Because the 4870 has the same VRAM but more memory bandwidth and shader power.
Anandtech showed 4x AA @ 2560x1600 goes to the 4870 when comparing it to a 9800 GTX+. Even against the 9800 GX2, the 4870 still takes the lead in that situation.
You must be looking at a different review, as the 8800GT never beats the 9800 GTX+ in that review. As for other reviews, I've seen plenty that show a significant difference between the 1GB and 512MB 4870s. I guess others have as well in order to justify that $50-100 price premium for an additional 512MB RAM.
Does the Guru3D even state what AA or settings they're using? I've scanned it a few times but even that review shows 1) a significant difference between 4870 1GB and 512MB and 2) the 8800GTX beating the 512MB 4870 at higher resolutions.
You're right, it doesn't beat it, but they're tied. Bottom of the Fallout 3 page @ 1920x1200 with "48 FSAA" (Guessing its a mistype for 8x FSAA)
Original Link
Specific graph
Right, so you've found a situation that's so limited by VRAM that it outweighs all other considerations...now extend that to another comparison being made between the 1GB 9800GTX+ and 512MB 4870.
It's 4x AA, look at page2. My point was that while in the review you posted the 8800gtx beats the 4870 512mb at 1680x1050, in the guru3d review its slower even at 1920x1200 so something's wrong in one of them, and i for one trust guru3d more. My point is that the 512mb 4870 didn't tank in performance until it hit 25x16 so while it does benefit from more ram, it's not killed by having only 512mb until you hit that resolution and as such it should beat a 1gb 9800gtx+
Ya, it beats the 4870 512 at 1680 with 8xAA which makes sense given the additional VRAM requirements of AA. At 1920 with 4xAA it behaves nearly identically to your Guru3D benchmark with the 512MB 4870 edging out the 8800GTX. In any case, I think the point has been clearly made, the 4870's performance isn't so much greater that a difference in VRAM wouldn't juxtapose relative performance in VRAM limited situations (which is exactly what I said in my first post about this....)
I have? If it was "so limited by VRAM that it outweighs all other considerations" why is a 4870 512 MB still beating a 9800 GTX+ 512 MB?
Answer: Because bandwidth matters.
Perhaps with the 4800s, I was referring to the comparison between the 8800gt and 9800GTX+, which is again confirmed when looking at the 9800GTX+ 1GB results.
If you're going off of strictly VRAM amounts and not bandwidth or shader power, then why is the GTX 260 performing better than the 9800 GTX+ 1GB?
Answer: Because bandwidth matters.
I'm only going off VRAM amounts when strictly comparing between the same architecture, obviously once you expand the scope to encompass other features the results will vary. Its clear bandwidth, shaders, rops, tmus, fillrate and anything else aren't enough to compensate in some instances, which is why I brought up the 1GB 9800GTX+ to 512MB 4870 comparison to begin with.
But lets look at some of the examples you've provided. Even in your 4870 to 4850 comparison, it'd be inaccurate to say that difference is solely because of bandwidth, as the difference in performance closely mirrors the difference in clockspeed. If the clocks were the same and you saw that same 20% difference in performance then yes, you could assume only "bandwidth matters".
Same for the 9800GTX+ 1GB to the GTX 260. While its true the 9800GTX+ 1GB may equal or exceed the 260 in numerous areas like VRAM, Texture fillrate, SP performance the GTX 260 still holds the edge in bandwidth, pixel fillrate and ROP/AA performance.
The conclusions are simple, some features or aspects of performance are more important than others and that advancements in one area might not be enough to overcome deficiencies in others.
What is this RV740? Been looking through all the cards at newegg, and I settled to buy a 4830 after I build a new system sometime next week. Better performance? Same power requirements?
the 4870/512 for 150 is just too much value for the cash. it's closer to a gtx260 and competes with the rebranded nvidia 9800gt(x) lines. man this is just the best value out there.
Qbah
Diamond Member
- Oct 18, 2005
- 3,754
- 10
- 81
It's a new mid-range Radeon that's rumored to hit retail in April. 40nm process, 640 shaders, 128-bit bus with GDDR5 memory (the same as a 256bit one with GDDR3). There's a preview somewhere showing that it sits between a HD4830 and HD4850 in terms of performance and is expected to be a sub-100$ card. Instant winner in my eyes.
There's still no name for it, but the code name is RV740.
I'm guessing ATi is going to EOL HD4830 and HD4850 cards. Maybe there will be an entry level card too. And it will be like:
- ?? - entry level
- RV740
- HD4870
And something to compete with the GTX285. There's a 950MHz card rumored, I think it's a HD4890?
So little information... I'm hoping for an instant-hit like the HD48x0 series.
I hate being too single-minded, but I suuuure hope they make some low profile versions of these cards! Any rumours?
TBH, i'm not buying the pc hardware review. Look at this page for example: http://www.pcgameshardware.com...viewed/Reviews/?page=5
We're looking at Call of Duty: World at War, 1680*1050, 4xfsaa. A 8800gt is faster then a HD4850 ? The 8800gtx, which is known to be on par with a HD4850, is beating a HD4870? I dunno man, afaik the HD4850 roughly equaled the 9800GTX+, and the HD4870 more or less equalled the GTX260 216. Maybe new driver updates from nvidia really worked their magic and things drastically changed, maybe the review is off. I'll wait for another 12 hours or so, till the NDA is up ( and we hopefully get a msrp for the GTS250 ).
We're looking at Call of Duty: World at War, 1680*1050, 4xfsaa. A 8800gt is faster then a HD4850 ? The 8800gtx, which is known to be on par with a HD4850, is beating a HD4870? I dunno man, afaik the HD4850 roughly equaled the 9800GTX+, and the HD4870 more or less equalled the GTX260 216. Maybe new driver updates from nvidia really worked their magic and things drastically changed, maybe the review is off. I'll wait for another 12 hours or so, till the NDA is up ( and we hopefully get a msrp for the GTS250 ).
Yet "only going off of VRAM" is exactly what you did when you claimed a 9800 GTX+ 1GB will beat a 4870 512MB in VRAM limited situations.
I'm not saying it won't, but it would be surprising since even at high resolutions with AA in VRAM-demanding games differences in performance are based on factors outside of VRAM amounts like memory bandwidth and GPU architecture.
"My" 4870 to 4850 comparison? I never claimed it to be a 4870 to 4850 comparison. I claimed it to be a 4870 512MB to 9800 GTX+ 512MB comparison at a VRAM-stressing situation.
When comparing the two products you're wanting to (i.e., the 4850 and the 4870 512), yes it would be inaccurate to say the difference is soley because of VRAM.
However, the benchmark is clearly stressing VRAM the most as there is a 63% difference between the 4870 512MB and 4870 1GB.
COD5 consistently favors Nvidia parts, you can confirm this in pretty much any review. If you're looking at that game as an example of an outlier you're only going to find disappointment. But ya at this point it wouldn't hurt to just wait and see.
Yep, when strictly comparing those parts to explain the similarity in performance in that specific benchmark. It wouldn't make any sense to extend that comparison to all resolutions or settings as the 9800GTX+ clearly outperforms the 8800GT in situations VRAM isn't the biggest limiting factor. Just as it wouldn't make sense to compare the 4870 512MB to the 9800GTX+ 1GB in situations where VRAM wasn't the limiting factor. The 4870 is and should be faster in that case.
How would it be surprising? We've seen it makes enough difference in the past with like architectures when comparing 8800GTX to 9800GTX and 4870 512 to the 1GB version. When you look at the 9800GTX+ to the 4870 the difference isn't so insurmountable that a change in VRAM wouldn't be enough to change the relative performance.
Oh right, I was just a step ahead of you. Again, you're comparing between different architectures in that case where the 4800s don't require additional VRAM as much as the G92. But even in the case of your 4870 to 9800GTX+, I wouldn't say bandwidth is the greatest limiting factor again using the 4850 and 1GB 9800GTX+ as examples.
Yep, meant to read bandwidth, which was the inaccurate assumption you made. Again, clearly false as the 4850 scales to clockspeed despite having half the bandwidth and the 9800GTX+ with the same bandwidth outperforms the 512MB 4870 once you increase its frame buffer.
Which just confirms VRAM is still the greatest limiting factor at that particular resolution and setting.
Will this card be the same power requirement about as the 4830? I imagine it would be, but do not want to make assumptions waiting for a card I can not use. I am building a budget system and as long as the power requirements are not a large jump, I should be ok.
AzN
Banned
- Nov 26, 2001
- 4,112
- 2
- 0
pchardware results are usually amplified. Results isn't what you would see in actual game play. I get an average of 50fps in COD5 that's 1920x1080 4xAA with my 8800gts.
512mb of vram is enough for 1920x1200 4xAA. There are 1 or 2 titles that has little more impact like farcry 2 and fallout 3 maxed out. Even than it's less than 20% with ATI cards. Nvidia cards are worse.
Upgrading for future games has been a dumb idea and always is. By the time you need more vram you would need a faster GPU at the resolution and settings you are trying to play at.
512mb of vram is enough for 1920x1200 4xAA. There are 1 or 2 titles that has little more impact like farcry 2 and fallout 3 maxed out. Even than it's less than 20% with ATI cards. Nvidia cards are worse.
Upgrading for future games has been a dumb idea and always is. By the time you need more vram you would need a faster GPU at the resolution and settings you are trying to play at.
AzN
Banned
- Nov 26, 2001
- 4,112
- 2
- 0
http://www.guru3d.com/article/...ics-performance-test/6
It is a little off. 4850 should tie 9800gt at worse. 4870 is barely slower than GTX260 in this game. Chizow loves to only point out these kind of benches that show Nvidia cards in light. With later revisions ATI said they have fixed performance issues with this game. I don't know exactly how much.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 21K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter