videocardzAMD Radeon R9 290X confirmed to feature 64 ROPs

[GaiaHunter]

From my own testing with Crysis 3 @ 1080P high settings and 2XSMAA - I saw a 14% increase going from 5Ghz to 7.4Ghz on the memory with the same core clock. That percentage might grow at a higher resolution? Not all games respond that way though.

But that is like a 50% OC and it was your highest increase?

With my 7870 XT I just bother with the core.

Also, for example, Titan has 75% more shaders but only 50% more bandwidth than the GTX680.

 

[lopri]

Crazy clocks. My GTX 580 barely did 850 MHz (825 MHz was rock-solid) and HD 7950 did 1100 MHz (1050 MHz was rock-solid). Both were well ventilated.

 

[Slomo4shO]

You mean 280X Toxic, right? (for the link)

Yes, thanks for catching that.

A few things.

There is only a 512 bit bus GDDR5.

One can claim the 384 bit buses matured since we see the improvement from the GTX480 to today Tahitis and GK110s. No doubts about that.

What problems are associated to 512 bit bus with GDDR5 we don't know.

Second the 512 bit bus is actually smaller in die size to the 384 bit bus of the Tahitis.

So the point is to save die space and save on power/cheaper memory, since 5Gbs GDDR5 chips give 320GB/s bandwidth.

6GBs chips are cheaper than 7GBs.

Additionally 512bit bus allow for a 4GB configuration.

That is the stock value.

If the memory can be overclocked to just 6GHz the bandwidth will be 388GB/s.

I am hoping that high memory overclock potential is the case with the 290X considering that reviews of the 280X DCII, Matrix, Toxic, Gaming and WindForce have been able to OC the memory to above 7GHz:

http://www.kitguru.net/components/g...phire-r9-280x-toxic-edition-oc-3gb-review/21/

http://www.kitguru.net/components/graphic-cards/zardon/asus-r9-280x-matrix-platinum-review/21/

http://www.techpowerup.com/reviews/MSI/R9_280X_Gaming/29.html

http://www.guru3d.com/articles_pages/gigabyte_radeon_r9_280x_windforce_review,26.html

http://www.guru3d.com/articles_pages/asus_radeon_r9_280x_top_review,26.html

 

[rgallant]

Crazy clocks. My GTX 580 barely did 850 MHz (825 MHz was rock-solid) and HD 7950 did 1100 MHz (1050 MHz was rock-solid). Both were well ventilated.

I have 3 gaming profiles for my 2 580's , 772,860,900 all stable and under 40c.
nothing compare to the new cards but 940 could be used , but for me it wouldn't make a difference so why beat the cards up.

 

[omeds]

I had 3-way reference 580's that all ran at 930 mhz np.

 

[krumme]

Neither one makes sense for a top tier GPU. It appears that this card may be deliberately throttled so that there may be a subsquent release with better performance in the even that Nvidia releases the Titan Ultra. I guess we shall have more details in a week.

Agree. Very low clocks for both core and especially mem. It makes sense to wait for nv to answer to 290, eg titan ultra, and know how high to go on freq and how to price that response.

If the oem are alowed for amd we will see some very high factory oc from day one on 290.

Still it looks like nv is in the same production cost disadvantage at the 290 end as amd is on the 280 end.

 

[RussianSensation]

860 is the default frequency of that model of 580, lol classic "high OC"...
:thumbsup:

The shocking part is that when someone points out to you that the 7970 @ 1125mhz is yielding huge 40-50% gains over 580 @ 860mhz, you state this test doesn't show 580 in the best light (because some of them can hit 930-950mhz), but then you ignore the other side which is overclocking 7970 beyond 1125mhz. In simplest terms, the 50% increase over 580 oc will remain because the 7970 can scale more as well.

Every time someone points those benchmarks, you ignore them and yet you what you say doesn't change the conclusion at all unless you LN2 the 580.

7970 overclocked also beats 6970 overclocked by 75-80% in many modern games. Benchmarks are everywhere. Check TR's review of R9 280X where it absolutely crushes the 6970.

It takes a GTX680 boosting to 1280-1290mhz to just match a 7970 at 1165mhz or so. A 580 oc never stands a chance. In BF4 beta, the 580 gets absolutely crushed by 7970GE. If R9 290 is $499 and once it is overclocked, it can trade blows with 780/Titan overclocked, that will absolutely be a win for AMD if this comes true.

I never made any arguments regarding correlations between performance and memory bandwidth, that is your own assumption. I was simply stating that memory speeds have increased over time and that 320GB/s bandwidth over a 512 bit bus is not impressive by any means.

That's the entire point of my post that you still missed. You look at memory bandwidth as only a number and a function of how wide the bus width is vs. the total memory bandwidth attained. Without looking at the context of the GPU speed, comparing 288GB/sec bandwidth of GPU A vs. 320GB/sec of GPU B is a useless exercise. It's like comparing a car with 288 hp vs. 320 hp but not looking at the curb weight....waste of time. You can have a GPU like Tahiti that has 288GB/sec but only benefits from 200GB/sec because the GPU is a major bottleneck. The only thing you are comparing is a number on a piece of paper in theoretical terms, but the only thing that matters is how efficiently that GPU can actually utilize the memory bandwidth in real world programs.

 

[BallaTheFeared]

Nobody knew the full extent of what was possible on some Thati products, what was known was what we'd had for over a year at that time. The 580 OC potential wasn't reflected at all in that review. Your inablity to gain context from text is astounding. As is your ability to go off on tangents which have nothing to do with the quote you're discussing.

I don't know what else you said, but honestly I don't even care anymore.

 

[RussianSensation]

The 580 OC potential wasn't reflected at all in that review. Your inablity to gain context from text is astounding.

Shocking how I outline logically that overclocking 580 beyond 860mhz doesn't matter at all since 7970 can overclock beyond 1125mhz as well. You have no point at all.

As is your ability to go off on tangents which have nothing to do with the quote you're discussing.

Sure it does. You continue to downplay 7970's performance increase over 580 for the last 2 years and you are doing it again with your unsubstantiated claims that a card with 64 ROPs and 512-bit bus is unbalanced.

I don't know what else you said, but honestly I don't even care anymore.

Of course you don't. You can't rebuttal an argument that is presented with facts and logic that opposes your opinion that has no facts. Essentially you constantly spout your opinion as fact and then when confronted, have no response other than "I don't even care anymore."

More amusing is that you fixated on this whole 64 ROP/512-bit bus and rushed to 28nm statement but all the rumours are pointing to AMD's <440mm2 chip trading blows with NV's 561mm2 one. For all we care, this thing might as well come with 1 billion ROPs and memory bandwidth units. What's impressive is that a chip with a die size a whopping ~30% less is going to give NV's big daddy a performance run, and cost less. Hopefully NV will lower prices and/or release an even faster GK110 variant.

 

[GaiaHunter]

960 MHz core over 860 MHz core is like 8%.
1200 MHz core over 1125 MHz core is like 6%.

The 860 vs 1125 seems a pretty accurate description.

And of course, that was with early drivers.

 

[BallaTheFeared]

RS the dicussion is about what took place in early 2013 before anyone other than reviewers had samples in hand.

Your use of BF4 is a clear indication you aren't actually following the conversation.

 

[BallaTheFeared]

960 MHz core over 860 MHz core is like 8%.
1200 MHz core over 1125 MHz core is like 6%.

The 860 vs 1125 seems a pretty accurate description.

And of course, that was with early drivers.

860 > 960 is a 11.7% increase

1125 > 1200 is a 6.7% increase

Of course none of this matters in the context of where the argument came from.


Which I'm no longer going to discuss.

 

[wand3r3r]

860 > 960 is a 11.7% increase

1125 > 1200 is a 6.7% increase

Of course none of this matters in the context of where the argument came from.


Which I'm no longer going to discuss.

If you want to debate, put facts on the table. Statements like you made about the die size and architecture already demeaning an unreleased and as of yet card without reviews simply demonstrates bias. Subjective statements will not lead anywhere. (Show a 580 at your supposed ubiased oc vs. 7970 at your unbiased oc etc.)

 

[BallaTheFeared]

I'm out after this, people are taking perceived slights towards companies personally and that never ends well.

Here is the comment that sparked the response.

I remember people defending heavily OC'ed GTX 460's against HD 6870, but you couldn't use an OC'ed 7970 against a GTX 580.

My reply:

No the issue at the time was [H] used a factory overclocked 580 against a manually overclocked 7970, but changing history isn't anything new.

Anything I said there wrong, anything I said there have anything to do with 1500MHz 7970s? Anything said there have anything to do with performance today? Performance in BF4? Anyone have any idea 1125 was a below average OC for a 7970 on January 9th? Anything at all with what I said that gave the impression that 7970 wasn't way faster than the 580, OC or not? Anything I said in there that made it seem like the 7970 wasn't much faster than the 580 OC vs OC then or now?

Anything at all, to warrant what has been said since?


I also said I would take back what I said about the ROPs and BUS until we see it in reviews and we have more information, but that was of course ignored much like the comment I made was ignored and a new one was spun up for straw man attacks.

 

[Erazor51]

 

[Skurge]

Snip

So Tess performance has also been doubled for those people that care.

 

[Arkadrel]

@Erazor51

Intresting slides.
Seems everything got 30~90% increase compaired to the 7970 1ghz.

I wonder how much that giant Pixel fill rate, and Geometry proccessing boost, will impact its performance and in which types of scenarios.

 

[AtenRa]

Thx for that slides, it answer my question about 16x RBEs

Also, they have even doubled the Front End, Tessellation performance will raise tremendously.

Now all we need is performance numbers

 

[Grooveriding]

What is the 'Crossfire XDMA' in that block doing in relation to Crossfire ?

 

[BallaTheFeared]

I was thinking the same to myself but didn't want to say anything.

I thought maybe it was frame metering, but then I thought it was probably hardware to deal with the connection taking place through PCIe instead of across a bridge.

 

[Tempered81]

^sideporting without the interconnect bridge

This thing is going to be blazing fast. Launching with 3-4% more performance than titan. Memory headroom like crazy. Mantle. Probably 300mhz headroom on core. Good PCB. Over 4x 7770 specs, over 2x 7870 specs, probably going to gain 40% performance over time like 7950 did with 5-10 months of driver improvements to quad geometry efficiency. Somewhere between $550-$730. 300-500 cheaper than titan.

AMD will need something fast to double up when it comes time to match what Nvidia has planned with Maxwell GM100/110 on 20nm

 

[AtenRa]

OK, Cache Hierarchy has been changed. It seams they have changed the i$(Instruction Cache) and K$ (Scalar Cache) ???

Tahiti had 16KB of i$ and 32KB of K$ per 4x CUs,

Now it seams they have decoupled those from the CUs. They have grouped 11 CUs + Caches(i$, K$ and L1) + 4x RBEs in to a single group called Shader Engine, much like NVIDIAs SMX in Kepler architecture.

Im sure they have gained a great deal of density by doing that, but they dont have the same number of i$ and k$ cache per CU anymore.

 

[csbin]

gtx titan

Geometry Processing : 4.4 billion primitives/sec

Compute&#65306;4.7 Tflops

Texture fill rate: 196 Gtexels/sec

Pixel fill rate: 42 Gpixels/sec

Peak Bandwldth: 288GB/sec

 

[BallaTheFeared]

So if this was deadliest warrior we'd say...

Geo Pro: Edge Titan

Compute: Edge 290X

Texture Fill: Edge Titan

Pixel Fill: Edge 290X

Peak Bandwidth: Edge 290X



Should be interesting, especially if there is a response on 28nm from Nvidia before the holiday season besides price cuts.

 

[NTMBK]

Should be interesting, especially if there is a response on 28nm from Nvidia before the holiday season besides price cuts.

Don't expect anything apart from a fully enabled GK110.