[staryoshi]

GTX 460 1GB vs. GTX 465: Battle of the Fermis

With the release of the GTX460 now a month behind us the dust has started to settle and the nVidia GPU pricing landscape is falling into place. This leaves consumers with a few compelling choices in the performance mid-range segment. One of the most hotly contested and intriguing of them is the competition between the 1GB GTX460 and GTX465 lines of graphics cards.

Both present an outstanding value and both have their unique strengths and weaknesses. I have compiled a short but telling list of benchmarks comparing these two cards at varying clock speeds spanning several popular synthetic and game benchmarks to give you a good idea of their relative performance. I'll also chime in on the ancillary attributes of the cards as well and provide a brief summation of my opinions. I hope you find this review relevant and useful for your GPU-purchasing process.

As always, please feel free to ask me any questions and don't hesitate to post!

Thanks for dropping by,
StarYoshi

Introduction

With the introduction of the GTX460 line of graphics cards, the GF104 Fermi architecture has been unleashed upon the world. How does this compare to the previous GF100 architecture? Lets investigate briefly.

GF100
The original GF100 architecture, as seen in the GTX465, GTX470, and GTX480 lines of graphics cards, is comprised of 4 graphics processing clusters (GPC), each containing 4 simultaneous multiprocessors (SM). Within each SM are 32 CUDA processing cores. This would theoretically allow for a GPU powered by 512 CUDA cores. However, nVidia's current flagship graphics card, the GTX 480, has a single SM disabled resulting in a total of 480 processing cores. Each GF100-based GPU features a unique memory bus and size as well. The GTX480 sports a 384-bit bus and 1536MB of memory, the GTX470 320-bit and 1280MB, and the GTX465 256-bit and 1GB.

The GTX465 model is significantly cut down, with 5 SMs disabled. This results in a total of 3 GPCs, 11 SMs, and a total of 352 CUDA cores.

Pictured below is a high-level block diagram of the GF100 architecture

Below is the GTX465 GF100 diagram, grabbed from Tom's Hardware.

GF104
The new architecture debuting in GTX460 graphics cards has been dubbed GF104. It's Fermi alright, but it certainly differentiates itself from its older brother. The GF104 features 2 GPCs (Reduced from four), each containing 4 SMs (As with GF100). However, each SM contains 48 CUDA processing cores, a 50% increase from GF100's 32 cores. Theoretically this will allow for a graphics card containing 384 CUDA cores. Like the GTX480, the GTX460 has a single SM disabled. This results in a graphics card featuring 336 CUDA processing cores. The GTX460 comes in two flavors: one featuring 1GB/256-bit and one with 768MB/192-bit memory and bus widths respectively. There are other technical details which I will not address in this review. For further information consult one of the many online articles regarding the Fermi architecture.

Pictured below is a high-level block diagram of the GF104 architecture

GTX460 vs GTX465 - Specifications
Here's a quick chart I put together for both cards. It's far from comprehensive but it contains useful info.

Images: GTX460 and GTX465
Here are a few images for your viewing pleasure. Note differences in PCB length, heatsink/fan size, connectors, etc.

Test System Configuration

Processor: Intel Core i7 860 @ 4.0Ghz (w/HT)
Memory: G.Skill ECO DDR3-1333 7-7-7-21 @ DDR3-1600 8-8-8-24
Motherboard: Asus Maximus III Gene
Hard Drive: Samsung F3 Spinpoint 1TB
Power Supply: Corsair 850HX
Operating System: Windows 7 Home Premium 64
Drivers
-Display: Forceware 258.96
-PhysX: 9.10.0513

Overclocking

Both the GTX460 and GTX465 have proven to be very capable overclockers. The MSI Cyclone model comes clocked at core/memory speeds of 725/900(3600 Effective) - My particular model reads 726Mhz for the core. The PNY GTX465 is clocked at 607/802(3208) by default. I was able to reach a maximum overclock of 825/1000(4000) on the GTX465 at 1.087v. Temperatures were very warm and the fan was run at high-speed. It is not a configuration I would run 24/7 due to the noise pollution. For testing, I ran the GTX460 at overclocks of 825/1000(4000) for direct comparison clock-for-clock and 925/1050(4200) at 1.087v for a maximum overclock. Temperatures and noise levels are immensely more tolerable for this card at its maximum overclock. I will go into more detail regarding fan profiles, noise, and temperatures later in the review.

Synthetic Benchmarks

-3dMark 06
--This reliable standby is mildly interesting at this point. It's too CPU-limited to be a true measure of GPU strength at its default settings, but I'll include anyway. I have the performance figures, why not?

-3dMark Vantage
--Much more relevant than 3dmark06, Vantage is a great gauge for relative GPU strength as it is capable of tearing up even the most high-end of graphics cards. Testing was conducted at performance and extreme presets.

-Stone Giant
--A DirectX 11 benchmark featuring varying grades of tessellation. Tested at 1680x1050 and 1920x1200 resolutions with tessellation set to high.

-Unigine Heaven
--The most prevalent tessellation benchmark, like Stone Giant it's a good measure of a graphics card's tessellating capabilities.

Game Benchmarks

-Batman Arkham Asylum
--A fantastic game, Batman AA implements PhysX well and should prove to be a useful benchmark.

-Dirt 2
--Run at 1680x1050 and 1920x1200 resolutions using the High preset and 4xAA. Dirt 2 is a good measure of DX11 GPU performance, utilizing new technologies to deliver a graphically-breathtaking experience.

-Far Cry 2
--A bit outdated, but still a good measure of GPU performance. Tends to favor nVidia cards.

-Just Cause 2
--One of the most graphically demanding games out there featuring one of the largest virtual landscapes to date with incredible detail, Just Cause 2 is my favorite sandbox game in years. I disabled GPU water simulation for the purpose of benchmarking. Testing was run at 1680x1050 and 1920x1200 resolutions at the specified settings.

-Mafia II
--The newly released Mafia II demo features a built in benchmark. The game makes heavy use of PhysX, which would appear to be poorly-implemented currently, but that should change. Relative performance measures are still possible to obtain and will be useful.

-Resident Evil 5
--Not a game that I care for, but the benchmarking tool is useful. Testing was run at 1680x1050 and 1920x1200 resolution at max settings using the fixed benchmark.

-Stalker: Call of Pripyat
--Testing was run at 1680x1050 and 1920x1200 resolution at the pictured settings. The "Day" and "SunShafts" benchmark scores are included.

-Street Fighter IV
--While not the most resource demanding of games, SFIV scales very well and is useful for measuring relative performance. Testing was run at 1920x1200 resolution at max setting using the "Ink" effect.

 

[staryoshi]

Benchmark Results and Analysis
-Enjoy!

Synthetic Benchmarks
-3dMark 06, 3dMark Vantage, Stone Giant, Heaven

3dMark06

As I stated earlier, 3dmark06 is too CPU limited to be truly relevant for most modern GPUs at the default (1280x1024) setting. We see a CPU-bottleneck here, and that's with 8 threads of core i7 goodness running at 4.0Ghz. Both cards post respectable scores at stock and jump a bit higher when overclocked. The overclocked GTX465 manages to top 25k 3dmarks. Very respectable.

3dMark Vantage

I have chosen to list only the GPU scores for the Vantage runs. The CPU score was around 24,500 for all runs, so their overall scores would all be bumped up similarly. (GPU PhysX was disabled). Here we see the GTX460 best the GTX465 by nearly 2000 points in performance mode at stock setting, a 14.3% difference. However when run at the same overclocked setting of 825/1000 we see a much different picture. The GTX465 edges the GTX460 by about 600 points, or 4%. However, once further overclocked, the GTX460 takes back the crown, managing a 1000 point victory (6.6%) over the GTX465. When run on Extreme, we see a very similar pattern. At factory settings, the GTX460 wins by 15.7%, an even larger gap than Performance. Overclocked to 825Mhz core, the GTX465 edges the GTX460 by 2.9%. After another clock increase to its maximum overclock, the GTX460 strikes back with an 8.4% victory. What we see here is the following trend: At stock settings (Factory OC for the GTX460 - Should hold true at its reference 675Mhz core clock speed) the GTX460 thoroughly beats the GTX465. However, once both are overclocked to the same 825Mhz and 1000Mhz core and memory clocks respectively, the GTX465 manages to hold off the GTX460 and pull out a small victory. When the GTX460, not running at its full potential, squeezes out another 100Mhz core clock increase and another 50Mhz memory clock increase it again pulls ahead of the GTX465. We will see if this trend continues in further testing.

Stone Giant

Making heavy use of tessellation, the Stone Giant benchmark allows the GTX465 to flex its muscles, taking advantage of its GF100 architecture. At factory settings, the two cards perform exactly equal, managing the same frame rates at both resolutions. At the clock-for-clock (from here on abbreviated CFC) overclock, however, the GTX465 pulls away, managing 10 and 9 higher FPS averages at 1680x1050 and 1920x1200 resolutions respectively. When overclocked further, the GTX460 manages to pull within 3 FPS at both resolutions. The GTX465 has a higher capacity for tessellation than the GTX460, but due to its cut-down specifications the GTX460 is able to hang with it when clocked further.

Heaven

Heaven follows the established trends. At factory setting, the GTX460 tops the GTX465 by 8.8% and 10.2% at 1680x1050 and 1920x1200. At the CFC setting the GTX465 wins by 7.7% and 6.1% respectively. When overclocked further, the GTX460 nudges the GTX465 by 1/2 a frame at 1680x1050 and by a frame and a half at 1920x1200. An interesting trend to note: The performance gaps widen or shorten in favor of the GTX460 as the resolution increases from 1680x1050 to 1920x1200, which would suggest to me that aside from situations when the GTX465 can take advantage of the GF100's architectural power, the GTX460's GF104 design processes more efficiently... My two cents anyway.

Game Benchmarks
-Batman Arkham Asylum, Dirt 2, Far Cry 2, Just Cause 2, Mafia II, Resident Evil 5, Stalker: Call of Pripyat, Street Fighter IV

Batman Arkham Asylum

Batman AA is an exception to the CFC testing trend. The GTX460 is capable of besting the GTX465 at all clock settings, even if only marginally so at the CFC setting. The GTX460 wins convincingly here, sometimes managing double-digit average frame rate improvements. It may have to do with each card's Physx processing capabilities as well.

Dirt 2

Ah, Dirt2. A reliable staple of my DirectX 11 benchmark suite. It's still one of the prettiest games out there in my opinion. Here the GTX460 wins by 11.7% and 14.7% at factory setting run at the respective resolutions. CFC however, the GTX465 overtakes the GTX460 (seems to be a recurring trend, doesn't it?) by slim margins of 2.9% and 1.6% respectively. Upon further OC the GTX460 storms back to win by 6.9% and 8.2% respectively. Both previously-mentioned trends are fully realized here.

Far Cry 2

Far Cry 2 is an oldie (relatively speaking) but a goody. It still looks darn good in my opinion. Traditionally, it does tend to favor nVidia cards and as we see here we get some very satisfactory frame rates from both cards. The GTX460 nets gains of 14.2% and 15% at factory setting. CFC the GTX465 oh-so-slightly nudges the GTX460 by 1.8% and 0.9%. A victory is a victory no matter how small! (I suppose) Once cranked, the GTX460 wins again by 7.7% and 9.2%.

Just Cause 2

How does the GTX46x series handle the jaw-dropping visuals of Just Cause 2? Very well, actually! We see a small victory by the GTX460 at factory setting, netting almost 2FPS gains at both resolutions. However at CFC setting the story becomes very different. Here the GTX465 wins by 14.2% and 13.3%, a very significant lead. Even overclocked further the GTX460 cannot overcome the GTX465, who retains a lead of 3.6% and 0.4% respectively. This is the first game that the GTX460 has been unable to overtake the GTX465, even if it's only by a super-slim margin. Will we see it happen again?

Mafia II

I'm sure that the engine is still very much so a work in progress, as these figures are disheartening, but we can still see relative performance differentials here so it's fairly useful. The GTX460 nudges the GTX465 at factory and maximum overclocked settings, but the GTX465 strikes back hard when running CFC, managing gains nearing double digit percentages. I am very interested to see the game running at fully-optimized levels, as the demo was quite enjoyable.

Resident Evil 5

We return to normalcy with Resident Evil 5. The GTX460/465/460 victory pattern holds. I'm going to be lazy on this one! Factory: GTX460 by 9.7% and 10%. CFC: GTX465 by 5.7% and 5.1%. Max OC: GTX460 by 3.1% and 4.1% Nothing exciting or surprising here.

Stalker: Call of Pripyat

We see another trend-fighter here with Stalker: CoP. There are four tests run by the built-in benchmark. For the sake of space I have only included two of the tests. The "Day" test yields the highest frame rates on average and the "SunShafts" tests yields the lowest. At factory setting the GTX460 nudges the GTX465 by between 2.1% and 4.1% across all settings. CFC the GTX465 by 10-15%, with a wider gap at 1680x1050 than at 1920x1200. The GTX460 is still beaten when taken to its highest OC, losing by 1.8-3.7%. Like Just Cause 2, Stalker favors the GTX465 when overclocked.

Street Fighter IV

A reliable measure of GPU-scaling, SFIV has served me well. We're talking about frame rates well past 100 FPS, but we can still see helpful relative performance trends. At factory setting the GTX460 nudges the GTX465. However at CFC and maximum overclock settings the GTX465 overtakes it by 14.5% and 4.2% respectively.

That concludes my benchmarking figures. See the next post for further analysis and my conclusion.

 

[staryoshi]

Scalability Analysis

Both the GTX460 and GTX465 offer impressive overclocking potential. With this, they also offer high scalability.
(Scaling performance measured on average frame rates)

This chart measures the relative performance of the two cards at stock and overclocked settings. Both against themselves and each other. Blue scores denote a GTX460 performance victory, green for GTX465.

Both of these cards are impressive in their own right. The GTX465 has awesome overclocking potential, perhaps even more than many GTX460s depending on the model and factory overclock. At factory settings the GTX460 benefits from a 50Mhz overclock. As a result, it claims victory in all of the benchmarks I ran. However, the tables turn when both are set to the same clock rates, resulting in performance wins in every benchmark except Batman. At its maximum overclock, the GTX460 generally beats the GTX465 at its best with a few exceptions.

Despite being a cut-down GF100, the GTX465 is a very capable GPU with overclocking headroom to spare. If you're lucky enough to unlock it to a GTX470 (I did not try with this model as it has a blue PCB) then you'll tear through games like nobody's business.

Both cards manage to scale very well relative to their core/memory clock speed increases. After receiving a 27.6%/16.7% core and clock boost respectively (37% higher than the reference core clock), the GTX460 manages an average improvement of 23.2% across all benchmarks and settings. The GTX465 managed a whopping 35.9%/24.7% core and memory clock speed increase respectively. This resulted in an average improvement of 30%. The scalability of this card is staggering.

Thoughts
Despite the tedium known as benchmarking (It's an awesome thrill until it becomes monotonous and tiresome), I had a great time benchmarking these cards. Some results were expected, others were not. In the end, I was able to unleash the power of these cards without worry of a CPU-bottleneck and gather useful data.

Temperature and Noise (informal brief)
Regarding heat and noise, the GTX460 is nothing short of a feat of engineering with regard to both AIB partners' heatsinks and nVidia's GPU/PCB design. The MSI Cyclone runs like a dream. It idles at a fan speed of 40%, which is not audible over my other case fans. This is important to me, as reducing noise pollution is one of the key factors in my component selection. Furmark testing yielded temperatures in the mid 50's (C) when left on auto. The fan speed did not exceed ~70%, which is very tolerable when the GPU is under that amount of stress. Idle temperatures vary from ~28-29C with the fan set to auto depending on ambient temperatures (measured @ 24C ambient). When overclocked to 925/1050, I was seeing game-load temperatures of about 70C with the fan set to 70%. That's impressive as well. This particular GTX465, however, features the reference turbine heatsink. I am terribly biased against this design, as it lacks both acoustic and thermal efficiency. When in idle, the fan speed is set to 40%. This does not change until the card reaches loads exceeding 70C. 40% is too loud for idle use in my opinion, as it is audible over my other fans. Load fan speeds are fairly unbearable for my taste. With am ambient temperature of 24C, my auto fan speed temperatures were 37C idle, 77C (49% fan speed) Furmark load. That's a respectable showing, but like I said, too loud for my taste. At a fan speed of 100% (unbearable!) the temperatures were reduced to 34C idle 62C load. Not too shabby if noise is not an issue. When overclocked, the GF100 chip showed me what it's like to keep a thermal-nuclear reactor next to my monitor, managing a load temp of 88C in Furmark at 100% fan speed. I was able to tolerate it for the purpose of benchmarking, but just barely.

Beyond Performance
For me, performance is just one metric important to graphics card selection. One should also consider PCB length, power requirement, heat and noise output, included outputs, features, etc. When it comes to a comparison of ancillary benefits, it is my belief that the GTX460 wins hands down. Featuring a PCB shorter by 1.25" (15%), a TDP lower by 40w (GTX460 160w / GTX465 200w), and higher factory clocks (almost always) the GTX460 has much to offer beyond its graphical performance. One of the few areas that it does not improve upon GF100 is that it lacks the ability for Tri-SLI. It's a feature reserved for enthusiast models, and it's not particularly missed in my opinion. (The GTX460 scales incredibly well in SLI as well, check out my MSI GTX460 SLI review if you're interested).

Conclusion
The Fermi line of graphics cards continues to evolve. Both in its expansion to performance mainstream and mid-range GPUs as well as its continual driver support. The GF104 is particularly impressive and a triumph for nVidia. Their failing, however, is that the release of the GTX460 (and other Fermi models) is several months late. They are winning back the hearts of many, but they surely have lost out on countless sales due to their lagging release schedule.

Both the GTX460 and GTX465 present an amazing value in the DirectX 11 performance segment. The price of the GTX465 have plummeted, making them very compelling. Which one is best for you depends on your needs. If you're space-limited, power-limited, concerned with power consumption, heat output, or noise levels, the GTX460 1GB is definitely the graphics card for you. If you are not interested in overclocking, it will perform better than the GTX465 out of the box too. It is also a blast to overclock as well. If your only concern is frame rates, pure frame rates, and you're an aggressive overclocker the GTX465 is the way to go. This card has oodles of overclocking headroom. However, be prepared to deal with burning hellfire and wind torrents if you stick with reference cooling. A key advantage for the GTX465 is aftermarket cooling compatibility. Since it's a cut-down GTX470, your options will be more plentiful than the GTX460, which has a glaring void of aftermarket solutions available (however, man AIB partners provide plenty of factory non-reference solutions for the GTX460).

I can honestly say that both cards present an awesome mix of value and performance. Depending on your needs, either would be awesome addition to most any rig. Based on my personal preferences however, I would recommend the GTX460 1GB model as my GPU of choice in the $200-250 market segment every time. It's a triumph of engineering that, albeit came rather late, it's making the GPU market more competitive and the consumer will reap the benefits.

I hope these benchmarks prove useful in aiding your GPU purchase decision making process or satiating your curiosity. I had to know the truth for myself!

As always, please comment and ask any questions you may have.

Thanks, Happy Gaming!
-StarYoshi

 

[Kand]

Super! Subscribed. =D

Just 825Mhz at 1.087V? My 465 achieves that at stock volts (0.987v). =P

 

[Jplaz]

You should add some pics of the 465 w/o the shroud (when you get the time)
Sub'd.

 

[man from atlantis]

Uber mega anti gravity 128 bit 16 cores reserved.. lol

i think another mighty review gonna come as usual by Staryoshi

 

[Coolio831]

Wow, Nice!

 

[staryoshi]

Benchmarks are up, stay tuned for a bit more... I have an awesome grid-graph and more coming

 

[james8]

very nice. this should increase GTX 465 reputation a bit. i means i've heard a lot of GTX 465 bash ever since 460 is released.

 

[Kand]

The results are just as I would imagine. Good show. Very in-depth observations as well. I notice that you may have swapped the Gpac cou... Oh wait, you noticed that. :3

 

[Rising]

+1

Love these reviews. Keep them coming!

 

[Jeffredo]

Thanks very much star. Excellent and relevant review so far.

 

[Redwoodz]

looks like the core clock wins for the most part.Would be interesting to see the MSI 460 against the MSI 465 lim. edition,as I believe they have the edge when it comes to overclocking the GTX.Good work!

 

[venom55520]

excellent review man. i knew the 465 wasn't the pushover everyone was making it out to be

it costs 20 bucks less, and it's within 5% of the gtx 460 90% of the time after a decent OC.

 

[VW_TDI_02]

I didn't see it in the review so far but you have to factor in the fact that the 460 is limited to 2-way SLI which IMO is by far it's biggest draw back. Also I think it was only the black PCB PNY 465s that unlocked but you should try it. Hmm, what the MSI Gold Edition one...

 

[PROBN4LYFE]

gtx 465 ftw.

marginal OC>massive OC on gtx460

GTX465 Tri Sli> GTX460 Sli

LMAO

 

[venom55520]

Quote:

Originally Posted by PROBN4LYFE gtx 465 ftw. marginal OC>massive OC on gtx460 GTX465 Tri Sli> GTX460 Sli LMAO

hell yea man, i'm gonna get a triple sli setup when these get dirt cheep

 

[staryoshi]

My review is now complete. Please read the third and final section and tell me what you think!

Thanks for reading guys.

Quote:

Originally Posted by james8 very nice. this should increase GTX 465 reputation a bit. i means i've heard a lot of GTX 465 bash ever since 460 is released.

The GTX465 is a solid performer and a great value now that it's price has dropped. While it doesn't offer the same evolutionary benefits of GF104, it's no slouch at all.

(aka power consumption, die size, PCB length, etc)

Quote:

Originally Posted by Kand The results are just as I would imagine. Good show. Very in-depth observations as well. I notice that you may have swapped the Gpac cou... Oh wait, you noticed that. :3

Thanks, I notice you just changed your avatar picture

Quote:

Originally Posted by Redwoodz looks like the core clock wins for the most part.Would be interesting to see the MSI 460 against the MSI 465 lim. edition,as I believe they have the edge when it comes to overclocking the GTX.Good work!

I do not have any experience with the MSI GTX465 that everyone is raving about (they have good reason to!), but I imagine it would do better than a reference GTX465... However it may not overclock quite as well when unlocked (at which point who cares, cause you got a GTX470 upgrade!

)

Quote:

Originally Posted by VW_TDI_02 I didn't see it in the review so far but you have to factor in the fact that the 460 is limited to 2-way SLI which IMO is by far it's biggest draw back. Also I think it was only the black PCB PNY 465s that unlocked but you should try it. Hmm, what the MSI Gold Edition one...

I added a tiny blurb in my last update about 2-way/3-way SLI. In my opinion if you're going for Tri-SLI, you're better off going for the GTX470/480 anyway. Personally I'd rather have SLI GTX 470 than Tri-SLI GTX465... for the sake of power, heat, and real-estate (case-space) efficiency

I picked up the GTX465 on a whim from Best Buy to review it, hoping it had been sitting on a shelf forever and could possibly be the black PCB. I had no such luck, so I didn't bother trying to unlock it. With GTX470s on sale for $290 before MIR and better, I wouldn't bother at this point... that's IMO anyway.

 

[banging34hzs]

nice job

 

[Kand]

Quote:

Originally Posted by staryoshi Thanks, I notice you just changed your avatar picture

Heh. The old one was getting stale, I thought it was time for a new one.

Anyways. The 465 you reviewed with the reference cooler attatched, I'm biased as well to the "hotbox" design. I'd always go for aftermarket/custom heatsinks on videocards, thus my Palit GTX 465 Dual Fan. I really think the heatsinks designed by Nvidia are somewhat inadequate for their videocards. Take the GTX 470 for example, the heatsink is tiny, covers such little space with the fins suffocating due to the density of heatpipes within, the fan barely pushes air through, even with that cutout on the PCB, they should have cut vents on the shroud instead of the PCB.

I think 465s are capable of hitting at least 900Mhz stable at 1.087V which is the usual stock maximum for any 4xx card without a flash of a special bios. Though, it's also probably luck of the draw as well that I'm able to achieve the same clocks as your Overclock on that blue PNY with stock volts. =P

I should give that a try.. This 465 of mine is able to handle 900Mhz with some... visible artifacting on 0.987v...