G80 past stock speeds have very little impact on performance which is probably because of the massive memory bus.

Lastly, the G8x stream architecture also benefits more from a wider memory interface with channels dedicated to independent stream processors, so direct comparisons to past parts isn't fully relevant.

From AT's G80 Overview:

"The memory interface has been dramatically redesigned to support the access patterns of all of G80's independent stream processors. Given the theme of increasing granularity within G80 it's no surprise that we are now seeing 5 and 6 channels of GDDR rather than the 2 or 4 channels we have been used to for the past few years. 8800 GTX will have a 384 bit bus (6 x 64-bit channels), while the 8800 GTS will have a 320 bit wide connection to DRAM (5 x 64-bit channels). We would love to delve further into the details of G80's new memory interface, but NVIDIA isn't discussing the details of this aspect of their hardware."

Wish AT would follow up on this and fill in the blanks, since its been a while since there was a major difference in the bus size for same generation parts (7600GT and 9600Pro are the only ones that really stick out in my head). Hopefully they revisit this in their G84 review/overview. But clearly going with 2 x 64-bit instead of 4, 5 or 6 x 64-bit will save NV a lot of money without sacrificing too much performance from their point of view.

 

[Cookie Monster]

We will have to wait for the benches.

 

[Ichigo]

Originally posted by: Cookie Monster
Link

8600GTS: 7000
8600 GT: 5800

And for the love of god, they got that info from this site.

Link

Notice that those were scored using an A64 single core CPU?

So yea, i believe the 8600GTS will be a quite a match for the 7900GTX. And finally, stop with the 256bit nonsense. YOU DONT NEED 256bit if you can compensate it with using higher clocked memory. The only reason it wasnt possible back then is because DDR2 was hot for a memory chip, DDR1 couldnt be clocked high due to limitations and lastly they were expensive at that time. But nowadays where GDDR3 is mature as you get, dont even need cooling on the memory, and reaching 2000mhz (1.0ns) theres no reason why you should increase memory interface because the performance/price ratio for such an addition isnt worth it overall. (die size increases, transistor budget, more complex PCB, no of memory chips etc)

Show me the situations of why you think the card is being bandwidth starved. Current gen hardware seems to indicate that there arent many bandwidth starved scenarios. A clear example is seen by the X1950XTX compared to the X1900XTX.

I dont see why your stressing that the mid range card has to be 256bit. How about 7600GT vs the X1800GTO. Memory interface/bandwidth may matter, but its the architectural power thats the most influential in performance. Just look at the 7600GT vs the 6800ultra or how about X1650XT vs the X850XTPE.

Let's not ignore this post.