At what price difference is a 5850 a justifiable upgrade over a 4890?

[cbn]

I'm also surprized this hasn't came up more.
Some are saying its the drivers and they will unleash more power when Fermi arrives.

Who knows?

Edit:it's also wierd that a 5770 in crossfire beats a 5870.
or 2 5750 beats a 5850. There must be some crossfire overhead yet the single cards loose?
Don't make sense.

But what happens when we compare a single HD5750 to a single HD4850?

Both have the same processing power (720 stream processors @ 700 Mhz vs 800 stream processors @ 625 Mhz), yet HD5750 is faster?

So I am not sure if we can say HD5xxx stream processors are always weaker. In the case of the lower cards Evergreen stream processors actually stronger than RV870 stream processors.

P.S. I agree these relationships between HD4xxx and HD5xxx are pretty weird. I am definitely scratching my head as to why this is happening?

 

[happy medium]

But what happens when we compare a single HD5750 to a single HD4850?

Both have the same processing power (720 stream processors @ 700 Mhz vs 800 stream processors @ 625 Mhz), yet HD5750 is faster?

So I am not sure if we can say HD5xxx stream processors are always weaker. In the case of the lower cards Evergreen stream processors actually stronger than RV870 stream processors.

P.S. I agree these relationships between HD4xxx and HD5xxx are pretty weird. I am definitely scratching my head as to why this is happening?

Funny I was just thinking of my card vs the 4850 and scratching my head also.

edit: the 4850 is clocked at 625 core. Mabe thats it? The 5750 is at 700 core.

 

[MagickMan]

However, I personally would upgrade to at least a 5870 / GTX 470+

I can't really see why. A 5850 and 5870 have roughly the same OC ceiling, and at that ceiling the 5870 has a 4-6% advantage (due to number of shaders and memory clock). Is it worth $100 for 4-6%? For a few, maybe, for most, no. Now, if the person is a newb or lazy and doesn't OC at all, you have a point.

 

[cbn]

edit: the 4850 is clocked at 625 core. Mabe thats it? The 5750 is at 700 core.

Yes, but HD5750 only has 720 stream processors. HD4850 has 800 stream processors.

When comparing HD5750 against HD4850 we also need to keep VRAM the same. I might have accidently compared HD5750 1GB to HD4850 512 MB @ 1920 x1200.

 

[happy medium]

Yes, but HD5750 only has 720 stream processors. HD4850 has 800 stream processors.

When comparing HD5750 against HD4850 we also need to keep VRAM the same. I might have accidently compared HD5750 1GB to HD4850 512 MB @ 1920 x1200.

I really don't have a clue. Very odd :hmm:

 

[cbn]

I really don't have a clue. Very odd :hmm:

I just checked the Anandtech benches.

At 1680 x 1050, HD5750 is anyehere from 7-8% slower to 11 % faster than HD4850 despite having less than 1% extra processing power.

Crysis warhead was the game HD5750 beat HD4850 by 11% at 1680x 1050 resolution. Hopefully 512MB isn't limiting Crysis FPS at that resolution.

 

[toyota]

I just checked the Anandtech benches.

At 1680 x 1050, HD5750 is anyehere from 7-8% slower to 11 % faster than HD4850 despite having less than 1% extra processing power.

Crysis warhead was the game HD5750 beat HD4850 by 11% at 1680x 1050 resolution. Hopefully 512MB isn't limiting Crysis FPS at that resolution.

the 5750 has slightly more bandwidth(15%) than the 4850 too.

 

[happy medium]

the 5750 has slightly more bandwidth(15%) than the 4850 too.

Thanks Toyota,
Mabe the higher core clocks and higher memory bandwidth make up for the 80 less shaders?.
Could also explain why it wins some loses some depending on the game.

 

[cbn]

Thanks Toyota,
Mabe the higher core clocks and higher memory bandwidth make up for the 80 less shaders?.

HD4850= 1.0 TFLOPs
HD5750= 1.008 TFLOPs

I'm sure the memory bandwidth of HD5750 helps too. (In BFG10K's HD5770 article, he found the card mostly balanced with a slight bias towards the core. Downward clocking memory affected performance about 75% as much as core if I recall correctly).

However, none of this really explains what is happening with the higher end cards? Unless memory bandwidth is affecting Cypress? But then if this were true we would be seeing different results when comparing HD5830 to HD4890.

 

[toyota]

HD4850= 1.0 TFLOPs
HD5750= 1.008 TFLOPs

I'm sure the memory bandwidth of HD5750 helps too. (In BFG10K's HD5770 article, he found the card mostly balanced with a slight bias towards the core. Downward clocking memory affected performance about 75% as much as core if I recall correctly).

However, none of this really explains what is happening with the higher end cards? Unless memory bandwidth is affecting Cypress? But then if this were true we would be seeing different results when comparing HD5830 to HD4890.

remember though even the low end cards acted weird. http://www.anandtech.com/video/showdoc.aspx?i=3734&p=15

This is especially evident on the 5450, where the 5450 has a 50MHz core speed advantage over the 4550, and yet with everything else being held equal it is still losing to the 4550 by upwards of 10%. This seems to the worst in shader-heavy games, which leads us to believe that actual cause is that the move from DX10.1 shader hardware on the 4000-series to DX11 shader hardware on the 5000 series. Or in other words, the shaders in particular seem to be what’s slower.

 

[cbn]

remember though even the low end cards acted weird. http://www.anandtech.com/video/showdoc.aspx?i=3734&p=15

This is especially evident on the 5450, where the 5450 has a 50MHz core speed advantage over the 4550, and yet with everything else being held equal it is still losing to the 4550 by upwards of 10%. This seems to the worst in shader-heavy games, which leads us to believe that actual cause is that the move from DX10.1 shader hardware on the 4000-series to DX11 shader hardware on the 5000 series. Or in other words, the shaders in particular seem to be what’s slower.

That's a good find. Thank you.

 

[RussianSensation]

I can't really see why. A 5850 and 5870 have roughly the same OC ceiling, and at that ceiling the 5870 has a 4-6% advantage (due to number of shaders and memory clock). Is it worth $100 for 4-6%? For a few, maybe, for most, no. Now, if the person is a newb or lazy and doesn't OC at all, you have a point.

Good points.

Personally I don't overclock videocards (so consider me a noob on that end). Here is why I don't overclock videocards:
- Overclocking is not guaranteed
- With stock coolers that already buzz in my ear on ATI cards at load (and that's at stock speeds), I cant imagine running 5850 @ 900mhz+ comfortably
- 5850 is $310+, far above its $259 MSRP. For $310, you can get 2 5770s
- Other than Metro 2033, 5850 OC or 5870 don't really offer 2x the performance over OP's 4890. I would consider them more favourable if they were priced lower. I only mentioned 5870 as the minimum base to upgrade, since in stock form, I do not believe that 5850 offers enough of a performance improvement.

 

[MagickMan]

Good points.

Personally I don't overclock videocards (so consider me a noob on that end). Here is why I don't overclock videocards:
- Overclocking is not guaranteed
- With stock coolers that already buzz in my ear on ATI cards at load (and that's at stock speeds), I cant imagine running 5850 @ 900mhz+ comfortably
- 5850 is $310+, far above its $259 MSRP. For $310, you can get 2 5770s
- Other than Metro 2033, 5850 OC or 5870 don't really offer 2x the performance over OP's 4890. I would consider them more favourable if they were priced lower. I only mentioned 5870 as the minimum base to upgrade, since in stock form, I do not believe that 5850 offers enough of a performance improvement.

Considering noise, that's on a card-by-card basis. Some seem to handle the temps much better than others. Take, for example, the 2 ASUS 5850s I purchased recently. Mine seems to work exceptionally well up to 1020MHz and ~1.29v, not making much more noise at load than when it's running stock. OTOH, my brother-in-law's card turns into a hair dryer at the same speeds. It's a puzzler and possibly due to TIM application, but I didn't want to void his warranty trying to fix it. He doesn't OC anyway (lazy), so it's just as well for him.

As for a pair of 5770s, I'd need to purchase a board with 2+ PCI-E 16x slots, and I'd rather not put more money into S775 right now. I'll just wait and buy an i7 board, CPU, and RAM at the end of this year and get another 5850. I figure at that time a pair of them OCed will still be pretty impressive.

I do somewhat disagree about OCing, 5850s@1GHz are as close to a gimme as I've ever seen in video cards. Personally, I think they're all capable, given a modest amount of effort.

 

[SunnyD]

Well, I've decided to pass on the 5850 deal at this point due to various reason. The fact remains I have an apparent serious inquiry into my 4890 @ $195... but I have bigger fish to fry for now.

So unless a killer deal falls in my lap in the next few weeks... 4890 it is.

 

[RussianSensation]

$195 is a killer deal on a used 4890. I am amazed someone would offer that.

 

[SunnyD]

$195 is a killer deal on a used 4890. I am amazed someone would offer that.

Not like it matters much...

http://www.google.com/products/cata...ct=result&resnum=3&ved=0CBIQ8wIwAg#ps-sellers

If you post that link on Craigslist, people think they're getting a steal at $195. (Yes, that is the exact card I have)

 

[Munky]

At 2560x1600 4x AA, HD5850 is showing only a 16% improvement over HD4890 (using 3.33 Ghz Core i7 for the CPU).

That tells me the CPU is the bottleneck even at very high resolution.

This means the only way for a person to get their moneys worth out of such a Video card would be either 1) Increase the number of monitors and/or 2) Overclock the CPU.

This begs the question why are these gigantic GPUs even being produced?

Your calculations and assumptions would be very wrong. The fact that it's showing only a 16% increase is inline with the fact the a 5850 is not a huge leap in gpu power beyond a 4890.

 

[cbn]

The fact that it's showing only a 16% increase is inline with the fact the a 5850 is not a huge leap in gpu power beyond a 4890.

Based on GPU power alone I would have expected at least 30% improvement (not 16% improvement).

Something is holding HD5850 back. Maybe its shader design? Maybe its memory bandwidth? Maybe its drivers? I don't know.

 

[nyker96]

for me probably around 175.

personally I wait until it gets below 100. but that's me. :]

 

[RussianSensation]

Not like it matters much...

http://www.google.com/products/cata...ct=result&resnum=3&ved=0CBIQ8wIwAg#ps-sellers

If you post that link on Craigslist, people think they're getting a steal at $195. (Yes, that is the exact card I have)

Hhee I got the exact same card. It's pretty loud when Crysis stresses it! Maybe I should post it on sale as well.

 

[SunnyD]

Hhee I got the exact same card. It's pretty loud when Crysis stresses it! Maybe I should post it on sale as well.

Actually someone else posted on in FS/FT this morning for $150 OBO. I was tempted.

 

[happy medium]

Actually someone else posted on in FS/FT this morning for $150 OBO. I was tempted.

Sell the 4890 195$ and grab 2 5770's for 300$. 105$ buys you 5870 speeds and good overclocking.

There is a guy in the for sale forums that selling 2 5770's with aftermarket cooling and ramsinks for 320$.

If you overclock the 5770's they should match 4890 crossfire performance.
The 5 series scales better then the 4 series.

 

[SunnyD]

Sell the 4890 195$ and grab 2 5770's for 300$. 105$ buys you 5870 speeds and good overclocking.

There is a guy in the for sale forums that selling 2 5770's with aftermarket cooling and ramsinks for 320$.

If you overclock the 5770's they should match 4890 crossfire performance.
The 5 series scales better then the 4 series.

Actually, I just picked up a 5870. So now all I need to do is find a home for the 4890.

 

[happy medium]

Actually, I just picked up a 5870. So now all I need to do is find a home for the 4890.

Nice! and good luck.

 

[RussianSensation]

Actually, I just picked up a 5870. So now all I need to do is find a home for the 4890.

Sweet card!