How fast will the radeon 5770 be upon release?

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

T2k

Golden Member
Feb 24, 2004
1,664
5
0
5770 is the new affordable 4800 - ergo it's going to be 4850-4870 level, sans memory bandwidth, plus DX11 featureset.
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
The hd5750 due to less shaders and lower clocks may lose to the hd4830 and hd4770

How are 720 shaders less than 640 shaders?

If those slides are right then HD5750 will have the processing power of HD4850 but 15% more bandwidth (although HD4850 is 256 bit it uses GDDR3) .

HD5770 will have the processing power of HD4890....But almost half the bandwidth (basically HD5770 looks like a Cypress cut in half)

 

toyota

Lifer
Apr 15, 2001
12,957
1
0
Originally posted by: LCD123


I didn't think the rumors of 1120sp was accurate. I had said 800-960 and I was correct. With only 800sp and 128 bit memory, there is NO WAY the 5770 will touch a hd4870. I am not even sure it can touch a hd4850 either. I therefore stand by my vote that it can at least match the hd4830, well it may be slightly faster than hd4830 and hd4770, but that's as far as my prediction goes. The price of $159 is because of the dx11 premium and supporting the "latest and greatest" features.

The hd5750 due to less shaders and lower clocks may lose to the hd4830 and hd4770, however it should outperform the hd4750 and hd3870. But for around $119, it's a fairly cheap dx11 card. If you don't need dx11, grab a used hd4830 or hd4850 for cheap!

Don't sell your hd48x0 too cheap or you will be sorry unless you plan to upgrade to hd58x0. Anything less is a sidegrade or downgrade!

what? the 5770 will most certainly be as fast or faster in many cases as the 4870. the only thing it gives up to the 4870 is 50% bandwidth which may not hurt it too bad in most cases. in every other regard the 5770 is as good and actually has faster clocks. the 5770 will easily beat the 4850 as it has way faster specs in EVERY regard. hell the 4770 was only 10% slower than the 4850 and has less shaders , tmu and slower clocks than the 5770.
 

crisium

Platinum Member
Aug 19, 2001
2,643
615
136
Originally posted by: toyota
hell the 4770 was only 10% slower than the 4850 and has less shaders , tmu and slower clocks than the 5770.

That's why we're worried, especially about the 5750. It's just a step above the 4770 really, and probably will barely outperform the 4850. I think it's safe to say the 5750 will be inferior to the 4870. Now the 5770 is just a step above the 5750... so at best it will barely outperform the 4870.

At least there's room for a 5790 in the future so that we actually can get 4890 DX11 performance. There's gonna be a bigger gap between 5850 and 5770 then most people like I think.
 

toyota

Lifer
Apr 15, 2001
12,957
1
0
Originally posted by: Crisium
Originally posted by: toyota
hell the 4770 was only 10% slower than the 4850 and has less shaders , tmu and slower clocks than the 5770.

That's why we're worried, especially about the 5750. It's just a step above the 4770 really, and probably will barely outperform the 4850. I think it's safe to say the 5750 will be inferior to the 4870. Now the 5770 is just a step above the 5750... so at best it will barely outperform the 4870.

At least there's room for a 5790 in the future so that we actually can get 4890 DX11 performance. There's gonna be a bigger gap between 5850 and 5770 then most people like I think.
the gap may not be so bad so just wait and see the actual performance. if it is then I think we will likely see a 5830 in the future just to fill the performance and price gap. IMO they would be much more likely to cut down the 5850 and make 5830 then to add anything to the 5770 and make it a 5790.

 

bunnyfubbles

Lifer
Sep 3, 2001
12,248
3
0
Originally posted by: LCD123
I didn't think the rumors of 1120sp was accurate. I had said 800-960 and I was correct. With only 800sp and 128 bit memory, there is NO WAY the 5770 will touch a hd4870. I am not even sure it can touch a hd4850 either. I therefore stand by my vote that it can at least match the hd4830, well it may be slightly faster than hd4830 and hd4770, but that's as far as my prediction goes. The price of $159 is because of the dx11 premium and supporting the "latest and greatest" features.

The hd5750 due to less shaders and lower clocks may lose to the hd4830 and hd4770, however it should outperform the hd4750 and hd3870. But for around $119, it's a fairly cheap dx11 card. If you don't need dx11, grab a used hd4830 or hd4850 for cheap!

I think you're way off on your "predictions" here. I say it that way because it really doesn't sound like you know what you're talking about, lets start from the bottom up on your predictions and see where it takes us:

========
5770: 850MHz GPU w/ 800SPs 40TMUs 16ROPs, GDDR5 @ 1200MHz & 128bit for 76.8GB/sec

vs.

4830: 575MHz GPU w/ 640SPs 32TMUs 16ROPs, GDDR3 @ 0900MHz & 256bit for 57.6GB/sec
or
4770: 750MHz GPU w/ 640SPs 32TMUs 16ROPs, GDDR5 @ 0800MHz & 128bit for 51.2GB/sec

In every situation (except for ROPs), the 5770 completely outclasses the 4830 (or 4770). This isn't even considering improvements made to increase the efficiencies anywhere. What makes it even worse is that while some 4830/4770s might have 1GB of memory, most will not, whereas all 5770s should have 1GB. What really makes you look extremely foolish is that the 4770 is ATI's pioneer part on the 40nm process, so it could be considered as a cousin of sorts to the 5700s, however both 5700s clearly outclass the 4770 in just about every aspect possible, yet you're spouting off nonsense that the 5770 might not be as fast or barely faster...

Ok, so you've gotten it way wrong thus far, lets see how you do with the 5770 vs. 4850.
========

========
5770: 850MHz GPU w/ 800SPs 40TMUs 16ROPs, GDDR5 @ 1200MHz & 128bit for 76.8GB/sec

vs.

4850: 625MHz GPU w/ 800SPs 40TMUs 16ROPs, GDDR3 @ 0993MHz & 256bit for 63.55GB/sec

Well, the GPUs look like they have the same raw power, however the 5770 has a much faster clock rate on the GPU and its memory bandwidth is equally greater. Again, you're way off. Also, just like with the 4830, we're most likely comparing a 512MB 4850 to a 1GB 5770. Are still unsure as to whether or not it can keep up with the 4850?
========

========
Finally we have 5770 vs. 4870, cards that will be similarly priced...

5770: 850MHz GPU w/ 800SPs 40TMUs 16ROPs, GDDR5 @ 1200MHz & 128bit for 76.8GB/sec

vs.

4870: 750MHz GPU w/ 800SPs 40TMUs 16ROPs, GDDR5 @ 0900MHz & 256bit for 115.2GB/sec

Again, the 5770 has a faster GPU, but less memory bandwidth. Are you still really confident in your claim that the 5770 can't even touch the 4870? I will bet you money the 5770 will be able to match or win in several games at lower resolutions that match up with its price point.
========


Originally posted by: LCD123
Don't sell your hd48x0 too cheap or you will be sorry unless you plan to upgrade to hd58x0. Anything less is a sidegrade or downgrade!

This is the only thing I could agree with you at all on, however as I've shown these 5700s will clearly be upgrades for anyone who doesn't have a 4870 or better. I would still agree with you though, if I'm coming from a 4850, I wouldn't want to settle for less than a 5850.
 

bunnyfubbles

Lifer
Sep 3, 2001
12,248
3
0
Yeah, the higher specs were aligned with the assumption of a $149-199 price range.

The problem I can see now is that there is a pretty significant void between 5770 @ $159 and 5850 @ $259...

I thought the 5600s would have taken up an $80-135ish range with performance just an edge below what the 5700s will offer with the 5700s occupying that higher price range with a slightly higher performance.
 

happy medium

Lifer
Jun 8, 2003
14,387
480
126
Originally posted by: bunnyfubbles
Yeah, the higher specs were aligned with the assumption of a $149-199 price range.

The problem I can see now is that there is a pretty significant void between 5770 @ $159 and 5850 @ $259...

I thought the 5600s would have taken up an $80-135ish range with performance just an edge below what the 5700s will offer with the 5700s occupying that higher price range with a slightly higher performance.

yea I can smell a 5830 comming soon. ~200 and besting 4890 performance
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
Originally posted by: happy medium
Originally posted by: bunnyfubbles
Yeah, the higher specs were aligned with the assumption of a $149-199 price range.

The problem I can see now is that there is a pretty significant void between 5770 @ $159 and 5850 @ $259...

I thought the 5600s would have taken up an $80-135ish range with performance just an edge below what the 5700s will offer with the 5700s occupying that higher price range with a slightly higher performance.

yea I can smell a 5830 comming soon. ~200 and besting 4890 performance

Yep a 5830 with 256bit DDR5 would be a good chip.

Otherwise I can't wait to see the reviews on HD5770/HD5750 and how they compare to HD4870/HD4890. My guess is at these levels bandwidth will be a limiting factor making a "Voltage Tweak" HD5750 the best bargain.
 

LCD123

Member
Sep 29, 2009
90
0
0
bunnyfubbles,

I think you're way off on your "predictions" here. I say it that way because it really doesn't sound like you know what you're talking about, lets start from the bottom up on your predictions and see where it takes us: 5770: 850MHz GPU w/ 800SPs 40TMUs 16ROPs, GDDR5 @ 1200MHz & 128bit for 76.8GB/sec vs. 4830: 575MHz GPU w/ 640SPs 32TMUs 16ROPs, GDDR3 @ 0900MHz & 256bit for 57.6GB/sec or 4770: 750MHz GPU w/ 640SPs 32TMUs 16ROPs, GDDR5 @ 0800MHz & 128bit for 51.2GB/sec In every situation (except for ROPs), the 5770 completely outclasses the 4830 (or 4770). This isn't even considering improvements made to increase the efficiencies anywhere. What makes it even worse is that while some 4830/4770s might have 1GB of memory, most will not, whereas all 5770s should have 1GB. What really makes you look extremely foolish is that the 4770 is ATI's pioneer part on the 40nm process, so it could be considered as a cousin of sorts to the 5700s, however both 5700s clearly outclass the 4770 in just about every aspect possible, yet you're spouting off nonsense that the 5770 might not be as fast or barely faster... Ok, so you've gotten it way wrong thus far, lets see how you do with the 5770 vs. 4850.

For all your quotes, let's assume ATI will clock the hd5770 this high. Youd be correct that the hd5770 would own the above two.

5770: 850MHz GPU w/ 800SPs 40TMUs 16ROPs, GDDR5 @ 1200MHz & 128bit for 76.8GB/sec vs. 4850: 625MHz GPU w/ 800SPs 40TMUs 16ROPs, GDDR3 @ 0993MHz & 256bit for 63.55GB/sec Well, the GPUs look like they have the same raw power, however the 5770 has a much faster clock rate on the GPU and its memory bandwidth is equally greater. Again, you're way off. Also, just like with the 4830, we're most likely comparing a 512MB 4850 to a 1GB 5770. Are still unsure as to whether or not it can keep up with the 4850?

36% faster clocks and 21% more bandwith than the hd4850, if those specs manifast themself, then yes performance will be at least 21% higher, possibly 30% faster than hd4850!

Finally we have 5770 vs. 4870, cards that will be similarly priced... 5770: 850MHz GPU w/ 800SPs 40TMUs 16ROPs, GDDR5 @ 1200MHz & 128bit for 76.8GB/sec vs. 4870: 750MHz GPU w/ 800SPs 40TMUs 16ROPs, GDDR5 @ 0900MHz & 256bit for 115.2GB/sec Again, the 5770 has a faster GPU, but less memory bandwidth. Are you still really confident in your claim that the 5770 can't even touch the 4870? I will bet you money the 5770 will be able to match or win in several games at lower resolutions that match up with its price point.

The hd4870 has a whopping 50% more memory bandwith vs. the hd5770 having only 13% faster core. The hd4890 matches the hd5770 in core clocks but has over 50% more bandwith, that we agree the hd5770 won't come close. For the hd4870 being ahead, let's do the math on hd4770 vs. hd4830 to explain why:


hd4770 has 30% faster core but 11% less bandwith. Therefore it still comes out about 10% ahead of hd4830.

But the hd5770 has only 13% faster core but 33% less bandwith. Therefore it will be 20% behind a hd4870.

Again, the 5770 has a faster GPU, but less memory bandwidth. Are you still really confident in your claim that the 5770 can't even touch the 4870? I will bet you money the 5770 will be able to match or win in several games at lower resolutions that match up with its price point.

I don't care about 800x600, let's see how it does at 1360x768(high definition) and especially at 1920x1080(FHD)

Again, ill agree that the hd5770 should outperform the hd4850 at those clocks the hd5770 is claimed to be at. However I still stand by the fact it can't touch 50% more bandwith of the hd4870, a 13% increase in core clocks doesn't come close. Even a 30% increase in core clock of the hd4770 only put it 10% ahead of the hd4830 due to it having less memory bandwith which is really the limiting factor here.
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
Originally posted by: LCD123
less memory bandwith which is really the limiting factor here.

The reviews will be interesting.

Even though Juniper is essentially half of Cypress.....there seems to be a absolute bottom limit as far as games tolerating a certain amount of bandwidth.

 

cbn

Lifer
Mar 27, 2009
12,968
221
106
Sorry this post was a duplicate of the one above. DELETED.
 

happy medium

Lifer
Jun 8, 2003
14,387
480
126
If the 5770 had the same exact core clocks, memory bandwidth, and amount of sp's as the 4870 ,would it perform better? Is the architexture better per clock? Something like a Athlon x2 vs Core 2 duo @ 3.2? The Core 2 duo is faster .

 

cbn

Lifer
Mar 27, 2009
12,968
221
106
Originally posted by: LCD123



The hd4870 has a whopping 50% more memory bandwith vs. the hd5770 having only 13% faster core. The hd4890 matches the hd5770 in core clocks but has over 50% more bandwith, that we agree the hd5770 won't come close.

Good point. I wonder if we will be seeing a refresh of HD5770 in the future using 7 Gbps GDDR5? (especially if clocking ability of 40nm silicon improves even more as it matures)

Switching from 5 Gpbs DDR5 to 7 Gpbs DDR5 will gain 40% more bandwidth. Hopefully something like will be used if 950 Mhz stock clocks start becoming possible.
 

happy medium

Lifer
Jun 8, 2003
14,387
480
126
Looks like they do have a 5830 planned.?

"The HD 5870, 5850 and 5830 series will be the top end with 1,600 shader units, or almost double what the current top end HD 4000 series have. The HD 5770, 5750 and 5730 series will be the performance range, the HD 5650 and 5600 series will be the mainstream parts, and the HD 5470, 5450 and 5430 series will be for the entry level market."

http://www.theinquirer.net/inq...introduces-radeon-gpus
 

crisium

Platinum Member
Aug 19, 2001
2,643
615
136
One thing worth mentioning is the 4670 vs 3850. The cards are equal in performance. Same shaders, but 4670 has 100MHz higher clock and 172MHz higher memory on a 128-bit bus. 4670 also has half of the ROPs! 5770 vs 4870: same shaders, but 4770 has 100MHz higher clock and 300MHz higher memory on a 128-bit bus. It has the same ROPs - not any less!

Now obviously the 4000 series had a lot of improvements over the 3000, but if you do base it off of what happened then, then the 5770 cannot be slower than the 4870. If anything it would be a little faster because it is better off vs the 4870 than the 4670 was to the 3850. But I'm not sure if it's a relevant comparison. Thoughts?
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
Originally posted by: happy medium
If the 5770 had the same exact core clocks, memory bandwidth, and amount of sp's as the 4870 ,would it perform better? Is the architexture better per clock? Something like a Athlon x2 vs Core 2 duo @ 3.2? The Core 2 duo is faster .

People say the HD5xxx has a better memory controller.

This in combination with the new 5 Gbps GDDR5 probably helps the situation a bit.



 

bunnyfubbles

Lifer
Sep 3, 2001
12,248
3
0
Originally posted by: LCD123
The hd4870 has a whopping 50% more memory bandwith vs. the hd5770 having only 13% faster core. The hd4890 matches the hd5770 in core clocks but has over 50% more bandwith, that we agree the hd5770 won't come close. For the hd4870 being ahead, let's do the math on hd4770 vs. hd4830 to explain why:


hd4770 has 30% faster core but 11% less bandwith. Therefore it still comes out about 10% ahead of hd4830.

But the hd5770 has only 13% faster core but 33% less bandwith. Therefore it will be 20% behind a hd4870.

Again, the 5770 has a faster GPU, but less memory bandwidth. Are you still really confident in your claim that the 5770 can't even touch the 4870? I will bet you money the 5770 will be able to match or win in several games at lower resolutions that match up with its price point.

I don't care about 800x600, let's see how it does at 1360x768(high definition) and especially at 1920x1080(FHD)

Again, ill agree that the hd5770 should outperform the hd4850 at those clocks the hd5770 is claimed to be at. However I still stand by the fact it can't touch 50% more bandwith of the hd4870, a 13% increase in core clocks doesn't come close. Even a 30% increase in core clock of the hd4770 only put it 10% ahead of the hd4830 due to it having less memory bandwith which is really the limiting factor here.

Except you still don't show that you understand what is going on here. Memory bandwidth doesn't directly correlate to increased performance, it is only a bottle neck if the GPU needs more than what is provided. You could give a GPU 1750MHz GDDR5 on a 512bit bus for an insane 448GB/sec of memory bandwidth, but if the GPU only needs 60GB/sec of bandwidth then you have a surplus of 388GB/sec, it won't be making the card any faster. The perfect example here is the 4870 vs. 4850, where its quite obvious that 50% more memory bandwidth does not equate to 50% more performance. Your super scientific method of calculating performance doesn't work there now does it.

Also, 1360x768 is not high definition, that's a tiny resolution, barely 1 million pixels, I'd definitely bet you big money that a 5770 would be faster than a 4870 in just about every game at that resolution. 1440x900 is a far more common resolution for gamers and is just about the lowest resolution we'd ever see benched, and even that resolution is way too low to consider a memory bandwidth bottleneck. What we really need to see is how the 5770 does on 1680x1050 (not too worried) and 1920x1080 (given the recent influx of affordable panels at this resolution). However even those resolutions haven't posed too much problem for cards.
 

edplayer

Platinum Member
Sep 13, 2002
2,186
0
0
My vote is for average scores higher than that of the 4890

I also vote: This thread sucks.

 

Remi fa Sola

Junior Member
Sep 17, 2009
3
0
0
Although I didn't vote (can't be bothered), you guys would proberbly interested in this link:

http://www.xbitlabs.com/articl...-hd4850-1024mb-gs.html

It has the HD4850 (1 gigabyte videoram) overclocked to 750 Mhz, thus equal to the HD4870.
Ram speed is at 1200Mhz, resulting in 76.8GBps (well, isnt that convienent?)

Altough the main clock is lower than the likely 850mhz, it does give insight in core clock scaling and memory speed dependency of the design

This results in a 0~10% performance difference in average framerates in most games. (4850 @750mhz vs 4870 @ stock)
(noticeable exception here is Call of Duty 4: almost 18%)

Overal, I think its pretty important to note that the framerates lie closer to the 4870 than they do to the 4850 @ stock, often by a significant margin.

The 4850 GS @700mhz shows pretty interesting results too.
Its memory is not overclocked (if im not mistaken, that is) and the card gives a definite performance boost over the standard version.
So performance still scales with an increased core clock...

Given these two "facts" and the 850mhz core clock rumor, I think its safe to extrapolate the HD5770 ending up as fast or faster (on avarage) than the HD4870.

But, there are a lot of uncertainties. Memory controller for one.
Another factor is the latency of the memory, which as far as i know does give an advantage.

Now, we must also not forget the secondary benifits of a the new architecture:
A massive reduction of heat, thus noise (40nm) and the price can (over time) drop even more!

Lastly, I dont want to be misquoted, so I will say this:
Do i think (when discussing a 5770) an increase in bandwidth will give a good increase in performance? - YES, without doubt.
Do i think the lack of more bandwidth will make the card a failure, or not competitve? no, not in the current climate, but we must first see what fermi brings to the table...
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
Originally posted by: Remi fa Sola
Although I didn't vote (can't be bothered), you guys would proberbly interested in this link:

http://www.xbitlabs.com/articl...-hd4850-1024mb-gs.html

It has the HD4850 (1 gigabyte videoram) overclocked to 750 Mhz, thus equal to the HD4870.
Ram speed is at 1200Mhz, resulting in 76.8GBps (well, isnt that convienent?)

Altough the main clock is lower than the likely 850mhz, it does give insight in core clock scaling and memory speed dependency of the design

This results in a 0~10% performance difference in average framerates in most games. (4850 @750mhz vs 4870 @ stock)
(noticeable exception here is Call of Duty 4: almost 18%)

Overal, I think its pretty important to note that the framerates lie closer to the 4870 than they do to the 4850 @ stock, often by a significant margin.

The 4850 GS @700mhz shows pretty interesting results too.
Its memory is not overclocked (if im not mistaken, that is) and the card gives a definite performance boost over the standard version.
So performance still scales with an increased core clock...

Given these two "facts" and the 850mhz core clock rumor, I think its safe to extrapolate the HD5770 ending up as fast or faster (on avarage) than the HD4870.

But, there are a lot of uncertainties. Memory controller for one.
Another factor is the latency of the memory, which as far as i know does give an advantage.

Now, we must also not forget the secondary benifits of a the new architecture:
A massive reduction of heat, thus noise (40nm) and the price can (over time) drop even more!

Lastly, I dont want to be misquoted, so I will say this:
Do i think (when discussing a 5770) an increase in bandwidth will give a good increase in performance? - YES, without doubt.
Do i think the lack of more bandwidth will make the card a failure, or not competitve? no, not in the current climate, but we must first see what fermi brings to the table...

That is a good article.

In Fallout 3 there is a large difference in performance between HD4850 1 Gb clocked at 750 mhz and the HD4870 1 GB.....but then in STALKER frame rates are much closer together.

We can probably expect the same with HD5770 and HD4890.
 

LCD123

Member
Sep 29, 2009
90
0
0
Originally posted by: Just learning
Originally posted by: LCD123
less memory bandwith which is really the limiting factor here.

The reviews will be interesting.

Even though Juniper is essentially half of Cypress.....there seems to be a absolute bottom limit as far as games tolerating a certain amount of bandwidth.

The only way a hd5770 can beat the hd4870 is in games that don't care for bandwith or in very low resolutions. The hd5770 will never match/equal the hd4890 because both cards have identical core clocks and shader processors but the hd4890 has way more bandwith. I think those voting that the hd5770 as being equal to the hd4890 were assuming the rumors of 1120 shader processors would be true but it now looks like there will be only 800 shader processors. I did not think it would have more than 1000.

Originally posted by: Just learning
Originally posted by: LCD123



The hd4870 has a whopping 50% more memory bandwith vs. the hd5770 having only 13% faster core. The hd4890 matches the hd5770 in core clocks but has over 50% more bandwith, that we agree the hd5770 won't come close.

Good point. I wonder if we will be seeing a refresh of HD5770 in the future using 7 Gbps GDDR5? (especially if clocking ability of 40nm silicon improves even more as it matures)

Switching from 5 Gpbs DDR5 to 7 Gpbs DDR5 will gain 40% more bandwidth. Hopefully something like will be used if 950 Mhz stock clocks start becoming possible.

We may see the 7g ram on the hd5890. Too expensive to use on a sub $200 card, it will make it's way in a card costing at least $400.

Originally posted by: happy medium
Looks like they do have a 5830 planned.?

"The HD 5870, 5850 and 5830 series will be the top end with 1,600 shader units, or almost double what the current top end HD 4000 series have. The HD 5770, 5750 and 5730 series will be the performance range, the HD 5650 and 5600 series will be the mainstream parts, and the HD 5470, 5450 and 5430 series will be for the entry level market."

http://www.theinquirer.net/inq...introduces-radeon-gpus

That's alot of different cards for every price range. Would be interesting to see how the 56x0 perform. The 54x0 probably will be too slow for modern games above 800x600, those will find their way in cheap sub $500 PCs.

Originally posted by: bunnyfubbles

Except you still don't show that you understand what is going on here. Memory bandwidth doesn't directly correlate to increased performance, it is only a bottle neck if the GPU needs more than what is provided. You could give a GPU 1750MHz GDDR5 on a 512bit bus for an insane 448GB/sec of memory bandwidth, but if the GPU only needs 60GB/sec of bandwidth then you have a surplus of 388GB/sec, it won't be making the card any faster. The perfect example here is the 4870 vs. 4850, where its quite obvious that 50% more memory bandwidth does not equate to 50% more performance. Your super scientific method of calculating performance doesn't work there now does it.

Also, 1360x768 is not high definition, that's a tiny resolution, barely 1 million pixels, I'd definitely bet you big money that a 5770 would be faster than a 4870 in just about every game at that resolution. 1440x900 is a far more common resolution for gamers and is just about the lowest resolution we'd ever see benched, and even that resolution is way too low to consider a memory bandwidth bottleneck. What we really need to see is how the 5770 does on 1680x1050 (not too worried) and 1920x1080 (given the recent influx of affordable panels at this resolution). However even those resolutions haven't posed too much problem for cards.

The GPU will need more bandwith than the hd5770 provides at anything above 800x600 and for all modern games. If there was no need, ATI would have gone 128 bit for the high end cards, but the fact ATI has gone 256 bit ram since the radeon 9700s tells me that there is a need for bandwith! Nvidia is actually going 384 bit for their new cards and they have gone above 256 bit on their current high end cards. A 50% increase in memory bandwith doesn't always give 50% more performance unless the game requires every bit of bandwith then the bandwith will be there!

High definition is 1280x720, full high definition is 1920x1080. My 32" LCD is considered to be high definition with a native resolution of 1360x768. A tiny resolution would be 800x600 or lower which is what I had on my CRT monitors ive owned before I finally went LCD in late 2008. There's plenty of LCDs(small monitors and medium TVs) with 1360x768 so this will be the lowest common resolution except for those still on a CRT or those running non native resolutions on LCDs. 1920x1080 is a common resolution nowdays and can be found on LCD monitors in the 23" range and on LCD TVs 32" and up.
 

bunnyfubbles

Lifer
Sep 3, 2001
12,248
3
0
Originally posted by: LCD123
The GPU will need more bandwith than the hd5770 provides at anything above 800x600 and for all modern games. If there was no need, ATI would have gone 128 bit for the high end cards, but the fact ATI has gone 256 bit ram since the radeon 9700s tells me that there is a need for bandwith! Nvidia is actually going 384 bit for their new cards and they have gone above 256 bit on their current high end cards. A 50% increase in memory bandwith doesn't always give 50% more performance unless the game requires every bit of bandwith then the bandwith will be there!

High definition is 1280x720, full high definition is 1920x1080. My 32" LCD is considered to be high definition with a native resolution of 1360x768. A tiny resolution would be 800x600 or lower which is what I had on my CRT monitors ive owned before I finally went LCD in late 2008. There's plenty of LCDs(small monitors and medium TVs) with 1360x768 so this will be the lowest common resolution except for those still on a CRT or those running non native resolutions on LCDs. 1920x1080 is a common resolution nowdays and can be found on LCD monitors in the 23" range and on LCD TVs 32" and up.

ATI went from a 512bit bus with their 2900XT to a 256bit bus with the 3870 and saw virtually no drop off in bandwidth intensive situations.

Then there was the G80 -> G92 transition where nVidia dropped from a 384bit bus to a 256bit bus, again with the 256bit bus being more than enough. And now they're also dropping from 512bit to 384bit from GT200 to Fermi

Also, you can delude yourself about your TV resolution all you want, anything lower than 1680x1050 is a tiny resolution per PC gaming standards. Like I said before, 1440x900 or 1280x1024 are both much larger than your tiny, non HDTV resolution, and both of those resolutions are as small as you'll see tested on most review sites.