CF works much better than SLI these days, with far more games supporting it.Crossfire and "works well", "oh my"
Look at the list of 2016 SLI games and the one for CF. 3 game for SLI, a whooping 20 games for CF. That includes all the main AAA titles
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Poll: ** The AMD VEGA Thread **
- Thread starter TNA
- Start date
- Status
Look at the list of 2016 SLI games and the one for CF. 3 game for SLI, a whooping 20 games for CF. That includes all the main AAA titles
You think running a GPU at 100 percent load for months on end doesn't stress it? Especially when it's stuffed in amongst a load of other cards and wasn't designed for that kind of life.
Heat cycles are only one aspect of wear and tear on electronics.
Can you quantify the amount that power/heat cycling reduces the lifespan of the product?
From what I've read, constant heat will reduce a product's lifespan by a factor of 2 for every +10 degrees hotter it runs. Mining cards will be constantly 20-40 degrees (more?) hotter than gaming cards which will be idling most of the time.
That means the power/heat cycling would have to reduce longevity by a factor of 4 to be comparable.
What I see is a computing card running with the Fiji to Vega modded drivers at 1440Mhz speed , providing the same performance as the GTX1080 @ 1785 with it's optimised gaming drivers......
In addition, the CPU per CORE overhead is way less. Look at how some cores are running at 90% on the GTX1080 while same scene same fps, not a single one goes over 50% on Vega.
Many of the scenes the 7700K @ 4.5 is running at 17% overall on the Vega GPU compared to 30% on the GTX1080/1080Ti (with many cores hitting even 90%). In addition it consumes 1GB less VRAM.
I would say, jump on the pre-order train before miners grab them all. Especially Ryzen owners going to see a lot of performance increase
How come the Vega memory frequency is so low compared to the 1080's?
Different memory technologies and ways of measuring the frequency.
Because it's HBM2 memory. It has a super-wide bus (2048 bits) and runs at a low clock.
GDDR runs at higher clocks but has a much smaller bus (typically 192 to 384 bits)
Nobody should jump on any train until reviews are in, unless they have money to gamble and lose.What I see is a computing card running with the Fiji to Vega modded drivers at 1440Mhz speed , providing the same performance as the GTX1080 @ 1785 with it's optimised gaming drivers......
In addition, the CPU per CORE overhead is way less. Look at how some cores are running at 90% on the GTX1080 while same scene same fps, not a single one goes over 50% on Vega.
Many of the scenes the 7700K @ 4.5 is running at 17% overall on the Vega GPU compared to 30% on the GTX1080/1080Ti (with many cores hitting even 90%). In addition it consumes 1GB less VRAM.
I would say, jump on the pre-order train before miners grab them all. Especially Ryzen owners going to see a lot of performance increase
The idea that AMD are still sandbagging is getting more and more absurd by the day.
There's also zero evidence that Vega will work better with Ryzen at this time.
Would be indeed stupid for AMD to deliberately slow Vega's performance on Intel platform considering big majority of gaming PCs have Intel CPU and majority of reviewers will no doubt use Intel CPU.
HBM isn't exactly new.D'Oh! Completely forgot it uses the new memory
HBM1 just didn't become common because of small (1GB) capacity per stack...
And now getting mass produced HBM2 out hasn't proceeded at that good pace.
http://www.anandtech.com/show/11690/sk-hynix-customers-willing-to-pay-more-for-hbm2-memory
I do not see the AMD slowing down the Intel performance with the Vega drivers. On the contrary I see the Nvidia drivers slowing down Ryzen & SkylakeX performance.
Something many are saying months now.
Last edited:
Soldato
- Joined
- 18 May 2010
- Posts
- 23,648
- Location
- London
I posted this earlier but it seems to be more relevant to the current topic of conversation.
Soldato
- Joined
- 17 Jun 2012
- Posts
- 5,951
I mine and they're nowhere near as stressed as gaming, and I'm not throwing money away because the hash rate is the same at 55-65% TDP as it is at 120% TDP. All that matters is the memory overclock.
Like I said, 100w per card, 60% TDP, 1300-1400mhz core clock vs 120% and 2000-2100 core, they're barely warm at between 52-60c temp wise compared with 80c for gaming. Some serious mining myths here.
This is the thing I see. Nvidia basically use CPU power to do what they didn't want to do on the GPU. They saved themselves time and money and cost of NOT putting the scheduler on the card. I.e. they cheaped out, but gained fps by doing it. It feels like a dirty cheat, but the dirty cheat worked, so who can really complain.
You forgot power efficiency. By removing transistors for hardware schedulers you gain in power efficiency.
I wouldn't say they cheaped out because now they need to spend time writing drivers to ensure that their cards are competitive. But it does mean that EOL cards aren't worth writing optimisation code for
.Nah most of the hardware scheduling is still there - some bits have been moved to software "assistance" - what nVidia is doing is actually a lot more complicated and while somewhat more reliant on CPU it provides the same and potential for better GPU performance (as long as the CPU it upto it). There is a lot of misunderstanding about what nVidia is doing and people are quick to assume and criticise it.
Correct, not relevant to what was being discussed, but correct
If you're actually using some type of card that doesn't take any hashrate hit from lowering performance that that's cool, but it's an anomaly and by far the exception to the rule.
They didn't cheap out and its not a dirty trick.
Nvidia used to have the entire scheduling pipeline in hardware with Fermi, but hardware based solutions lack a lot of flexibility and can be very complex. With Kepler they moved a small amount of that out to the driver stack, which made the hardware much more efficient. The task scheduling is still done by the GPU in hardware but the drivers can influence more control to better load the hardware
Actually it's standard mining practice - despite the many pages of rhetoric from people who don't know about mining talking like experts on the matter.
So you're putting a 24/7 overclock on the memory.
Sure you might not be stressing the core, but you sure as hell are stressing the memory.
In the ex-mining card that I had, it was indeed the memory that failed, a few months after purchase. Anecdotal and not proof of anything, of course. But you can't argue that a 24/7 overclock of your memory is less stressful for that subsystem than periodic gaming.
- Status