• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

HD 4850 Shootout! 512mb vs. 1gb

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

sno.lcn

Senior2 Member
Joined
Nov 21, 2005
Location
Atlanta, GA, USA
HD 4850 Shootout: 512mb VS. 1gb
How much difference does the extra 512mb make?

Sure, I've seen reviews comparing the new 1gb HD 4850 and HD 4870 models
to their 512mb counterparts, but I wanted to see what the additional 512mb
did for myself. I would like to thank Gigabyte US for letting me test these cards.



First, I'd like to introduce the cards I'll be testing.
The 512mb is a Gigabyte model, from when the HD 4850 first came out.
My review of this card are here. Here's the 1gb model it'll be going up against :)


Model-specific specs
Model: GV-R485MC-1GH
Stock clocks: 625mhz core, 1920mhz mem
Memory size: 1gb
Cooling: Gigabyte's Multi-Core™ Cooling Technology


The box hasn't changed, but to include mention of the upgraded heatsink on this card.

01theboxkv4.jpg



As we all know, the reference cooler on all of the original HD 4850's was poorly suited for overclocking.
This massive heatpipe cooler is the reason I asked for this card. It's meant to be passive,
so the heatsink is bigger, with much more surface area/plenty of fins.
Think what this passive cooler could do with a little airflow.
The only obvious issue is the lack of memory and pwm cooling.
As well as with Gigabyte's other GPU heatsink, there's little room for memory cooling.
I rarely use ram sinks, but when I'm dumping on the clocks, I usually have a high-cfm
fan throwing enough air across the chips to keep them cool. I'm not adding pwm sinks
for the shootout since it didn't come that way, but I will be for serious benching later.
Plus, looking at the pwm layout on this card, I'll end up cutting a heatsink down to custom sizes for them.
I'll add some pictures when I get them on.

02frontcardag5.jpg


04heatsinkbackgv4.jpg




The back. Everything's been moved around! That means no volt mod out
of the box like I was planning. It'll come soon enough though.

03backym8.jpg




Sitting in a Gigabyte EP45-UD3P (still reviewing the board). They match perfectly.

05inmobooj8.jpg




To briefly test the cooler out, I set the core clock to 750mhz, and ran 3DMark 03
(the first 4 tests) and left rivatuner open to monitor core temps.
I hovered the mouse where the temps were the highest for the screenshot.
In case you can't tell, the one on the left is with no fan, the one on the right
is with a low-cfm 80mm fan resting against the heatsink. I haven't tested
for max clocks yet, but will add them to this post when I do.
Card clocks top out at 780/1200 on stock voltage, which is better than the 512mb did.

7501140loadnofankr9.jpg
7501180loadlowcfmsd4.jpg
 
Last edited:
The cards

101comparefrontrz1.jpg
102comparebackiy6.jpg



Testing configuration
CPU: Intel e8600 @ 4.5ghz (8.5 x 530)
Mobo: Gigabyte EP45T-Extreme
RAM: OCZ Platinum DDR3 @ 707mhz, 5-5-5-15 2T
CPU Cooling: TR Ultra-120 Original
GPU Cooling for 512mb: AC Accelero S1
GPU Cooling for 1gb: Gigabyte Multi-Core Cooling
PSU: Gigabyte ODIN Pro 1200w
HDD: 36gb WD Raptor
Card clocks will be 640/1000 for the first round of testing, then 740/1170 for the second round.


To obtain a comparison as accurate as possible, I'm using very specific and identical settings and testing methods.
Core and mem clocks on both cards will be identical to each other. I'll be doing testing at close to stock, and at overclocked speeds.
Each card will have a brand new OS install, and all software and drivers installed identically, even in the same order.



Benchmarks

Unigine Tropics 3D, 1680x1050, DX10
Aquamark
3DMark 01
3DMark 03
3DMark 05
3DMark 06
3DMark Vantage Performance
Crysis High, 1680x1050, DX9, no AA
Crysis High, 2048x1536, DX9, 16xAA
Crysis High, 1900x1200, DX10, 16xAA
Crysis Very High, 1900x1200, DX10, no AA
Crysis Very High, 1900x1200, DX10, 16xAA
Devil May Cry bench: DX9, 1280x960, 8x AA, all other settings on highest


If you want to see any other benchmarks, let me know and I'll do my best to get them added.




512mb scores in red. 1gb scores in blue.


640/1000 Testing

Tropics bench
318 - 12.6 fps
Screenshot
547 - 21.7 fps
Screenshot

Aquamark
253,628
Screenshot
251,339
Screenshot

3DMark 01
62,316
Screenshot
62,015
Screenshot

3DMark03

40,120
Screenshot
44,718
Screenshot

3DMark05

24,265
Screenshot
24,320
Screenshot

3DMark06

13,510
Screenshot
13,642
Screenshot

3DMark Vantage
P7316
Screenshot
P7294
Screenshot

Crysis High 1680x1050, DX9, no AA
36.405 fps
Screenshot
36.52 fps
Screenshot

Crysis High, 2048x1536, DX9, 16xAA
25.62 fps
Screenshot
25.465 fps
Screenshot

Crysis High, 1900x1200, DX10, 16xAA
23.185 fps
Screenshot
23.87 fps
Screenshot

Crysis Very High, 1900x1200, DX10, no AA
15.825 fps
Screenshot
15.77 fps
Screenshot

Crysis Very High, 1900x1200, DX10, 16xAA
12.52 fps
Screenshot
12.865 fps
Screenshot

Devil May Cry
512mb

1gb



740/1170 Testing

Tropics bench
359 - 14.2 fps
Screenshot
634 - 25.2 fps
Screenshot

Aquamark
260,768
Screenshot
261,172
Screenshot

3DMark 01
64,017
Screenshot
63,814
Screenshot

3DMark03
51,366
Screenshot
51,315
Screenshot

3DMark05
26,187
Screenshot
26,207
Screenshot

3DMark06
15,212
Screenshot
15,387
Screenshot

3DMark Vantage
P8275
Screenshot
P8298
Screenshot

Crysis High 1680x1050, DX9, no AA
41.495 fps

Screenshot
41.77 fps
Screenshot

Crysis High, 2048x1536, DX9, 16xAA
29.61 fps
Screenshot
29.565 fps
Screenshot

Crysis High, 1900x1200, DX10, 16xAA
26.22 fps
Screenshot
27.705 fps
Screenshot

Crysis Very High, 1900x1200, DX10, no AA
18.095 fps
Screenshot
18.17 fps
Screenshot

Crysis Very High, 1900x1200, DX10, 16xAA
14.325 fps
Screenshot
14.935 fps
Screenshot

Devil May Cry
512mb

1gb


 
Last edited:
Got all the initial benches done. Definitely not the outcome I expected. Even though the 3DMarks and Aquamark are the only ones I care about, I'll grab a few more benching apps and keep going.
 
Wow the 1GB is actually slower in several cases :confused: I wonder if the timings on the RAM have anything to do with that? The denser chips on the 1GB have slower timings?
 
It should technically be slower in everything where the VRAM used is less than 500MB due to the slower timing.

Once you hit crazy resolutions though will be where it pulls away.
 
so basically for most, the card is not powerful enough to really use the 1G of ram was was expected, even the 4870 showed little gain from 512 to 1G even at higher res.
 
I got some more scores up. I hate Crysis bench with a passion now :beer:


I tried maxing out the resolution, and both cards choked on it. I might try it with crossfire later though.


I'll do some more testing tomorrow :)
 
What's this? In (H)ardOCP, they said the 1gb was smoother. Can you confirm this by playing games please?

Devil May Cry
512mb

1gb

In (H)ardOCP, they said the 1gb was smoother. Can you confirm this by playing games please?
 
Difference seems much less than I thought. Even at 1900 or 2k+ res in Crysis DX10 with 16xAA there's a 1.5 fps gain at best.

For more benchmarks I'll recommend Crysis Warhead Bench Tool, and run Avalanche or Frost benches on it (Not flythrough). Those might give more accurate results as they are recorded while playing in-game, rather than being a flythrough which the Crysis Benchmark_GPU is. Some review sites have said that the difference between 512mb and 1gb is more noticable in-game, especially when going through levels/switching zones/etc, it's more seamless.
 
Just got done testing with Tropics 3D benchmark, and the 1gb slaughtered the 512mb card.


I'll load up some games right after I run a few more benches guys.
 
Back