wouldnt be the worst thing their marketing dept's come out with.
-
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.
AMD Navi 23 ‘NVIDIA Killer’ GPU Rumored to Support Hardware Ray Tracing, Coming Next Year
- Thread starter Gregster
- Start date
- Status
wouldnt be the worst thing their marketing dept's come out with.
Soldato
Could be just a 10 second clip of "the" card then it ends. That would be enough to set tongues wagging.
Associate
Is it just me or does anyone else hate this sort of tease. I am on the cusp of spending £650 on your competions product which will mean it will be 2 years at least before I am remotely interested in what you have to offer and the best you can come up with is some boring fortnite tease and a song about tomorrow which may actually not mean tomorrow but the next day.
Some free advice for you AMD. Stop messing about with this pointless and annoying marketing BS. Just tell us you are having a press conference / release on this date at this time, see you then.
This sort of BS that you are doing now just makes me want to buy a 3080 out of spite.
Rant over.
Some free advice for you AMD. Stop messing about with this pointless and annoying marketing BS. Just tell us you are having a press conference / release on this date at this time, see you then.
This sort of BS that you are doing now just makes me want to buy a 3080 out of spite.
Rant over.
Soldato
Yes I agree, I hate this childish game playing crap.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 29,803
It's simply not that hard to issue/leak some slides and numbers. Otherwise it will be simply too late.
If there's a market for 1m of top tier cards worldwide (Total guess) the that number starts dropping really quickly from the 17th
Could it be that AMD are simply a long way from a functioning card ?
Associate
- Joined
- 28 Sep 2018
- Posts
- 2,259
In reality, they can't do much to stop nvidia's momentum. Nvidia has marketshare and mindshare. AMD on the GPU side is well on the back foot to draw attention away.
I do agree, set the date, announce your products professionally and let them speak for themselves. Part of what you're seeing is buyers anxiety. When people are spending a notable figure like 650+ they want to know if they're spending that money and getting the best product they can for it. This is a lot easier to do when all your options are infront of you.
honestly who knows at this point, IF they have a full product stack ready to go on rdna2 for the pc im shocked at the total lack of leaks already.
if they are months off so the new year from a functioning card god help them.
or a third way, they have a handful of units of varying performances and where waiting for nvidia to release specs and are now scrambling to figure out what sort of product stack to release which could be good in having slightly faster cards than the respective nvidia ones (up to a point) but still means its next year for any to hit stores. and by then we could have another round of supers dropped by nvidia.
still say i wouldnt be surprised if there was a non compete from sony and ms over what level cards amd could ship within a month or two either way of the console launches.
Associate
- Joined
- 19 Jun 2017
- Posts
- 1,029
Die size comparisons don't provide a solid basis for comparing architectures.. especially if they are based on different lithographies
- IGP Vega. 8 CUs (1 CU = 64 ALUs. 1 ALU = 2 FP32 ops per CLK). 1.8 TF @ 1.75 GHz CLK
- Tiger Lake IGP. 96 EUs (1 EU = 8 ALUs. 1 ALU = 2 FP32 ops per CLK). 2 TF FP32 @ 1.35 GHz CLK (Real world performance delta is more than TF delta)
- HPG 1 tile. 512 EUs (RX Vega 64 class). 10 TF FP32 @ 1.3 GHz CLK
- HPG 2 tile. 1024 EUs. 21 TF FP32 @ 1.3 GHz CLK
- HPG 4 tile. 2048 EUs. 42 TF FP32 @ 1.3 GHz CLK
Also note that, above stats are for datacenter products with a good amount of EU silicon reserved for FP64 and matrix MADs, so the FP32 perf for gaming products is expected to be much higher, as they can pack more ALUs per EU. The IGP makes me confident that these stats might actually translate to real world performance as they seem to have reached a respectable level of driver maturity.
But, enough of that, I see the hype train is picking speed
I did think the same thing!
Hopefully not.
Associate
It's without a doubt going to be a "coming soon" tease, nothing more. It's going to be generally worthless fodder until or just before 17 Sept from AMD.
I'm trying to catch up on the rumored performance of Big Navi (23/6900) -- what's the consensus? My basic math says, without any gains in core efficiency, with at least 72CUs, we should see performance around 50% better than 5700 XT (likely better with IPC gains). Am I off base here?
I'm trying to catch up on the rumored performance of Big Navi (23/6900) -- what's the consensus? My basic math says, without any gains in core efficiency, with at least 72CUs, we should see performance around 50% better than 5700 XT (likely better with IPC gains). Am I off base here?
Soldato
Die size comparisons don't provide a solid basis for comparing architectures.. especially if they are based on different lithographies
Here are few stats making rounds after Intel hosted its annual architecture day in August. These are stats relevant to HPG (its still a guess extrapolated from datacenter products, close enough for me) based on SuperFin. The HPG parts will be fabricated using TSMC 6nm (rumour, well the TSMC bit is a rumor but Intel has publicly acknowledged that HPG will be contracted to external fabbers)
- IGP Vega. 8 CUs (1 CU = 64 ALUs. 1 ALU = 2 FP32 ops per CLK). 1.8 TF @ 1.75 GHz CLK
- Tiger Lake IGP. 96 EUs (1 EU = 8 ALUs. 1 ALU = 2 FP32 ops per CLK). 2 TF FP32 @ 1.35 GHz CLK (Real world performance delta is more than TF delta)
Currently as things stand the 4 tile version is rated at 500W TDP under SuperFIN, but that might change if they move to TSMC. This is the card from a gamers' wild fantasies and I hope they launch it, maybe clock it lower for all I care.
- HPG 1 tile. 512 EUs (RX Vega 64 class). 10 TF FP32 @ 1.3 GHz CLK
- HPG 2 tile. 1024 EUs. 21 TF FP32 @ 1.3 GHz CLK
- HPG 4 tile. 2048 EUs. 42 TF FP32 @ 1.3 GHz CLK
Also note that, above stats are for datacenter products with a good amount of EU silicon reserved for FP64 and matrix MADs, so the FP32 perf for gaming products is expected to be much higher, as they can pack more ALUs per EU. The IGP makes me confident that these stats might actually translate to real world performance as they seem to have reached a respectable level of driver maturity.
But, enough of that, I see the hype train is picking speed
Again it is relevant. Because again Intel had beaten AMD in IGP performance by using more die area dedicated to the GPU. Then they used eDRAM/L4 cache to supplement main memory.That was 5 years ago.
You are still comparing an AMD SOC with a lower percentage of the die area dedicated to graphics to an Intel SOC with more of the die area to graphics,and one with only 4 cores,which consumes less of the package power in a power limited mobile scenario.
In both cases the design parameters are different and this plays a bigger roll in performance differences.
Renoir has the least percentage area dedicated to the GPU of any AMD APU. It literally has the same Vega design from the previous Zen+ based APUs but has nearly 30% less cores.Only Intel is dedicated more of the die resources towards GPUs this round and AMD has made a decision double the cores is more important than a faster GPU.
GPUs by their very nature achieve much higher transistor densities than CPUs.
You are basically comparing Apples to Oranges and getting Pears. Until intel gets out its own dGPUs,then it's hard to infer much about uarchs. Then the second part of your answer regarding the desktop variants comes into play.
At that point it will be easier to isolate other factors and test in a more Apples to Apples way.
If intel does show good scaling then yes we could be in for some fun as they have access to more advanced packaging technologies than AMD or Nvidia do.
Last edited:
Associate
- Joined
- 19 Jun 2017
- Posts
- 1,029
i think we are talking pears.. i am trying to extrapolate IGP Tiger Lake's performance to the big Intel GPU launch in 2021.. i am just using Vega IGP as a basis for arguing that it looks like a credible entry, their basic building blocks are in place..
i am not comparing Tiger Lake with Ryzen 4800U, and why would i do that in a graphics card thread?
i am not comparing Tiger Lake with Ryzen 4800U, and why would i do that in a graphics card thread?
This was an interersting watch. It seems that having ray tracing in the consoles will help AMD on the desktop since the developers will optimise the raytracing render pipe line.
I hope he does this again once RDNA 2 is out.
I've been saying this for ages. The console development must give them an advantage given the input MS and Sony will have had. I think it's been down played as most people are invested in Nvidia's ecosystem.
Associate
also comes with free bike
- Status