I made a BIOS with just synced xbar and L2 and saw 0 difference in performance at 1531MHz synced vs default. Applying the power limit/voltage increases did help, however- the core speed maintains the max boost clock when I max out the memory oc, which slightly improved scores. In my case, I...
True. The reference BIOS sets the temperature target to 80C, meaning you start throttling at that point. The reference fan curve is geared for low-noise, so it's pretty easy to each the 80C limit. Precision or Afterburner let you adjust both of those, or you can get a custom BIOS with the...
I'll try again- SLI and crossfire will not tell you any more information over a single card. If you say scaling is 95% for SLI, then we don't need to run the experiment. Each card contributes the same amount of FPS, and you scale it down by that 5% of the scaling factor. Done.
Each card has...
SLI and Crossfire use alternate frame rendering right? That means per-frame, you only get 1 card's cores and VRAM. Neither the cores nor the VRAM are additive across multi-GPU setups; comparing two 8GB 290X's is not equivalent to testing a theoretical 8GB card with twice the # cores.
There's...
NVidia's board partners used to make versions with double the stock memory, but I don't know if they did that for the 780's. I don't see why they won't do it again, but it might be a couple more months.
I think Marcdaddy pretty much hit the nail on the head- by the time you really need more...
Hexus looked at the new 8GB card from AMD on Shadows of Mordor, for those interested in whether 4GB is too little for NVidia's cards:
http://hexus.net/tech/reviews/graphics/76685-sapphire-radeon-r9-290x-vapor-x-8gb/?page=9
I'm also unsure what the impact of DSR is. From NVidia's description, it is doing real 4k (or w/e you choose), plus an additional downsampling step at the end of each completed frame. That's the best I can do in terms of testing at higher than 1080p. And Crysis 3 is just one game, there are...
Firestrike was 70pts more on the gfx subscore for me, a 0.5% increase. Nearly as much performance as adding 13Mhz on the core.
Core clock is of course the first thing to do since you benefit across the board (both geometry, pixel, and compute shader horsepower). But the story doesn't change...
I haven't encountered any cases where 1% core speed increase is better than 14% mem increase, even at my native 1080p. In order for 15MHz to get you 5-6 more fps, you'd have to be getting 500fps already.
Overclocking the RAM gets you about the same perf increase regardless of the resolution. It's better to sacrifice one or two core clock bins if you can get the mem to 8000MHz or more.
!!!
I wouldn't wait on 20nm, I think we're going straight to 16nm. TSMC only has a cell phone SOC version of 20nm, and all that capacity is currently devoted to cell phones. For graphics, you're potentially looking at something marginally better than 28nm for a significantly higher cost...NVidia...
Increased memory from +750 to +797, got some more points. +829 complete also, but was slightly slower.
http://www.3dmark.com/3dm/4361814
I'm 85th in the hall of fame: http://www.3dmark.com/hall-of-fame-2/fire+strike+3dmark+score+performance+preset/version+1.1/1+gpu
lol
Thank you.
In other words, unless you have an identical CPU setup, then you can't even ask the question "Why are my identically overclocked GPUs giving me a crappier graphics subscore?" We all know the overall score will differ.
Turns out that the graphics score improves with more powerful CPUs.
When I further overclocked my 4770k from 4.7GHz to 4.8GHz, I went from 15955 on the graphics subscore to my current best of 16162. Suicide run, in the sense that I get an artifact here and there. Rarely crashes though...
Core speed is more important, but it's not worth sacrificing a ton of mem speed just for one or two core clock bins. In each frame of a game, there are always regions where the cores are waiting on memory. I would expect the speedup of memory oc to be the same % no matter the resolution. In my...
Last night I set out to answer three questions:
1. What effect does voltage have on the realized boost clock?
2. What effect does memory oc have on the realized boost clock?
3. What combination yields the best peformance?
I ran Heaven for a couple minutes on each setting, and then quickly...
As a matter of fact...I have a lot of data points. Regarding artifacting, +797 has major artifacts, +702 is occasional, +654 is rare, +591 is good to go.
Firestrike
+152MHz +0MHz @ 125/86 +87mV: 13910
+139MHz +654MHz @ 125/86 +87mV: 15941
+139MHz +591MHz @ 125/86 +87mV: 15918
+139MHz...
20nm won't happen for high end graphics cards, I'm betting on a new $700 price range card next year and a maxwell 2 refresh for the mid and lower GPUs. As far as the $500 range goes, I don't think we'll see any significant performance difference for at least a year. And the price for 28nm parts...
Good article Brent, it must've been a painstaking process to find the clocks for all those cards.
I'm curious if Brent's findings are similar to other 980 owners out there- mine definitely doesn't behave like Brent's. Before we get into this, I have the evga sc edition, so it could all be due...
You could try precision, but none of these programs inspire confidence at the moment. If I leave precision running regardless of OSD on/off, Valley and Heaven hitch every second or two. It drops the min FPS score to 20, when it's normally 40! I haven't had issues with settings not sticking...
TDP means temperature to me, and I solved that with a custom fan profile. It's running out of power budget- it's at 122% power most of the time in the OSD, with max reported as 125%.
Here's what I'm getting at 1493MHz to 1519MHz / 8192MHz.
GPU-Z
Firestrike
Valley
Heaven...
+87mV isn't getting me a whole lot on my 980 SC. It lets me benchmark at +139, or makes +127 stable. With 0mV, +127 works for quite a while before crashing but +139 won't work at all. If I drop my memory back to stock and add 87mV, +152 will run with bad artifacting for a short time before...
Gotcha. It would be lots of fun if we could push it to the thermal limit. I'm seeing artifacts at the max clock I can get with +87mV, so the power limit probably isn't an issue for me.
Maxing TDP on your CPU or your 980s?
Sort of related:
http://www.pcper.com/reviews/Graphics-Cards/EVGA-Hydro-Copper-GTX-980-Water-Block-Early-Performance-Testing
Definitely true for memory, and in my case, also true for core speed. Although I can complete benchmarks at +267MHz core, the scores are lower, and there are artifacts.
980 owners- what oc are you able to actually game at? For Crysis 3, I'm using:
+242MHz core (factory overclock +127). I'm...
When I apply +87mV, GPU voltage is 1.250v often, where before it would be 1.2v with jumps to 1.225v. I don't think the Precision OSD is very...precise. The overvoltage is definitely working, I can't run anything at my card's max offsets with 0-50mV, but no problem at 87mV.
I looked at all the...
Looks like my 980 really does need the overvoltage to max out. With +152MHz on the "superclocked" default of 1241MHz (so +267MHz from reference cards?), it can't complete Firestrike without the +87mV. With the overvoltage, and no memory overclock, I'm seeing 1544MHz most of the time in the OSD...
Two bad GPUs would be pretty unlikely. That new 344.16 driver fixed some motherboard incompatibilities- that could mean more incompatibilities remain, and your setup is one of them.
I'd consider returning the cards or getting support from the vender, see if they've run into this.
My card maxes out at 1493-1531 boost range, I only got one bin farther with the 87mV bump. That's just for benchmark-stable, although Crysis 3 was fine for the half hour I played. 4112MHz for memory seems to be fine.
With that, Firestrike is at 13535 overall, Graphics Score 15955. I can't quite...
I could see a Ti version, of course at the Ti price point. Personally I look for 2x performance for ~$500.
I'm more curious how the lineup will look 6 months from now. If they produce a 20nm part, will it actually have good dollar value? 20nm is looking pricey, so I'm not sure we'll see a big...
Same here, only had a little time to play with it. I made some good progress on finding the max OC without overvoltage. I have the eVGA SC (1241MHz base clock), so my offsets will differ.
The best I got on Firestrike was 13349 overall, Graphics Score 15738, Physics Score 13243, Combined Score...
The cache needs more voltage to be at the same speed as the core, in general. I'm at 1.275v on both core and cache, but my cache is running 200MHz less than the core.
And if you have any stability issues, I would start by dropping the cache to 45. In fact, you're better off running cache at...
This one? http://hothardware.com/Reviews/Intel-Core-i74690X-Extreme-Edition-Ivy-Bridge-E-CPU-Review/?page=9
Personally I don't think running games like that at low resolution is useful. The picture changes too much when you're at 1080p or more with high settings. Bottom line- find the best...
Try changing your CPU speed, and then try slowing RAM down. I'd use a single GPU, since running two just taxes your CPU and memory unnecessarily. If I have Warhead on my haswell's drive I can find that spot and check FPS.