radeon 7 is also not good example because of massive bandwidth1,024 GB/s, generally 9.2tf navi would be aroud 1.8-2x faster than 6tf xbox one x gpu in games (xbox one x gpu is quite efficient as it's also not wide).
-
Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
-
We have made minor adjustments to how the search bar works on ResetEra. You can read about the changes here.
Next-gen PS5 and next Xbox speculation launch thread |OT6| - Mostly Fan Noise and Hot Air
- Thread starter Mecha Meister
- Start date
- Status
If we are comparing flops vs flops we should be removing bottlenecks. If we have a test to confirm the gigabytes / teraflops that Navi needs. Probably around 50GB/s per teraflop.
5700xt would be probably faster in 4k with 1,024 GB/s bandwidth than with 448 GB/s
Found the IPC test DrKeo
www.computerbase.de
This test is comparing between the rx 5700 and the rx 590 clocked at 1500MHz. Both are 36 CU GPUs so they have same TF at this state with the same design setup, and the rx 5700 performs 39% better on average.
AMD Radeon RX 5700 und RX 5700 XT im Test: Architektur-Vergleich, Preis-Leistung, Custom Designs (Update)
AMD Radeon RX 5700 (XT) im Test: Architektur-Vergleich, Preis-Leistung, Custom Designs (Update) / Navi vs. Vega vs. Turing vs. Pascal
www.computerbase.de
Seeing the position of the GTX 1080 Ti in charts is still incredibly impressive, it still ranks among the top performing cards today. Nvidia released a real monster back in early 2017.5700XT is actually 19% faster in 1440p and 18% faster in 4K. So 1.18x 12.5TF is 14.75TF.
In overall systems design, isn't the Xbonex bottlenecked by the weak CPU? shouldn't that impact overall game performance?
probably is, we are talking about gpu only now, cpu will be around 4x faster
Ah ok that one. Some folks were not happy that I posted it here because they think it is incorect regarding the comparison to the Nvidia architectures. So be aware to use it only for AMD comparisons.
Don't know it was posted:
I am only using the Polaris comparison, all the others don't mean much for me personally.Ah ok that one. Some folks were not happy that I posted it here because they think it is incorect regarding the comparison to the Nvidia architectures. So be aware to use it only for AMD comparisons.
It's wrong comparison as bandwidth isn't the same.Found the IPC test DrKeo
This test is comparing between the rx 5700 and the rx 590 clocked at 1500MHz. Both are 36 CU GPUs so they have same TF at this state with the same design setup, and the rx 5700 performs 39% better on average.AMD Radeon RX 5700 und RX 5700 XT im Test: Architektur-Vergleich, Preis-Leistung, Custom Designs (Update)
AMD Radeon RX 5700 (XT) im Test: Architektur-Vergleich, Preis-Leistung, Custom Designs (Update) / Navi vs. Vega vs. Turing vs. Pascal
Is it though? We are getting the same bandwidth increase on the consoles.
It means 9.2 tflops (36 CUs navi) are performing the same as 12.8 tflops (compared to polaris with 36CUs). So more than 2.13x XBX performance.Found the IPC test DrKeo
This test is comparing between the rx 5700 and the rx 590 clocked at 1500MHz. Both are 36 CU GPUs so they have same TF at this state with the same design setup, and the rx 5700 performs 39% better on average.AMD Radeon RX 5700 und RX 5700 XT im Test: Architektur-Vergleich, Preis-Leistung, Custom Designs (Update)
AMD Radeon RX 5700 (XT) im Test: Architektur-Vergleich, Preis-Leistung, Custom Designs (Update) / Navi vs. Vega vs. Turing vs. Pascal
But the most important is that 5700 (36CUs) tflops is on average 1% more efficient than turing tflops of nvidia 2070 (36 CUs) using the same clocks. So a 5700 at 2ghz (rumored PS5 specs) is performing the same as a 2070 running at 2ghz.
So PS5 GPU = 5700 XT = 2070
I would have expected nvidia 2080 performance, but this is not bad at all.
1500MHz is between its base and boost frequence, same for 5700
So when comparing Navi to GCN (Polaris which is in the PS4pro and XB1X) flops we should be multiplying by 1.39?Found the IPC test DrKeo
This test is comparing between the rx 5700 and the rx 590 clocked at 1500MHz. Both are 36 CU GPUs so they have same TF at this state with the same design setup, and the rx 5700 performs 39% better on average.AMD Radeon RX 5700 und RX 5700 XT im Test: Architektur-Vergleich, Preis-Leistung, Custom Designs (Update)
AMD Radeon RX 5700 (XT) im Test: Architektur-Vergleich, Preis-Leistung, Custom Designs (Update) / Navi vs. Vega vs. Turing vs. Pascal
That means a 9.2TF PS5 is equivalent to a 12.7TF Polaris?
considering how long away the next GTA is, this probably holds true for them!
maybe I read this test wrongly but I see 1.55x bandwidth advantage for 5700
yea, that is what i think will be the closest that we can get to calculating the generational leap of the GPU as of right now.
however, with the help of rapid packed math and VRS etc. there are also more boosts to take in for next gen. i assume rapid pack
Yea, also will happen between PS5 and ps4 pro
But it's not proper arch comparison lol, it's abvious that at some point you will be bandwidth limited and have to increase it
Its the closest approximation we can get right now, at least its much better than comparing to the vega 64
It is better I agree but its too optimistic for navi, I will write sumup in next post how we can compare 9.2tf navi to vega 64, vega 56, radeon 7 and radeon 580 in next post (will take into account 4k results of techpowerup)
According to 4k performance summary from techpowerup 9.2tf navi (36 or 40cu would be quite simillar) is eqiuvalent to:
14.6tf vega64
13.4tf vega56
11.87tf radeon vii
11.52tf rx 580
14.6tf vega64
13.4tf vega56
11.87tf radeon vii
11.52tf rx 580
Last edited:
You're making the mistake of thinking that performance scales linearly with Tflops. There's no such thing as a general performance/Tflops for GCN because different GCN card all have different performance/Tflops, even cards from the same architecture (ie Vega 56 vs Vega 64). Generally the more CUs you have the harder you drop off in performance/Tflops.
Look at the 580 (=X1X) vs Vega 64: 6 Tflops vs 12.6 Tflops but a Vega 64 only performs 65% better than a 580 and isn't actually twice as fast. The 5700XT isn't even twice as fast as the 580. It almost is though.
No, because we have no idea how a 12.7 Tflops Polaris card would perform. If it'd drop off as hard as Vega then a 9.2Tf Navi card would wipe the floor with it. Most likely it would drop even harder because Vega is better than Polaris at the Tflops as the test also shows.
The only thing you can say from that test is that a 36CU Navi card at 1500mhz is 39% better than a 36CU Polaris card at 1500mhz. Once you change the clocks & CUs this percentage changes.
Again this assumes that performance for Polaris wouldn't drop off hard at higher CUs & clocks and I wouldn't make this assumption.
Assumption is that you achieve flops only by chaning clock and that bandwidth is not a limit
That dropoff you are talking about isnt a problem if PS5 really is 36CU, as the only source for the increased performance would be higher clock speed and IPC, 2 things that do scale practically linearily
I'm curious, would most of you prefer native 1800p or checkerboarded 4K assuming the PS5 GPU is 9.2 Navi TFlops? While clearly very powerful in its own right, performance could be saved for other areas by rendering in sub native 4K
Anything but native 4K. CB or other forms of temporal reconstruction, dynamic res, whatever. Just don't fucking waste a lot of power chasing the absolutely marginal IQ gains native has.
Shipping a 2GHz Navi in a space/thermally limited box is a feat of engineering. I could see 36CU @ 2GHz with the extra die space going towards hardware RT acceleration being a good trade-off. It would also fit nicely with the lower CPU cache and HBM2 rumors. They could very well be squeezing every MHz from Navi like that.
Hmm, I've yet to actually see checkerboard rendering in action because I don't own a PS4 Pro or Xbox One X, but does it really deliver image quality close to native? Furthermore, it's usually quite evident for me when a game runs at 900p instead of native 1080p, but I guess that's less of an issue with much higher resolutions.
It looks much better than 1080p or even 1440p and it costs around the same as 1440p. 4K comparison isn't really fair because it takes significantly less resources than native resolution.
What he said. Colbert is already skeptical about just 36 CUs @2GHz due to the sheer amount of power it should take up.
Impressive, although I wonder how it holds up against a native 1800p framebuffer. Either way, next gen should be similar to the 7th gen where almost every game run subnative of the advertised resolution.
Gah.... This is not my intention. I'm literally just parroting stuff here. >°<
this is a comparison shot for horizon between native 4k with a lot of supersampling, checkerboard 4k and native 1512p (highest they could push on pro with a native resolution)
Gah, I'll have to look closely later, on mobile now and tiny screen won't make much difference, if any at allthis is a comparison shot for horizon between native 4k with a lot of supersampling, checkerboard 4k and native 1512p (highest they could push on pro with a native resolution)
Which was the whole purpose of this stunt: Attention seeking! Disgusting if you take into account what Putty wrote about his experience with him just recently! Sorry, my honest opinion!
Last edited:
This is not accurate. Since when a 2070 is almost as powerful as a 2080 ? Why spouting false information when you have all the accurate info (real benchmarks) on hand here and elsewhere ? thuway , I am disappointed.
- Status