• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

HD5770 and HD5750 predictions thread

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Originally posted by: StinkyPinky
So the 5770 will be slower than the 4890?

If so, there's no point upgrading from my 8800gt. Guess i'll have to save some cash for the 5850.
Click to expand...

I'd guess they'll be about equal. They will be launched in two days so I guess we'll see. The 5770 will have less memory bandwidth but more overall processing power. At 1920 x 1200 and below I would venture that it will beat the 4890 and higher resolutions will find them close to equal.
 
Originally posted by: Stoneburner
Originally posted by: StinkyPinky
So the 5770 will be slower than the 4890?

If so, there's no point upgrading from my 8800gt. Guess i'll have to save some cash for the 5850.
Click to expand...

I'd guess they'll be about equal. They will be launched in two days so I guess we'll see. The 5770 will have less memory bandwidth but more overall processing power. At 1920 x 1200 and below I would venture that it will beat the 4890 and higher resolutions will find them close to equal.
Click to expand...

I game at 1680 by 1050 so I'm hoping you're right!
 
Ive been doing math in my head. Wouldn't this give the hd57xx a memory bandwith of about 73GB/s? As for a core clock of 800MHz with 1120sp, this represents a 32% increase in core power vs. hd4890 but the hd4890 has 71% more memory bandwith! The hd57xx does have more memory bandwith than the hd4850(because of gddr3)

I checked hd4770 and it has a little less memory bandwith than the hd4830 but the core is almost as powerful as the hd4850. Thanks to gddr5 memory, the hd4770 is faster than I thought. Checking a review, it appears to perform between a hd4830 and hd4850. I guess the hd57xx could perform between a hd4850 and hd4870. Any chance ATI might opt for slower gddr3 memory instead of gddr5? I had based the hd5770's performance on it using gddr3 memory, but if they go with gddr5 then yea itll definately beat the hd4830 and possibly the hd4850!

wow the hd5770 looks like the hd5850 and hd5870 except it's shorter!
 
thank you model3, is that the case with hynix's 2.5ghz rating? it equals 5ghz? so the stock speed of that hynix ram is also 1250mhz? Thank you

I was so mad, why call it 2500mhz or 5000mhz ram when it runs at 1250mhz, too much confusion.
 
Originally posted by: Tempered81
thank you model3, is that the case with hynix's 2.5ghz rating? it equals 5ghz? so the stock speed of that hynix ram is also 1250mhz? Thank you

I was so mad, why call it 2500mhz or 5000mhz ram when it runs at 1250mhz, too much confusion.
Click to expand...

See the Hynix link below.

http://www.hynix.com/inc/pdfDo...ook_GraphicsMemory.pdf


About memory speed:

The memory may be certified for 2.5GHz, but i guess in this early state of GDDR5 it will be difficult for ATI to achieve the certified speed in 5770 SKU.
(the memory of 5870 is also certified at 5Gbps but ATI managed to clock them at 1,2GHz (2,4GHz effectively).

Let's hope that ATI will manage somehow to clock them at their certified speed.


 
well, here's what it looks like so far: 1300mhz ddr5 memory OC on 5870, 5850, 5770, and 5750... that is sweet. samsung .4ns on the 5800 and hynix .4ns on the 5700.
 
Originally posted by: Just learning
Model3,

What do you think is a greater issue with HD57xx: 16 ROPs or Memory bandwidth?
Click to expand...

both are issues.

It really depends on the game engine.

a 750MHz 4870 1GB is around +33% (4/3) (+20% up to +50%) faster than a 625MHz 4850 1GB. (1920X1200 4AA 16AF)
So the extra bandwidth (+80%) it brings an 0 up to +25% performance increase per MHz (core).

http://www.bit-tech.net/hardwa...h-radeon-hd-4850-1gb/3
http://www.hardwarecanucks.com...passive-review-10.html
http://www.bjorn3d.com/read.php?cID=1527&pageID=6566


5750 (1120SPs) will probably have a better memory controller (judging from 58XX series) than 48XX series.
And also it has +15% (according to Fudzilla) memory bandwidth (but it is GDDR5...) in relation with a 4850 1GB.

I suspect the 5750 (725MHz) will be +20% faster than a 4850 1GB in the worst case and +50% faster in the best case.

So it will be close to 4870 1GB (it will lose in some games that the bandwidth is an issue and will win in some other games that it is not, it depends also from the resolution and the AA & AF levels)

 
Originally posted by: MODEL3

I suspect the 5750 (725MHz) will be +20% faster than a 4850 1GB in the worst case and +50% faster in the best case.

So it will be close to 4870 1GB (it will lose in some games that the bandwidth is an issue and will win in some other games that it is not, it depends also from the resolution and the AA & AF levels)
Click to expand...

Well this is not a bad strategy then.

If ATI is aiming this card at the lower resolution mainstream market it can afford to give up some memory bus space for more stream processors on the die.

As long as higher resolutions aren't used the trade-off should result in higher performance compared to an identical die size GPU.

 
The problem here is the pricing is off or hopefully they at least come with immediate rebates like the 4850s if prices remain $150 and $200. This is assuming the 128bit buses and 16ROPs and also assuming these limitations hold the cards back. I still think these cards need to be able to push 1920x1080 with some AA on most games (especially a $200 5770). The 5770 really does seem a bit bizarre though, clock-rate alone can't justify the price difference. If these really are the specs, I'm now extremely curious as to how they'll perform as it seems like a bold move not to give the 5770 a higher memory bandwidth, its entirely possible an improved memory controller could be enough.


Originally posted by: Tempered81
thank you model3, is that the case with hynix's 2.5ghz rating? it equals 5ghz? so the stock speed of that hynix ram is also 1250mhz? Thank you

I was so mad, why call it 2500mhz or 5000mhz ram when it runs at 1250mhz, too much confusion.
Click to expand...

because you're confusing Gbps with GHz. 5Gbps = 2.5GHz, but when we account for DDR, its 1250MHz



Originally posted by: MODEL3
Originally posted by: Tempered81
thank you model3, is that the case with hynix's 2.5ghz rating? it equals 5ghz? so the stock speed of that hynix ram is also 1250mhz? Thank you

I was so mad, why call it 2500mhz or 5000mhz ram when it runs at 1250mhz, too much confusion.
Click to expand...

See the Hynix link below.

http://www.hynix.com/inc/pdfDo...ook_GraphicsMemory.pdf


About memory speed:

The memory may be certified for 2.5GHz, but i guess in this early state of GDDR5 it will be difficult for ATI to achieve the certified speed in 5770 SKU.
(the memory of 5870 is also certified at 5Gbps but ATI managed to clock them at 1,2GHz (2,4GHz effectively).

Let's hope that ATI will manage somehow to clock them at their certified speed.
Click to expand...

I'm pretty sure they can clock to their rated speed (and beyond, considering the claims of 1300MHz overclocks). The difference here is that instead of pushing them to the very brim, they're kept back a few % for stability purposes as well as to differentiate classes. Of course the 5850 being clocked back to 1000MHz (or to a lesser extent, a 5700 with mem @ 1150MHz) when it has chips certified to be capable of 1250MHz just like the 5870's 1200MHz chips...yeah, its mega win for the prudent overclockers 🙂

If they want faster memory they'll have to look into the newer 6 and 7Gbps chips, I know Samsung announced some earlier this year. Those chips would certainly work well in a 5890 to finally provide double the bandwidth (ie 224GB/s vs. 115-124GB/s) that many think the 5870 should have had.
 
Originally posted by: Just learning
Originally posted by: MODEL3

I suspect the 5750 (725MHz) will be +20% faster than a 4850 1GB in the worst case and +50% faster in the best case.

So it will be close to 4870 1GB (it will lose in some games that the bandwidth is an issue and will win in some other games that it is not, it depends also from the resolution and the AA & AF levels)
Click to expand...

Well this is not a bad strategy then.

If ATI is aiming this card at the lower resolution mainstream market it can afford to give up some memory bus space for more stream processors on the die.

As long as higher resolutions aren't used the trade-off should result in higher performance compared to an identical die size GPU.
Click to expand...

If the specs are what mymypc & Fudzilla are saying, i guess the probable thing would be that their sample 5750 card was real.
This probably means that their 5750 3Dmark was real.
This means that my performance prediction is not accurate. (based on the mymypc data)
But i don't see why is that.
I mean why a 5750 with 1120SP and 1150MHz GDDR5 to have so little performance difference in relation with a 4770 with 640SP and 800MHz GDDR?
It doesn't make sense.
Maybe mymypc new the specs but faked the results to get hits?
Anyway if the mymypc 5750 results are true you have to deduct at least -10% from my prediction.
 
Originally posted by: bunnyfubbles
Originally posted by: MODEL3
About memory speed:

The memory may be certified for 2.5GHz, but i guess in this early state of GDDR5 it will be difficult for ATI to achieve the certified speed in 5770 SKU.
(the memory of 5870 is also certified at 5Gbps but ATI managed to clock them at 1,2GHz (2,4GHz effectively).

Let's hope that ATI will manage somehow to clock them at their certified speed.
Click to expand...

I'm pretty sure they can clock to their rated speed (and beyond, considering the claims of 1300MHz overclocks). The difference here is that instead of pushing them to the very brim, they're kept back a few % for stability purposes as well as to differentiate classes. Of course the 5850 being clocked back to 1000MHz (or to a lesser extent, a 5700 with mem @ 1150MHz) when it has chips certified to be capable of 1250MHz just like the 5870's 1200MHz chips...yeah, its mega win for the prudent overclockers 🙂

If they want faster memory they'll have to look into the newer 6 and 7Gbps chips, I know Samsung announced some earlier this year. Those chips would certainly work well in a 5890 to finally provide double the bandwidth (ie 224GB/s vs. 115-124GB/s) that many think the 5870 should have had.
Click to expand...

Well, if they can clock to their rated speed, it will be super.
But probably because they have to buy handred of thousands ICs, i guess not all of them will be able to achieve their certified speed in a 5770 PCB (from a stability point of view)

About the 7Gbps certified ICs, it is on 50nm process and the quantity probably will be small and they are going to have higher power consumption/thermal characteristics (probably same as 60nm 6Gbps ICs) in relation with lower certified ICs (50nm).

My prediction is that in the best case scenario will see them (7Gbps) in Q3 2010.
 
Hynix announced their 54nm 7Gbps GDDR5 last November, and that Samsung announcement was also last winter. Will it really take that long to see 7Gbps chips implemented?
 
Originally posted by: bunnyfubbles
Hynix announced their 54nm 7Gbps GDDR5 last November, and that Samsung announcement was also last winter. Will it really take that long to see 7Gbps chips implemented?
Click to expand...

It doesn't say much when they announced it imo.

I suppose Samsung means that they will be able to reach 7Gbps with 50nm but they have to sell first their 60nm stock and also we don't know when the 50nm yields will be mature enough in order for samsung to be able to sell the 7Gbps certfied ICs with the right price and also having quantity to fill the demand.

To give you an example:

Samsung is planning to have 7Gbps ICs, in order to sell them.
Hynix had in Q2 2008 5Gbps ICs.
But we got cards with 5Gbps ICs just now (more than 1 year and one quarter later...)

http://www.hynix.com/datasheet...log_GraphicsMemory.pdf

 
well, that article from February was talking about Samsung putting designs in to mass production, and also talked about how they had developed their 50nm process to the point where its production efficiency was 100% higher than their older 60nm.

Maybe you're right, but I still think we'll see them sooner than Q3'10

its too bad though, 7Gbps chips in the 5700s would certainly eliminate 128bit bus bandwidth concerns.
 
Originally posted by: bunnyfubbles


7Gbps chips in the 5700s would certainly eliminate 128bit bus bandwidth concerns.
Click to expand...

I haven't seen benchmarks yet (except for the one at mymypc.com) but it almost looks like this idea of having the Juniper 180mm2 die with 1120 stream processors and 128 bit bus was made for 7Gbps memory.

Maybe we will see a HD5790 refresh with 7 Gps memory in the future?

P.S. Didn't a lot of people originally predict 180mm2 Juniper would be a DX11 shrink of HD48xx (ie, 800 stream processors with 256 bit bus). Instead ATI subtracts memory bus and adds stream processors. The memory bus is on the die right?

 
Just learning, I wouldn't count on faster memory any time too soon, otherwise it would have been used on at least the 5870. I also wouldn't count on seeing it on any other AMD card prior to a potential 5890.

That being said, if that slide is true, it looks like it will essentially be a 4890 with crippled memory bandwidth... $160 would be a fair-ish price, although a 4890 should definitely outperform it.

I'd also expect the 5750 to be something along the lines of 700MHz GPU, 1150MHz Mem, 720SPs 36TMUs 16ROPs, otherwise it could just be overclocked to 5770 speeds. Granted, I think I'd rather have seen a 5770 and 5750 parallel what we had with the 4870 and 4850 - ie same GPUs (but different clock rates) but different memory bandwidths - give the 5750 a 128bit bus but the 5770 a 256bit one (since this time the difference would have to be there, instead of in the memory type, ie GDDR3 vs. GDDR5).
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top