Poll: GT300 VS. HD5870

[Red Storm]

Originally posted by: Wreckage

Originally posted by: Red Storm
It should be faster since it's coming out after the 5800 series. Similar performance with a later release is not exactly a good thing.

The 4800 series came out after the GTX2xx series.

Aside from the fact that it came out a week later as others have already mentioned, the price/performance is what matters. If GT300 performs similar to 5800 but sells for cheaper, then that would be a victory for nVidia. However, I feel like that won't happen, and so I'm expecting nVidia to have a better performing card.

 

[Kuzi]

One important fact to keep in mind is that the GT300 is most likely much larger in size than the RV870, this means that it will run at lower clocks. That could be a reason why the advantage of the GT300 over RV870 may not be as large as some people believe it will be.

Looking at the performance of the HD5870, while it's pretty good, it's not really getting anywhere near it's theoretical limits. This suggest a bottleneck somewhere that may or may not be improved by software/drivers.

 

[darXoul]

Originally posted by: Kuzi
Looking at the performance of the HD5870, while it's pretty good, it's not really getting anywhere near it's theoretical limits. This suggest a bottleneck somewhere that may or may not be improved by software/drivers.

Memory bandwidth mainly. New drivers will surely improve performance like they did last gen but don't expect anything revolutionary.

I personally think that speculating about GT300's performance now, without any hard figures regarding specs, is pointless. We may be wiser tomorrow when they show their card. I'm pretty confident that the chip will be faster than Cypress but whether it can keep the last gen card's (relatively modest) advantage, remains to be seen.

A possible reason for "healthy but moderate enthusiasm" considering Fermi's performance is the fact that while some time ago nVidia heralded the chip as THE upcoming fps monster with unprecedented power, they now tend to emphasize parallel computing and features instead of raw speed.

We'll see...

 

[Staples]

Originally posted by: Astrallite
About the same as 4870 vs GTX280 I would say. The thing is nvidia cards aren't inherently faster but have more market support. On purely 3rd party tech demos the difference is far less.

Very well said. Older games were all made tailored to NVIDIA card optimizations and still plenty of games still are but the trend has been changing over time and most games are agnostic now. But there are a handful of games that run faster on NVIDIA cards, not because the NVIDIA cards are faster but just because the games are optimized for their architecture.

 

[Idontcare]

Originally posted by: darXoul

Originally posted by: Kuzi
Looking at the performance of the HD5870, while it's pretty good, it's not really getting anywhere near it's theoretical limits. This suggest a bottleneck somewhere that may or may not be improved by software/drivers.

Memory bandwidth mainly. New drivers will surely improve performance like they did last gen but don't expect anything revolutionary.

I personally think that speculating about GT300's performance now, without any hard figures regarding specs, is pointless. We may be wiser tomorrow when they show their card. I'm pretty confident that the chip will be faster than Cypress but whether it can keep the last gen card's (relatively modest) advantage, remains to be seen.

A possible reason for "healthy but moderate enthusiasm" considering Fermi's performance is the fact that while some time ago nVidia heralded the chip as THE upcoming fps monster with unprecedented power, they now tend to emphasize parallel computing and features instead of raw speed.

We'll see...

From further up in the thread:

Originally posted by: Idontcare

Originally posted by: Idontcare

Originally posted by: dguy6789
The limitation on the 5870 is most certainly memory bandwidth. Memory bandwidth is the only category in which the 5870 does not outclass the 4870x2 in.

Radeon HD 5870:

Pixel Fill Rate: 27200 MPixels/sec
Texture Fill Rate: 68000 MTexels/sec
Flops: 2720 GFLOPS
Memory Bandwidth: 153.6GB/sec

Radeon HD 4870x2:

Pixel Fill Rate: 24000 MPixels/sec
Texture Fill Rate: 60000 MTexels/sec
Flops: 2400 GFLOPS
Memory Bandwidth: 230.4GB/sec

I would think this theory would be fairly easy to test and either prove or disprove by anyone who has the 5870 in hand.

Just underclock the memory. Do stock, underlocked mem by 10-15%, underclocked mem by 20-30%, and plot the benchmark data.

If you get a linear line with nice R^2 and the slope is nice and large (slope of 1 on a % bandwidth delta versus % fps delta would be perfect) then you can easily conclude the performance would be improved by having higher bandwidth above stock.

What is interesting is that no one seems to have done this test in their reviews yet (or did I miss it?).

What would be even more interesting is if the data proved out this theory as surely AMD would have tested this and determined whether or not bandwidth was really going to be the limitation of the 5870 well before the launch. So why would they artificially handicap the 5870 by knowingly starving it of bandwidth? I don't think we should assume ignorance on their part here.

Firingsquad ran some tests along this line of thought:

OC?ing the graphics core delivered a performance improvement of 4-5% for the 5870 in CoD. In comparison, OC?ing the memory to 1330MHz improved performance by up to 2%.

http://firingsquad.com/hardwar...overclocking/page3.asp

If you can overclock the GPU cores and see a performance improvement that exceeds that which comes from increasing the memory clocks then that is about as close to proof you are going to get that your compute system is not memory bandwidth constrained.

So why does a 5870 not deliver 2X the performance of a 4870 despite having twice the compute units? Its not because of bandwidth, which is exactly what AMD stated from the beginning and who should know better than them? That leaves two contenders - immature drivers or actual bottleneck in the hardware itself (not enough dispatchers, etc).

 

[swing848]

ATI could have taken more time with the GPU and redesigned it heavily, including a 512 bus. The biggest problem with that is more time [NVIDIA would probably have their card out first], cost of R&D and cost of a new fabrication process that would be untried.

ATI played it smart by using the 40nm HD4770 as a test bed then basically doubled internal hardware of the HD4890 using 40nm. This saved a great deal of time and money.

If ATI can kick their marketing department in the rear they could go into overdrive with a lot of clout with software manufacturers before Nvidia's next generation hits the shelves. And this is happening, at least to some extent, because some companies are already writing code for ATI's new card.

The longer it takes Nvidia to bring a new card to market the more leverage ATI will have.

Again, as to which company will have the fastest video card is yet to be seen. Personally I like competition, it drives prices down and new products are brought to the market faster.

Dan

 

[Nemesis 1]

Not impressed.

http://www.techarp.com/article...rison/nvidia_4_big.png

 

[Idontcare]

Originally posted by: Nemesis 1
Not impressed.

http://www.techarp.com/article...rison/nvidia_4_big.png

Agreed, table contains hardly any data at all. Weak. A real table has at least 2^11 rows.

 

[OCGuy]

Originally posted by: Idontcare

Originally posted by: Nemesis 1
Not impressed.

http://www.techarp.com/article...rison/nvidia_4_big.png

Agreed, table contains hardly any data at all. Weak. A real table has at least 2^11 rows.

:laugh:

 

[Nemesis 1]

Originally posted by: Idontcare

Originally posted by: Nemesis 1
Not impressed.

http://www.techarp.com/article...rison/nvidia_4_big.png

Agreed, table contains hardly any data at all. Weak. A real table has at least 2^11 rows.

So true , But its = to all the other BS hype going around. Like don't buy ATI because NV is soon to release the 385. BS. This whole thread is NV hype about a card that won't appear for 6 months. By than ATI will be almost ready with the 6870 . So buy ATI now and latter when the 385 finely arrives ATI will have the 6870 readdy to slap the 385 around . This is more factual than anything written in this thread.

 

[OCGuy]

Originally posted by: Nemesis 1


So true , But its = to all the other BS hype going around. Like don't buy ATI because NV is soon to release the 385. BS. This whole thread is NV hype about a card that won't appear for 6 months. By than ATI will be almost ready with the 6870 . So buy ATI now and latter when the 385 finely arrives ATI will have the 6870 readdy to slap the 385 around . This is more factual than anything written in this thread.

Haha? The card that is reportedly being shown today is 6 months away from being released?

You and Kyle are going to be very embarassed if it isnt true.


6870 in 6 months? You are waaaaay out there man.

 

[Stoneburner]

Originally posted by: Wreckage

Originally posted by: error8

Originally posted by: Beanie46
Well, the GT300 better be about 2X as fast as the 5870 given its reputed retail price of $579.

Let's hope she is, then we'll see the 5870 price sliding gently to 200-250$, just the way we like it.
Even if it proves to be slower then 5870 and I seriously doubt it will happen, since now Nvidia it's forging that card as hard as possible to make it faster, then I'll probably have a 200$ Nvidia piece for me. Anyway your take it, it's a win win situation. But we really need those cards to be launched already.

I'm not a fan of paying high prices for video cards, but if ATI\AMD does not start making money soon they will be out of business by the end of next year.

How many ati cards have you owned? Let's be frank about this, ATI cannot become profitable when people such as yourself start bashing them despite releasing stellar products like the 4800 and 5800 series. I'm not personally attacking you because you are entitled to brand loyalty, but understand brand loyalty in the face of reason harms competitors.

 

[Kuzi]

Originally posted by: darXoul

Originally posted by: Kuzi
Looking at the performance of the HD5870, while it's pretty good, it's not really getting anywhere near it's theoretical limits. This suggest a bottleneck somewhere that may or may not be improved by software/drivers.

Memory bandwidth mainly. New drivers will surely improve performance like they did last gen but don't expect anything revolutionary.

As IDC and others posted, the HD5870 in most cases does not seem to be memory bandwidth constrained. One guy at XS forums gained only 2-3% performance for every 100MHz (400MHz effective) increase. I'm not saying more memory bandwidth won't help, and some games at higher graphics settings can use all the available memory bandwidth, but it's probably not the main reason for the lower performance on the HD5870.

Originally posted by: Nemesis 1
Not impressed.

http://www.techarp.com/article...rison/nvidia_4_big.png

GT300 with only 2 billions transistors? My expectation was it would be a little over 3 billion.

 

[swing848]

I have read at least one benchmark result that indicated the bottleneck was the Intel i7 CPU overclocked to 4GHz...

I also read that it will take an i7 [or like CPU] running at 4GHz to keep the HD5870 fed. If that is true what are people going to do when the HD5870 X2 and HD5850 X2 are released?

I have read so many reviews that I cannot remember which site I found that information on but I will look for it and if I can find it again I will post it here.

If the above is valid and as so many people are hoping for, if the GTX 380 is considerably faster than the HD5870 [that is still an unknown], what kind of CPU power will be needed to keep it fed?

 

[akugami]

Originally posted by: swing848
I have read at least one benchmark result that indicated the bottleneck was the Intel i7 CPU overclocked to 4GHz...

I also read that it will take an i7 [or like CPU] running at 4GHz to keep the HD5870 fed. If that is true what are people going to do when the HD5870 X2 and HD5850 X2 are released?

I have read so many reviews that I cannot remember which site I found that information on but I will look for it and if I can find it again I will post it here.

If the above is valid and as so many people are hoping for, if the GTX 380 is considerably faster than the HD5870 [that is still an unknown], what kind of CPU power will be needed to keep it fed?

That would be interesting. I think we need an in depth look at the overclocking of the 5850 and 5870 to see if there are any bottlenecks.

 

[Wreckage]

Originally posted by: swing848


If the above is valid and as so many people are hoping for, if the GTX 380 is considerably faster than the HD5870 [that is still an unknown], what kind of CPU power will be needed to keep it fed?

If the new GPGPU capabilities are as impressive as they seem. Maybe The GTX can even take some of that load off the CPU.

Imagine what kind of performance that would bring. :shocked:

No more CPU bottleneck.

 

[alcoholbob]

Performance gains roughly linearly with core/mem increases. However, most of the performance gain is attributed to core increase and not mem, roughly a 2:1 ratio.

The GT300/GF100 certainly has impressive specs, but how much of that power is allocated to GPGPU? And if their AA algorithm still sucks the HD5870 will still look very competitive in benchmark reviews.

 

[darXoul]

Well, maybe it's not bandwidth but I'm certainly not buying the "immature driver" argument.

 

[Keysplayr]

Originally posted by: darXoul
Well, maybe it's not bandwidth but I'm certainly not buying the "immature driver" argument.

While I'm sure there will be the usual increases in driver revisions, 5% here, 10% there, the immature driver argument doesn't hold water. I agree. It is obvious that it didn't take much to amp up AMD's 10.1 products to 11. At least that is what 10.1 proponents stated many a time. "DX10.1 is closer to DX11 than you realise", isn't that what they said?
Ok then, it's not illogical to believe that with only a few driver tweaks of the same drivers used for the 4xxx series would allow R870 to run with them. It's just two R770 cores (technically) with DX11 compliance. Driver immaturity probably doesn't apply IMHO.
Not ruling out all possibilities though.

 

[Kuzi]

Agreed. I'm sure the "driver immaturity" statement would apply more to a new architecture such as Fermi when it gets released early next year.

 

[lavaheadache]

Originally posted by: Wreckage
snip

Originally posted by: Wreckage
snip

Originally posted by: Wreckage
snip

Originally posted by: Wreckage
snip

Originally posted by: Wreckage
snip

Originally posted by: Wreckage
snip

Originally posted by: Wreckage
snip

I was reading through your posts and edited out the bs and left everything that was useful.

GT300 is surely gonna beat up on the hd5870 but at what price? I imagine by the time of release one could expect to pick up 2x hd5870's for rouhly the same price if an HD5890 or x2 isn't available. I know some of you don't like multicard solutions so this isn't for you. Amd has stated now for the last couple generations that this was thier plan. If you don't like it, buy something else. =)

 

[Keysplayr]

Originally posted by: Kuzi
Agreed. I'm sure the "driver immaturity" statement would apply more to a new architecture such as Fermi when it gets released early next year.

Uncertain about release time, but I agree about drivers for Fermi being a new arch.

 

[Idontcare]

Originally posted by: Nemesis 1
So true , But its = to all the other BS hype going around. Like don't buy ATI because NV is soon to release the 385. BS. This whole thread is NV hype about a card that won't appear for 6 months. By than ATI will be almost ready with the 6870 . So buy ATI now and latter when the 385 finely arrives ATI will have the 6870 readdy to slap the 385 around . This is more factual than anything written in this thread.

Nemesis where are you seeing people saying don't buy ATI because of NV's yet-to-be-released DX11 products?

I don't claim to read every thread or every post in the threads, but I can't say I've seen posts of this nature so if it is occurring it must be at a rather low frequency which would imply that is represents a a viewpoint held by a minority.

If I were needing/wanting to buy a new GPU today I'd probably be compelled to get the 5870...but if I was anywhere close to sitting on the fence in regards to my purchase timeline (meaning I could defer 3 months without really cramping my style) then personally I'd wait for no other reason than that the prices aren't going to go up in the meantime and if NV does release their card in that timeframe then more than likely either prices will go down or my choices will increase...again neither outcome being something I view to be a negative.

But I do understand there are consumers who happen to be in a position where they'll buy now regardless what the future may hold. When I bought my QX6700 right after release I knew AMD was going to release their quadcore at some point in the near-future (or so we were led to believe at the time)...but I was simply not in a position to delay that purchase, time was money in that case. And as it turned out AMD took nearly another year to get their quad out so in hindsight it was the right thing to do at the time.

Originally posted by: darXoul
Well, maybe it's not bandwidth but I'm certainly not buying the "immature driver" argument.

If its not bandwidth (as both AMD has stated and the data confirms), and if it is not a driver induced handicap (either intentional or unintentional), then what does that leave to explain the performance scaling discrepancy of Cypress versus 2xRV770?

Poor implementation of the architecture at the hardware (design/layout) level? (I'm not stating that to be the explanation, I am asking if that is the explanation)

 

[Tempered81]

Originally posted by: lavaheadache

Originally posted by: Wreckage
snip

Originally posted by: Wreckage
snip

Originally posted by: Wreckage
snip

Originally posted by: Wreckage
snip

Originally posted by: Wreckage
snip

Originally posted by: Wreckage
snip

Originally posted by: Wreckage
snip

I was reading through your posts and edited out the bs and left everything that was useful.

Too long for a sig, but pretty good!

 

[ochadd]

Does a six month wait really merit any kind of wait and see? If GT300 comes out with 20% or better frame rates over a 5870 then sell it and pick up the better card. Probably take a $50-100 hit from the purchase price and you have gotten six months of gaming @ max settings.