Report: GTX 870 Put Through Its Paces

Status
Not open for further replies.

vipboy28

Distinguished
Jan 24, 2007
184
0
18,690
Does anyone this that the 256-bit memory interface is fine? I feel like we dont need 384 to achieve similar results. Anyone feel like it should be 384?
 

xenol

Distinguished
Jun 18, 2008
216
0
18,680
Let's just hope NVIDIA prices these new cards "correctly". It's jarring to see system builders selling $3500-$4000 systems that comes with a $3000 Titan Z standard.
 

xenol

Distinguished
Jun 18, 2008
216
0
18,680
Does anyone this that the 256-bit memory interface is fine? I feel like we dont need 384 to achieve similar results. Anyone feel like it should be 384?
Maxwell has more cache, which is supposed to allow less of a need for higher memory bandwidth. It's part of the energy efficiency improvements they were doing.
 

oxiide

Distinguished


Well, it may be a 256-bit bus but its also 7 GHz memory. That's really, really high; in fact I've heard its the intended limit for the baseline GDDR5 specification. Especially when you consider that this card doesn't sound beefy enough for 4K gaming anyway, I really don't think it will be starved for memory bandwidth.
 

hardcore_player

Distinguished
Apr 11, 2011
44
0
18,530
the Maxwell architecture doesn't need more than 256 bit bandwidth to run effectively , unlike Kepler , which is bandwidth dependent . for instance take a look at GTX 760 .
 

hardcore_player

Distinguished
Apr 11, 2011
44
0
18,530
Maxwell architecture "GTX 800"series wont offer a huge step over GTX 700 , still were waiting till Pascal hits the shelves in 2016 or 4th quarter of 2015 . it presents a 3D Memory that offers several times greater bandwidth, more than twice the memory capacity and quadrupled energy efficiency of today's GPUs , it also has what so called NV-Link which puts a fatter pipe between the CPU and GPU , the flow of data between the CPU and GPU in this state allows data to flow at more than 80GB per second, compared to the 16GB per second available at current GPUs according to Nvidia .
so Maxwell is somehow a refinement to the GTX 700 , adds and sharpen some of the
the hardcore specs and functions of Kepler " GTX 600 " and Kepler Refresh architecture "GTX 700".
 

hannibal

Distinguished
Of course they are priced correctly. About the same speed as 780... about the same price as 780... That has been the trend in the resent years. Maybe they shave 10-20$ of if they feel generous...
All in all this is cheaper card to produce than 780 is, so they can get better margins. (I think so at least, because the chip is somewhat smaller.)
 

Chris Droste

Honorable
May 29, 2013
275
0
10,810
so, will AMD push harder to get some TONGA to the market in time for the holiday buying season? or will the GTX880 lanch just in time to take away single GPU share by besting a 290x in speed, efficiency, AND price? Stay tuned for the next "As the Wafer Yields!"
 

somebodyspecial

Honorable
Sep 20, 2012
1,459
0
11,310


If they price them where you want them (lower), they'd be making as much as AMD. Just about nothing. They haven't made as much as 2007 in 7 yrs. When will people start to understand R&D costs a lot more today, and they CLEARLY are not charging as much as R&D has went up or they'd be making MORE money than 2007 NOT less. Right? Simple math people. If they are gouging us they should be making more money, but that's not the case because they now spend more on R&D than AMD. Their R&D has increased ~50% in the last 4yrs, but profits have been completely stagnant. They are PAYING to give you better stuff, but NOT reaping any benefits financially on the bottom line.

Jarring to see such high prices, but without a the few who DO pay those prices (and can afford them easily), the rest of us would be looking at $1000 mid range cards and $500 bottom rung. You are not the target market for the $1500-3000 cards if you're complaining about them...LOL. I'm not someone who can afford those crazy cards, but I'm sure glad there are enough of them who laugh at those prices and buy immediately upon release so that NV can at least afford to give me a 780ti for $600-700. Without the truly rich buying the ridiculous stuff, that 780TI would surely be $1100+.

With R&D costs skyrocketing, the only way we'll keep pricing the same is by putting more gpus in other devices and expanding the market for gpus/cpus (IE, mobile etc, where lower income people/developing nations can get on the cpu/gpu train in cheaper devices). AMD made 80mil last 12 months. Do you think they're charging enough? Not enough to make you say "RAISE YOUR PRICES AMD!"??? OK, how about this: They lost 6Billion+ in the last 10yrs. Do you think they charge enough knowing that HUGE number?

While NV hasn't lost billions in the last 10, they are nowhere near their historical earnings either and R&D is blowing up for everyone. At some point they will have to PRICE to make more money and AMD should too! Immediately!

You want to talk gouging, start talking Intel who makes $9Bil and is raising prices (haswell $350 now, was $320 last year for top end ivy). Even they are giving a pretty good deal though and their profits are down from 12.3B, to 11B, now to 9B last 12mo (mobile losses is killing Intel's profits, 1.1B/quarter loss on mobile). Same trend as NV basically. Not making as much as before.
 

somebodyspecial

Honorable
Sep 20, 2012
1,459
0
11,310


This is wrong, you're judging maxwell on the 28nm versions which are just to reduce power and give the same perf. The real cards come soon with 20nm and that will be a huge jump in power AND perf (Q1?). AMD will get a good jump from 20nm also. But yes, Pascal will be impressive, but that doesn't mean 20nm Maxwell won't be impressive too, with most of the die shrink going to jacking up perf, where right now they're just dropping power with maxwell's new characteristics while giving basically the same perf or maybe a tad better (and higher margin I hope, if the chips are smaller due to a redesign).
 

Drejeck

Honorable
Sep 12, 2013
131
0
10,690
most useless benchmark ever lol
256 bit memory interface makes a lot of sense, maxwell architecture is really efficient about that.
I have the 750Ti and it outperforms the 6950 I had.
Addressing 2GBs of ram is faster than 4GBs. The 6950 1GB version was slightly faster than the 2GB.
Speaking of resolutions higher than 1920x1080 you need 256 to start without filters, 512 bit though is not yet capable of 4K gaming.
There are lots of limits today, any monitor above 1920x1080 increase a lot (really a lot) frametime variance, and the monitor itself has a lot more of latency. This goes far from pure gamer needs. Triple monitors and 4K aswell (not speaking of screen bezels and pricing).
Maxwell 128 bit seems efficient as 192 bit to me. 256 bit should be like 384.
The higher the bits the larger the chip, the more populated the PCB gets and the higher the power requirements.
My personal hope is that they never take again a chip made for HPC to act as gaming videocard.
They should simply build ad hoc gaming videocard. Not even megachips. To me the 290s are a bad choice. Audio resources on a GPU? I have yet to hear that working but I'd rather go for a 150-170 soundcard. It's cool on a budget APU and mainstream videocard. I hope 870 and 880 brutalize the GK110 spawns in gaming. Maybe when a Phantom comes out I'll think about selling my 770 Phantom.
 

Ninjawithagun

Distinguished
Aug 28, 2007
747
16
19,165
Why hasn't Nvidia advanced to a 512-bit memory bus yet?? I just don't understand!! I do understand that with specific GPU architecture and memory configurations that a 384-bit bus is not always ideal. Fine. So, why not use a 512-bit bus instead of a 256-bit bus. It can't possibly add that much complexity to the fabrication process or to the overall material to make the card. In which case, if either were true, then price would be the ultimate issue. I personally, would be willing to pay for a wider memory interface just for the sake of having that much more headroom to overclock the GPU and memory :D
 

xenol

Distinguished
Jun 18, 2008
216
0
18,680
Why hasn't Nvidia advanced to a 512-bit memory bus yet?? I just don't understand!!
It has nothing to do with manufacturing complexity. Maxwell was designed to not need a wide bus interface because adding a wider bus adds more power consumption. Besides, adding a wider bus may not add much performance anyway, simply because of the way it's designed.
 

semitope

Distinguished
Jun 2, 2006
35
4
18,535


If they price them where you want them (lower), they'd be making as much as AMD. Just about nothing. They haven't made as much as 2007 in 7 yrs. When will people start to understand R&D costs a lot more today, and they CLEARLY are not charging as much as R&D has went up or they'd be making MORE money than 2007 NOT less. Right? Simple math people. If they are gouging us they should be making more money, but that's not the case because they now spend more on R&D than AMD. Their R&D has increased ~50% in the last 4yrs, but profits have been completely stagnant. They are PAYING to give you better stuff, but NOT reaping any benefits financially on the bottom line.

Is this gpu r&d or just general for all their products? Because they can be gouging us in the gpu segment to support all the rest of their pet projects' R&D. As far as GPU, AMD spends enough. They clearly know how to do a lot with what could be a little.
 

xenol

Distinguished
Jun 18, 2008
216
0
18,680
If they price them where you want them (lower), they'd be making as much as AMD. Just about nothing. They haven't made as much as 2007 in 7 yrs. When will people start to understand R&D costs a lot more today, and they CLEARLY are not charging as much as R&D has went up or they'd be making MORE money than 2007 NOT less. Right? Simple math people. If they are gouging us they should be making more money, but that's not the case because they now spend more on R&D than AMD. Their R&D has increased ~50% in the last 4yrs, but profits have been completely stagnant. They are PAYING to give you better stuff, but NOT reaping any benefits financially on the bottom line.

Then how do you explain what I said? System builders are making gaming PCs with a Titan Z STANDARD (as in, it comes with it, not something you add on) for $3000 (Maingear announced a Titan Z build for that much) to $4000. The other components included in the machine don't add up to the difference. The only one I'd say is actually pricing their Titan Z box "realistically" is Falcon Northwest with their Tiki.

Normally if you buy a pre-built system, especially from a boutique, you end up paying quite a bit more than if you just built it yourself from ordering the parts on Newegg.
 

somebodyspecial

Honorable
Sep 20, 2012
1,459
0
11,310

OK, in that case you're complaining about the BOX builder, not NV. I thought you were talking NV pricing. NV/AMD puts out a price for their chips, what people do with them after that is their business.

Agreed, build it yourself and 9/10 times you'll save or just get better stuff in the box.
 

somebodyspecial

Honorable
Sep 20, 2012
1,459
0
11,310


I guess you'd say general for both, but since they both make cpus, gpus,/socs I'd say it's pretty much a wash between them at ~1.2-1.3B each. AMD used to spend more, but it's been coming down 100mil/year for the last 3-5yrs. They're not making any money, so they had to cut it down unfortunately.
 

semitope

Distinguished
Jun 2, 2006
35
4
18,535


I hadn't actually checked the figures. So they aren't even that far apart. I am surprised AMD was spending that much more than nvidia in the past but I guess that was more to do with competing with intel in other areas. The common claim that nvidia spends more on innovating is clearly not so true. Their money might be going elsewhere (marketing, publisher deals)
 
Status
Not open for further replies.