Soldato
I would love it is the forums would stop going down...
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.
Alright, basic gist, makes the new shared L1 in RDNA more effective (higher hit rates) by dynamically clustering CUs together, to pool their cache lines, to minimize duplication of cache use (+effective capacity) and increase hit rate. This reduces pressure on LLC/L2.
This makes the L2 more effective, thereby reducing pressure on memory bandwidth. This kind of design is enabled by the change in RDNA 1, with a 128KB shared L1 for the CUs within a shader array.
I still think the 128KB is too small for so many CUs, so one of the key changes I expect to see in RDNA 2 is at least 2x L1 size. Combine with this dynamic L1 changes, would be much more potent cache system, both L1 & L2 effective capacity & hit rate up.
Think of it as a multiplier for L1 & L2 cache, less duplication makes it "effectively bigger" & increasing hit rate makes it more efficient. Both of these will impact memory bandwidth efficiency (improve it). I like seeing these types of uarch changes, work smarter, not harder.
Section 5.2
DynEB enhances IPC by 22% on average over the private baseline. (Graph shows a max of just below 60% increase in IPC).
Overall, DynEB improves performance of all evaluated applications by 9%.
Therefore, DynEB improves performance-per-watt by 9% and the energy efficiency (performance-per-energy) by 20%, on average across all evaluated applications. For the shared-friendly applications, DynEB maintains the total power consumption (similar to baseline) and saves energy by 18%. Therefore, DynEB enhances performance-per-watt and energy efficiency for the shared-friendly applications by 22% and 49%, respectively.
Section 5.3
Effect of L2 Cache Size.We evaluate a boosted private L1 baseline with double the L2 cache size. We observe almost no performance improvement for the shared-friendly applications compared to the baseline. This is because performance is limited by the L2 reply bandwidth bottleneck [49,73,74]. Such a bottleneck is relieved with Shared++ and DynEB as the shared L1 organization utilizes the remote cores as an additional source of bandwidth.
We observe higher IPC improvement under increased core count because the overall L1 capacity increases with more cores, thus the available collective L1 bandwidth increases under shared L1 organization.
I would love it if we beat them too. I would also love it if the early adopters of the new nvidia cards regret their purchase
I would also love it if the early adopters of the new nvidia cards regret their purchase
All 3 of them who actually got a card?
Gibbo said 1,000 flew off the shelves
thats ur assumption, I have an idea what the performance is you can’t tell me what I do and don’t know what sort of non-sense is that. Anyone can easily estimate the performance of RDNA2 in a best case scenario and work back there is enough data out there already for that. All we lack really are the die sizes and CU counts at this pointYes performance is important but you have no idea how efficient an RDNA2 CU is. The number is just a number to us currently. 60CUs might be enough or it might not be but why does that number matter so much as long as performance is there?
I hope AMD BigNavi can compete against the rtx 3080 in 4k @ 60fps. I don't really care much for ray tracing, nothing against those who do. I'd just like a card that can handle higher game settings at 4k 60fps.
So, I'm prepared to wait and see what 6900 brings to the table. AMD have the better fab and node than Nvidia's rtx 3000 series.
Im going to hold off for AMD. Another month or so which is frustrating but will wait after the NV farce. AMD really need to get these chips baking and have oodles of stock. They could have a win here. Play it right AMD
I dont think we will hear much from AMD until the RTX 3070 and 3060 are launched. They are stalling / waiting at the moment I think.
They say they have their own schedule, I dont think that reflects what we are seeing, they are reacting to Nvidia mostly (and preparing for the console launches too).
AMD have no intention to react to Nvidia, it was clear before Ampere AMD had no intention to get in before or around the same time as Nvidia and they are still in no rush, just saying "we are on our own release schedule"
I don't know what to make of that, since the Ampere kitchen reveal the 80 CU Navi seems to be no more.
The 2080TI is 35% faster than a 5700XT at 1440P, i'm not going to use 4K for this given that every 8GB card falls further behind the 2080TI, they are not 4K cards.
But at 4K compared to the 3080 it is 32% faster than the 2080TI.
2080TI vs 5700XT 135%
3080 vs 2080TI = 132% (@4K)
Not so Big Navi is rumoured to have 60 CU's, that's 50% more than the 5700XT, after scaling make that +45%
5700XT +45% = 145%
145 / 135 = 1.074. so normalized: 60 CU Not so Big Navi vs 2080TI = 107%
3080 vs 60 CU Not so Big Navi: 132 / 107 = 1.23 (+23%)
RDNA1 vs RDNA2 +10% IPC = 60 CU RDNA2 @1900Mhz. 3080 vs 60 CU RDNA2 @ 1900Mhz: 132 / 117 = 1.128 (+13%)
The PS5 clocks to 2.23Ghz, let's assume that's the limit of RDNA2: 2230 / 1900 = 1.173 (+17%)
Now the 6800XT is 34% faster than a 2080TI, 2% faster than the 3080.
Moores Law thinks AMD think they can compete with the 3080 without the 80 CU Big Navi. i think he's right. With the enhancements of the improved 7nm node and improvements to the RDNA architecture AMD could make a still small 350mm2 GPU at around <2.3Ghz, 250 Watts, 16GB GDDR6 and at least trade blows with the 3080 at £150 less and still get good margins.
I think its a shame if AMD don't slam an 80 CU RDNA2 GPU in Nvidia's face anyway but i think AMD were expecting more from Ampere, they are not impressed.
https://www.techspot.com/review/2099-geforce-rtx-3080/
I wirk at a PC shop who specialises in custom builds, gaming pc mainly. Not on the same scale as OcUK but a decent size.Gibbo said 1,000 flew off the shelves