NVIDIA's GeForce GTX 465: Cheaper Isn’t Always Better
by Ryan Smith on May 31, 2010 3:26 AM ESTIn recent years it’s become customary to have 3-4 high-end cards on the market at the same time using the same GPU. For the GTX 200 series we had the GTX 260, GTX 275, and GTX 285, while for the Radeon HD 5000 series we have the 5830, 5850, and 5870. With the launch of NVIDIA’s GTX 400 series last month NVIDIA filled in the first 2 spots in their lineup with the GTX 480 and GTX 470, with obvious room to grow out the family in the future.
Above the GTX 480 is of course the “full” GF100 with all of its functional units enabled, and which is still missing in action on both the consumer and HPC markets. However there’s also room for a card below the $350 GTX 470, particularly with AMD being the sole inhabitant of the “bargain” high-end $300 point. NVIDIA is to the point in the Fermi rollout where they want a piece of that market, and they have a stash of further-binned so-so GF100 chips they want to fill it with. This brings us to today, and the launch of the GeForce GTX 465.
GTX 480 | GTX 470 | GTX 465 | GTX 285 | |
Stream Processors | 480 | 448 | 352 | 240 |
Texture Address / Filtering | 60/60 | 56/56 | 44/44 | 80 / 80 |
ROPs | 48 | 40 | 32 | 32 |
Core Clock | 700MHz | 607MHz | 607MHz | 648MHz |
Shader Clock | 1401MHz | 1215MHz | 1215MHz | 1476MHz |
Memory Clock | 924MHz (3696MHz data rate) GDDR5 | 837MHz (3348MHz data rate) GDDR5 | 802MHz (3208MHz data rate) GDDR5 | 1242MHz (2484MHz data rate) GDDR3 |
Memory Bus Width | 384-bit | 320-bit | 256-bit | 512-bit |
Frame Buffer | 1.5GB | 1.25GB | 1GB | 1GB |
Transistor Count | 3B | 3B | 3B | 1.4B |
Manufacturing Process | TSMC 40nm | TSMC 40nm | TSMC 40nm | TSMC 55nm |
Price Point | $499 | $349 | $279 | N/A |
In a nutshell, if you take a GTX 470 and disable some additional functional units, additional memory controllers, and additional ROPs, while turning down the memory speed any further, you get the GTX 465. NVIDIA has shut off another 3 Streaming Multiprocessors (SMs) from GF100, leaving the GTX 465 with 11 of them, giving it a total of 352 CUDA Cores/SPs, and 44 texture units. Meanwhile the ROPs have been cut down to 32 ROPs, and another memory controller disabled, making for a 256-bit memory bus attached to 1GB of 802MHz (3208MHz effective) GDDR5. All told the GTX 465 has around 78% of the texture/shader power of the GTX 470, 80% the ROP power, 76% of the memory bandwidth of the GTX 470, and 80% of the memory capacity. The loss of 256MB of RAM will be particularly interesting, as it means NVIDIA has surrendered its memory capacity advantage over AMD’s reference cards – both are even at 1GB.
With all of that in mind, compared to the GTX 470 the GTX 465 may be the more interesting card. While NVIDIA simply disabled some additional functional units compared to the GTX 480 to get the GTX 470, disabling even more functional units required a different strategy. Rather than disabling additional units from each of the GF100 GPU’s 4 Graphics Processing Clusters (GPCs), NVIDIA outright disabled one of the GPCs. This is the first time we’ve seen them disable a GPC on a GF100 card, making it an interesting first for Fermi. By disabling a GPC, not only does NVIDIA surrender CUDA cores, texture units, and polymorph engines, but they also surrender one of the 4 raster engines. As a result the GTX 465 takes a straight 25% hit in rasterization abilities compared to GTX 470, slightly greater than the loss for any other part of the GTX 465.
Top: GTX 465. Bottom: GTX 470
Along with similar clockspeeds as the GTX 470, the GTX 465 also shares the GTX 470’s design. It’s the same PCB and cooler – only the GPU has changed, with NVIDIA’s partners laying down one of NVIDIA’s GTX 465-binned GF100 GPUs.
With the disabling of additional functional units, the TDP has come down compared to the GTX 470. NVIDIA pegs the GTX 465 at 200W TDP, 15W below the GTX 470’s official TDP. We were not given the idle power consumption; however we’ll see quickly that it hasn’t improved when looking at our own power consumption numbers.
As we stated earlier, this is NVIDIA’s shot at the sub-$300 market, which is currently dominated by the Radeon HD 5850 at $289 and up, and the Radeon HD 5830 at $225 and up. Like the GTX 470, NVIDIA has built a product to slot in between AMD’s cards in terms of performance rather than taking AMD head-on, and the pricing reflects this. The MSRP of the GTX 465 is $279 accordingly, maintaining AMD and NVIDIA’s more-or-less neat division of the high-end market and putting the performance “sweet spot” for the GTX 465's performance at roughly 93% of the 5850.
Meanwhile this is the closest the two have come on pricing in quite some time, as a $279 MSRP puts the GTX 465 within $10 of the cheapest Radeon HD 5850. The pricing on the GTX 465 may change in the next month as NVIDIA’s North American partners are currently packing in Just Cause 2 with the card (a last-minute deal as we understand it), so there may be some flexibility on pricing once that promotion ends and NVIDIA’s partners no longer have to chip in for the game.
Finally, this is a hard launch, a very hard launch. In fact the cards started showing up on etailers 2 days before our NDA expired. After the farce that was the GTX 480/470 launch, it’s fantastic to see a proper hard launch. As far as we can tell you won’t have any problem finding a GTX 465 – thanks in large part to what looks to be quite the stockpile of GF100 GPUs that only meet GTX 465 specifications.
71 Comments
View All Comments
multivac - Monday, May 31, 2010 - link
also, you thanked Zotec at the endrscsr - Monday, May 31, 2010 - link
I read earlier the GTX465 review at computerbase.de and the load temperatures and power are well of both GTX 465 are well below the GTX 470.could it be that there is something wrong with your GTX 465 or your GTX 470 has a particular low power draw?
Ryan Smith - Monday, May 31, 2010 - link
It's always possible.The primary issue is that the quality of chips coming off of a wafer are going to be variable in the best of times, and TSMC's situation is not quite that good. So between any two cards there can be a lot of variability, which is hard to account for when we only have a handful of any given card.
I do not believe that our GTX 465 is "damaged" in any way. It functions just fine. But it consumes a lot of power, and it's hot. We may have simply received a card with a poorer-than-average GPU, which would color our results.
As for our GTX 470, looking at some of the other reviews of it, I don't believe it's particularly exceptional.
For anyone curious, under FurMark the fan on our 470 stabilizes at 76%. On the GTX 465 it stabilizes as 80%. This is the primary reason why our GTX 465 is louder than our 470.
CPUGuy - Monday, May 31, 2010 - link
"I read earlier the GTX465 review at computerbase.de and the load temperatures and power are well of both GTX 465 are well below the GTX 470.could it be that there is something wrong with your GTX 465 or your GTX 470 has a particular low power draw?"
Perhaps they were given a different sample card? Ever thought of that? You folks have to remember that these are not retail cards.
CPUGuy - Monday, May 31, 2010 - link
Why are you taking another review into consideration when they may have been given a different sample? I honestly don't think they had a retail version.MrSpadge - Monday, May 31, 2010 - link
It's said somewhere in the article that nVidia is using different voltages for these chips. So there you have it: Anand tested one at the upper bound of the voltage range (a worse chip) whereas computerbase likely got one which was more in line with the voltage on the 470/480s.Yes, voltage can make such a difference in power draw.
tviceman - Monday, May 31, 2010 - link
This review is sloppy and outright laughable at points. Although you state which Nvidia drivers used (3 of them) you clearly used different drivers and/or old benchmarks for the same game on many parts of the tests, despite the 257 drivers offering some noticeable across-the-board increases. Even if not by much, FPS scores should be slightly different and/or up. In some reviews, it was clear newer drivers were only used on certain cards as the gtx470 is way faster than the gtx480 at BFBC2. Nearly all the other benchmarks, though, have both cards scoring the exact same as they did at release.It's not an accurate, or credible, way to post a review. All tests should be done on the same configuration. same software, same hardware, same drivers. If you don't have time to rerun your gtx470/480 benchmarks with the latest drivers, then you should NOT include them in the review.
tviceman - Monday, May 31, 2010 - link
And just to say, it's not that I disagree with your analysis on the gtx465, but if you are going to have huge graphs showing how all the cards perform and where they fit vs. their counterparts and rivals, those graphs need to be accurate and not using different configurations that will skew results.Ryan Smith - Monday, May 31, 2010 - link
New drivers are always a thorny issue. In an ideal world I'd like to update our results with the latest drivers as they're released. But realistically it's 1-1.5 man-weeks of work to redo the entire benchmark suite and that's just not practical.As a result, for the most part we will keep the same results set for upwards of 6 months. If there's a significant change in performance due to a new driver we'll go ahead and rerun some numbers as necessary, which we have done for the 257.15 drivers. We will probably redo some more benchmarks, but it wasn't possible to get anything else done ahead of this review.
tviceman - Monday, May 31, 2010 - link
Well I think it's only fair, proper, and balanced to run all the cards with the same configuration. If you're unable to do so in the time frame alloted, then only include the cards which are in immediate competition and add more later.The BFBC2 benchmark looks plain WRONG with the gtx470 trouncing the gtx480. And since other gtx470 scores were exactly the same as your original review, it just looks like you randomly picked a few small benchmarks to update with the latest drivers, and left the rest untouched.
It just isn't accurate.