AMD's 3xx series tessellation "not" improved - performance boost comes from the drivers

[bogus]

It's the same Hawaii die on the same TSMC 28nm process, perhaps improved yields means better voltages/overclocking but other than that no improvements past PCB revisions and software improvements.

 

I suspect AMD just don't want to put money into making many new cards on 28nm with better processes right around the corner (which usually means a lot of cost, lower yields (so more wasted/failed chips) and much bigger performance improvements) when most of their cards are decently performing at their current prices.

Well what I stated was that replacing bios of a r9 290x and a driver update will not match a 390x because the 28nm process improved. Will it be a big difference? Let the benchmarks speak.

[patrickjp93]

So you are claiming that the same 28nm process is being unchanged since 2013? There were no improvements?

I didn't expect this coming from you.

I'm not saying they are not identical - what I'm saying is that the 28nm process improved.

I'm saying there's no new production. AMD had inventory channels of the 200 series still chock full. Most likely it had most of that inventory shipped back for rebadging.

And are you seeing this? The 290X and 390X match perfectly with the hacked driver. It's the exact same GPU, no process improvements.

[patrickjp93]

R9 290x and r9 390x is not the same.

R9 390x has higher memory clock speed and tighter timings (1500 vs 1250).

R9 390x has higher voltage about +50mv

Same GPU. They just changed the RAM and the board voltage.

[rcarlos243]

Same GPU. They just changed the RAM and the board voltage.

 

If it is the same, then people should not be having problems when they flash the R9 390x BIOS to R9 290x but so far everyone that tried flashing it encountered instability.

 

It is also weird why would AMD decide to increase the voltage in it.

[xAcid9]

If it is the same, then people should not be having problems when they flash the R9 390x BIOS to R9 290x but so far everyone that tried flashing it encountered instability.

 

It is also weird why would AMD decide to increase the voltage in it.

Same core, different memory.

[bogus]

I'm saying there's no new production. AMD had inventory channels of the 200 series still chock full. Most likely it had most of that inventory shipped back for rebadging.

And are you seeing this? The 290X and 390X match perfectly with the hacked driver. It's the exact same GPU, no process improvements.

That's a different story, and if you could back it up it would be great (on the no new production and the perfect match with the hacked driver, if you could).

I just stated the obvious - if the GPUs were being produced in a not so distant past (few months) they would definatly have improvements because the process itself improved.

[rcarlos243]

Same core, different memory.

 

If its the same core, why would AMD increase the VDDC. It doesn't make sense if it is the same core as it would just increase the heat output of an already very hot card.

[DEDRICK]

R9 290x and r9 390x is not the same.

 

R9 390x has higher memory clock speed and tighter timings (1500 vs 1250).

 

R9 390x has higher voltage about +50mv

 

People have already flashed their 290X into a 390X

[harrynowl]

If its the same core, why would AMD increase the VDDC. It doesn't make sense if it is the same core as it would just increase the heat output of an already very hot card.

Because 28nm production has got better. Better and less leaky chips I suppose.

[xAcid9]

If its the same core, why would AMD increase the VDDC. It doesn't make sense if it is the same core as it would just increase the heat output of an already very hot card.

Because they increased the memory clock to 1500mhz, memory voltage is tied to the core voltage for Hawaii cards.

I can overclock my memory to 1500mhz only if i increase the core voltage.

[patrickjp93]

If it is the same, then people should not be having problems when they flash the R9 390x BIOS to R9 290x but so far everyone that tried flashing it encountered instability.

 

It is also weird why would AMD decide to increase the voltage in it.

1000MHz more on the RAM (and thus a needed speed change to the memory controller too) might have something to do with it...

 

BIOSes are complicated, not to mention again the timing for the memory controller and VRAM. It's very possible it just needs tweaking to deal with the power delivery and board subsystems differences.

[patrickjp93]

That's a different story, and if you could back it up it would be great (on the no new production and the perfect match with the hacked driver, if you could).

I just stated the obvious - if the GPUs were being produced in a not so distant past (few months) they would definatly have improvements because the process itself improved.

Your second point amounts to nothing more than binning. Intel's processes improve in the middle of high-volume production as well.

 

As for the hacked driver, you're in the middle of the thread discussing it in the OP. Now, no two runs are ever perfectly the same, but the difference is minimal and very easily caused by the cheaper Elpida VRAM between higher latency and lower bandwidth relative to the Hynix parts on the 390X. In the OP there's already mentions of people hacking the driver onto the 290X.

 

As for the no new production, it is admittedly a theory, but based on the sudden plummeting of 200 series prices we saw 5 months ago and the still lacking sales and now within the last month the 200 prices steadily rising and warnings of low stock arising, I'm putting money on that particular horse of AMD having the inventory shipped back for board and bios swaps under a new name. Can I prove this? Not unless I shipping manifests from the last year from our various e-vendors and their warehouses along with AMD's. This is speculation. We know there was a glut of 200 series cards made right before the cryptocurrency mining crash. I don't think AMD would have gone into large-scale production of the same GPUs on a mildly improved 28nm process (was already very good from being the 2nd or 3rd GPU generation built on it) considering the bad press they'd get from extensive rebadging. It just doesn't fit with AMD's history. That said, it's all speculation on my part, though it is reasoned based on some of the facts at hand.

[rcarlos243]

People have already flashed their 290X into a 390X

 

People that flashed the 390x BIOS is getting instabilities.

even the modded downclocked BIOS wasn't stable.

[Humbug]

People have already flashed their 290X into a 390X

What's the point?

Why not just overclock the 290x and be done with it?

 

[LucidMew]

Unfortunately its near impossible to find GTX 980 non reference 1440p benchmarks with hair works on but we can see a Gigabyte GTX 980 Windforce score a 48fps on ULTRA 1440p no hairworks. (digital foundry has it at 47.5 fps average 36 min)  If we take the general assumption that turning hairworks on will remove 15-20 fps from a gtx 980 We'll have an average of 33fps at best assuming we drop an average of 15fps by turning hairworks on. 

 

These comparison are obviously sketchy as balls but i think my point still stands  is...(and goddamn somebody get all of these gpus and just make a giant very specific benchmark pls) I dont think the 290x and by correctional the 390x is as far behind as a lot of people might think.

 

I happen to have a GTX 980 non reference card (Gigabyte G1 Gaming) and a 1440p monitor, so I went ahead and installed TW3 and ran through the first bit, up to the first Ghouls part. About 25 mins give or take. Everything highest possible, except for small tweaks that I don't particularly like (Depth of Field, Chromatic Aberration), and of course Hairworks on.

 

Specs (from vid description):

OS: Win10 Technical Preview Build 10130

CPU: 5820k @ 4.3GHz

GPU: Nvidia GTX 980 Gigabyte G1 Gaming

GPU Driver: 352.84 (current)

OC Settings:

MoBo: ASUS X99 Deluxe

RAM: 16Gigs of RipJaw4 2133MHz CL15-15-15-35

HDD: Toshiba DT01ACA300 (3Tb Drive)

MSI Afterburner Graph:

Recorded with OBS @ 3000 KB/s

 

 

A couple of notes though.

• The boost should reach 1505+ in games, and for some reason it was peaking at 1366. It occasionally stays at 1228 in other games, from what I've seen, and I have no idea why. Perhaps the voltage is limiting it? Dunno.
• Temps are a bit warm (recorded it earlier today when it was upwards of 80+F in my room)
• And I forgot to change the bitrate up in OBS from 3000 so the quality isn't the greatest. Sorry

[zappian]

What is the point of buying the new gpus opposed to getting a cheaper / used 290x ( or equivalent card).

 

Sigh.

 

The only non rebranded cards are the flagships .

 

The same thing happened with the 2xx series , BUT in the 2xx series they lowered the price somewhat and made it somewhat compelling .

[stconquest]

What is the point of buying the new gpus opposed to getting a cheaper / used 290x ( or equivalent card).

 

Sigh.

 

The only non rebranded cards are the flagships .

 

The same thing happened with the 2xx series , BUT in the 2xx series they lowered the price somewhat and made it somewhat compelling .

 

The 300s will drop in price too, just give it time.

[MEC-777]

Have some relevant information to share...

 

I successfully flashed my 290 with a 390x BIOS that was modded to work with 4GB Vram (instead of 8). It was also modded with the lower default memory clock of 1250. Also installed the 15.15 beta drivers. 

 

At stock clocks of 1050/1250, I have not run into any stability issues. In fact, what I have found is that it is performing better at lower relative clocks compared to the 15.5 drivers and the original 290 BIOS. 

 

Here's evidence of this with Firestrike scores:

 

Max OC with 290 BIOS with 15.5 drivers and clocks at 1200/1600: score of 10135 http://www.3dmark.com/fs/5154393

 

Stable OC with 390x modded BIOS with 15.15 drivers and clocks at 1150/1500: score of 10134 http://www.3dmark.com/fs/5161600 - same score at lower clocks.

 

And finally, the max OC I could run with the 390x modded BIOS with 15.15 drivers was with clocks at 1150/1575: score of 10263 http://www.3dmark.com/fs/5162408 - higher score with still lower clocks vs 290 BIOS and drivers.

 

I could not OC any higher than that with the 390x BIOS and drivers as it would result in instability and crashes. However, at 1100/1500 with +0mV, it's totally stable in benchmarks and in-game. 

 

Not only did I observe gains in Firestrike but also in TW3 (the only game I've had time to test thus far, but it's one of the most demanding). I'm running it at 1440p with VSR, high graphics preset with AA and 32x tessellation override in AMD overdrive with hairworks off. On average I've noticed about a 2-4fps improvement (about 5%).

 

On top of that, I've also noticed the card runs cooler as well.

 

So this would seem to confirm that the majority of the changes from 290/x to 390/x is in the BIOS and drivers. The GPU dies are most likely binned to run at 1050 stock and the memory modules are most likely just better quality parts, able to sustain 1500 stable.

 

Make of this what you will. I'm happy that I was able to, essentially, turn my 290 into a 390 (save for the extra 4GB Vram, which I don't need anyways) for $0. But this looks quite bad for AMD. There are those among the tech community with the knowledge to figure these things out now-a-days. Not much can be kept hidden and this is becoming quite obvious - what they changed from the 290's to the 390's. 

 

There will come a day when the 200 series cards become hard to find, but until that day comes, you'll still be better off getting a killer deal on one of them instead of the equivalent 300 series cards. 

[Notional]

Have some relevant information to share...

 

I successfully flashed my 290 with a 390x BIOS that was modded to work with 4GB Vram (instead of 8). It was also modded with the lower default memory clock of 1250. Also installed the 15.15 beta drivers. 

 

At stock clocks of 1050/1250, I have not run into any stability issues. In fact, what I have found is that it is performing better at lower relative clocks compared to the 15.5 drivers and the original 290 BIOS. 

 

Here's evidence of this with Firestrike scores:

 

Max OC with 290 BIOS with 15.5 drivers and clocks at 1200/1600: score of 10135 http://www.3dmark.com/fs/5154393

 

Stable OC with 390x modded BIOS with 15.15 drivers and clocks at 1150/1500: score of 10134 http://www.3dmark.com/fs/5161600 - same score at lower clocks.

 

And finally, the max OC I could run with the 390x modded BIOS with 15.15 drivers was with clocks at 1150/1575: score of 10263 http://www.3dmark.com/fs/5162408 - higher score with still lower clocks vs 290 BIOS and drivers.

 

I could not OC any higher than that with the 390x BIOS and drivers as it would result in instability and crashes. However, at 1100/1500 with +0mV, it's totally stable in benchmarks and in-game. 

 

Not only did I observe gains in Firestrike but also in TW3 (the only game I've had time to test thus far, but it's one of the most demanding). I'm running it at 1440p with VSR, high graphics preset with AA and 32x tessellation override in AMD overdrive with hairworks off. On average I've noticed about a 2-4fps improvement (about 5%).

 

What if you run the modded 290x with the 15.5 beta and not the newest 15.15 one? Any difference then?

[MEC-777]

What if you run the modded 290x with the 15.5 beta and not the newest 15.15 one? Any difference then?

 

I think you meant "modded 390x" ?

 

I didn't try that, but I could. Simple enough to uninstall/re-install the drivers. Will give it a shot tomorrow after work and report back with the results. I'll also run some gaming benchmarks this time too to give more relevant data. Firestrike scores don't necessarily translate to gaming performance.

[xAcid9]

I think you meant "modded 390x" ?

 

I didn't try that, but I could. Simple enough to uninstall/re-install the drivers. Will give it a shot tomorrow after work and report back with the results. I'll also run some gaming benchmarks this time too to give more relevant data. Firestrike scores don't necessarily translate to gaming performance.

 

Care to revert the bios mod, install this driver and then redo the test?

 

It's essentially the same driver as 15.15 but with latest CFX profile. Modded by asder with support for all cards since HD5000 and signed by leshcatlabs so no need to enable driver test signing.

[MEC-777]

Care to revert the bios mod, install this driver and then redo the test?

 

It's essentially the same driver as 15.15 but with latest CFX profile. Modded by asder with support for all cards since HD5000 and signed by leshcatlabs so no need to enable driver test signing.

-snip-

 

You mean switch back to the 290 BIOS and install that driver? I could. Only thing is I like this 390x BIOS because the card runs a bit cooler and uses a bit less power.

 

With the additional CFX profile, it won't really make a difference in my case because I'm only running a single card. Unless there have been other improvements made as well?

 

Also, I don't know what enabling the driver test signing does. I never had to do that (nor know how to do it) with installing the 15.15 drivers. 

[xAcid9]

You mean switch back to the 290 BIOS and install that driver? I could. Only thing is I like this 390x BIOS because the card runs a bit cooler and uses a bit less power.

 

With the additional CFX profile, it won't really make a difference in my case because I'm only running a single card. Unless there have been other improvements made as well?

 

Also, I don't know what enabling the driver test signing does. I never had to do that (nor know how to do it) with installing the 15.15 drivers. 

I see.   Catalyst 15.15 only supported 300 series now right?

Btw did you check your VRM temp? Guys at Guru3D reported lower core temp but significantly higher VRM temp.

[MEC-777]

I see.   Catalyst 15.15 only supported 300 series now right?

Btw did you check your VRM temp? Guys at Guru3D reported lower core temp but significantly higher VRM temp.

 

15.15 is currently the only driver AMD has available for the R9 300 series as of right now (on their website). 15.20 might be the next iteration of that, but unless there have been other changes aside from added CFX profiles, I'll just stick with 15.15 for now.

 

VRM temps are about the same as before. My 290 is cooled by a Kraken G10 with H55 AIO, so my VRMs don't get any higher than the 70's at most (has a 92mm fan blowing directly on the VRM heatsink).