NVIDIA GF100 & Friends speculation

Any idea what they did to only have 4 rops? It appears it's 2 memory controllers, but only one L2 partition and one quad-rop block?
ROPs are dependent upon L2 as I understand it, so a single L2 means only 4 ROPs - i.e. no big deal.

Presumably there'll be an ultra-crappy version with only 64-bits of bus.

I thought this was pretty amusing (http://www.hardocp.com/article/2010/10/11/galaxy_geforce_gt_430_video_card_review/8) :

Based on the performance and pricing, it is clear that NVIDIA’s GeForce GT 430 GPU is not a product that is designed for hardcore gamers. Rather, it is designed for the vast majority of PC users who don’t do serious gaming. Maybe they watch Blu-Ray movies, or do some light video editing or encoding on a specific media PC. NVIDIA claims that the gaming PC market is half as big as the mainstream desktop video card market. For that desktop market, gaming support is nearly an afterthought.
Until the Fusion processors arrive to kill the OEM end of the mainstream desktop video card market...
 
nVnews goofed this morning:

Hanners at Elite Bastards said:
mikec.png
 
Hmm pricing this against HD5670 seems a pretty bold move, as it can't even quite keep up with HD5570. I don't think the ability of 3d bluray is really worth a 15$ premium...
At least idle (and movie play back) power consumption is very good, though load power consumption isn't nearly as good.
 
Considering I assume UVD3 includes this support, then Nvidia will enjoy a few weeks (lol) of exclusive feature where the only market it may make some money would be OEM MPC?
 
That's not good. Our legion of consumer advocates in the Barts thread will have a field day.
GT330 isn't a product. That's some OEM abomination where you never know what you get - could be g92b paired with ddr2 and 96 shaders or pretty much anything else. So yes GT430 is likely slower than SOME GT330 versions but probably not all. With much lower power draw, a better feature set, and what not (and hey at least the GT240 has higher last digits...).
 
ROPs are dependent upon L2 as I understand it, so a single L2 means only 4 ROPs - i.e. no big deal.

Presumably there'll be an ultra-crappy version with only 64-bits of bus.

I thought this was pretty amusing (http://www.hardocp.com/article/2010/10/11/galaxy_geforce_gt_430_video_card_review/8) :



Until the Fusion processors arrive to kill the OEM end of the mainstream desktop video card market...


that particular 430GT does a great showing in that review!, you would expect a lousy card too slow and expensive according to commentaries or impressions you hear but :

- it's unusually faster on the precise one game I would be interested in :)
- it's not that lousy for the rest, esp. given I would still use a 17" crt
- power use is excellent, on par with AMD, and overclock looks easy (first thing to do is bump memory from 800 to 900MHz and here's some free breathing room)
- it's silent! no need to hunt for a special model, here it seems that a pretty regular one will do the job!, you merely have to get a full height, single slot variant.

so you can count me in. I like the fact that I need no dongle nor power connector, thus no minor hassle ; the unusual quietness seals the deal. It's an amusing match with an Athlon II X2 - a CPU that comes with a small and silent stock cooler and has similar TDP. both allow a cheap and quiet PC with no need to buy fans, heatsink or a half decent case :)
 
Do you believe the bit about 4 ROPs in the review guide is true?

If you are looking at TPUs performance summaries it is where it should be with 96SPs and 128-Bit DDR3:
http://www.techpowerup.com/reviews/Zotac/GeForce_GT_430/28.html
=> a bit below ~50% of GTS 450 (192SPs 128-Bit DDR5 16 ROPs), 783 vs 700MHz.

Theoreticaly each ROP can write a single 32bit pixel in each clock with 128 bit bus. Thats quite efective usage of the bandwith. (at least without AA)
And on top of that 2SM-s can output 4 pixels anyway.
 
Why would they lie about the number of ROPs? What dastardly plan would that involve?
Maybe they let the interns do this part?

Theoreticaly each ROP can write a single 32bit pixel in each clock with 128 bit bus. Thats quite efective usage of the bandwith. (at least without AA)
And on top of that 2SM-s can output 4 pixels anyway.

But even with MSAA GT 430 can hold this ~50% of GTS 450.
8xMSAA @ BC2: http://www.guru3d.com/article/geforce-gt-430-review/9 or FC2@8xMSAA: http://www.guru3d.com/article/geforce-gt-430-review/11
 
Last edited by a moderator:
The BC2 numbers are 10 vs 21 FPS. Now thats over 100% difference. FC2 is then 17 vs 38 FPS. Thats also over 100%.

Like I said in first posting: GTS 450 783MHz vs GT 430 700MHz. Theoretically ~45% of TMU and SP performance, BW is 50%.

BC2: ~48%
FC2: ~45%
both @ 8xMSAA

So I really cannot believe that ROP/Z performance of GT 430 is ~22% of GTS 450.
 
Like I said in first posting: GTS 450 783MHz vs GT 430 700MHz. Theoretically ~45% of TMU and SP performance, BW is 50%.

BC2: ~48%
FC2: ~45%
both @ 8xMSAA

So I really cannot believe that ROP/Z performance of GT 430 is ~22% of GTS 450.

Bandwith is 28.8 GB/s vs 57.7 GB/s. And thats close to 100%. GTS 450 has 4SM-s and 8 pixels/clock. Thats also 100% increase.
Maybe a GTS450 would perform the same with just 8 ROP-s. Who knows.
 
Back
Top