btk2k2's forum posts

Avatar image for btk2k2
btk2k2

440

Forum Posts

0

Wiki Points

1

Followers

Reviews: 0

User Lists: 0

#1 btk2k2
Member since 2003 • 440 Posts

[QUOTE="btk2k2"][QUOTE="ronvalencia"]On the PC,

1. you have to factor in the PC driver issues and AMD GCN has a performance boost with recent drivers

2. you are dealing with legacy programs.

7970 murders 6970 in OpenCL.

OpenCL Compute Performance of AMD Radeon HD 6970 http://clbenchmark.com/device-info.jsp?config=11975403

OpenCL Compute Performance of AMD Radeon HD 7970 / 7970 GHz Edition http://clbenchmark.com/device-info.jsp?config=11905571

In Physics: SPH Fluid Simulation, 7970 / 7970 GHz Edition is about 4.3 times over 6970.

ronvalencia

I am just showing that a 40-50% increase in performance is associated with a generational leap on the PC. It used to be around 100% back in the Radeon 9700Pro days but back then you could increase power along with your process improvements where as now it is purely down to architectural efficiency and process improvements since we are close to the maximum of the ATX spec power for PCI-E cards. In any case that 40-50% difference between Xbox and PS4 will be noticed in games. There is no way around it because the PS4 is easier to program for as well.

Well, I'm showing massive difference for intended workload targets i.e. AMD made large investments for GpGPU workloads.

Xbox camp should be used to heterogeneous memory setup.

The intended workload is games. When comparing a set of fixed hardware where the difference in relative GPU performance is similar to the difference of the Xbox1 GPU compared to the PS4 GPU gives you the approximate level of scaling between the platforms. The rest of the system is close enough for this to be a valid comparison since I do not think the Xbox1 GPU or the PS4 GPU will be starved due to having insufficient bandwidth. This generation it is even easier because both systems have the same CPU and they are using the same GPU architecture. The PS4 will have an average 40-50% performance advantage over the Xbox1, the hardware bares out this fact and comparisons between PC's using GPU's with the same relative performance differences that are built on the same architecture show this to be the case. 50% more shader and 100% more fillrate performance with sufficient bandwidth = 40-50% more performance.
Avatar image for btk2k2
btk2k2

440

Forum Posts

0

Wiki Points

1

Followers

Reviews: 0

User Lists: 0

#2 btk2k2
Member since 2003 • 440 Posts
On the PC,

1. you have to factor in the PC driver issues and AMD GCN has a performance boost with recent drivers

2. you are dealing with legacy programs.

7970 murders 6970 in OpenCL.

OpenCL Compute Performance of AMD Radeon HD 6970 http://clbenchmark.com/device-info.jsp?config=11975403

OpenCL Compute Performance of AMD Radeon HD 7970 / 7970 GHz Edition http://clbenchmark.com/device-info.jsp?config=11905571

In Physics: SPH Fluid Simulation, 7970 / 7970 GHz Edition is about 4.3 times over 6970.

ronvalencia
I am just showing that a 40-50% increase in performance is associated with a generational leap on the PC. It used to be around 100% back in the Radeon 9700Pro days but back then you could increase power along with your process improvements where as now it is purely down to architectural efficiency and process improvements since we are close to the maximum of the ATX spec power for PCI-E cards. In any case that 40-50% difference between Xbox and PS4 will be noticed in games. There is no way around it because the PS4 is easier to program for as well.
Avatar image for btk2k2
btk2k2

440

Forum Posts

0

Wiki Points

1

Followers

Reviews: 0

User Lists: 0

#3 btk2k2
Member since 2003 • 440 Posts
[QUOTE="btk2k2"]As Ron says it is better to build a PC around the release of the PS4 to get a more accurate comparison. Life time costs will be lower for the PC if you buy enough games because they are cheaper. To have performance in line or greater than the PS4 for the lifetime of the console you will be looking at an i5 CPU with a Radeon 7950/Geforce 760 and 16GB of ram. I could build a system for around the £650 mark which is £300 more than the PS4 up front. Over time I will get that back and then some with cheaper games but it is a decent amount more to spend. If I were to use Linux instead of Windows that would come down to around £580 but my choice of games would be more limited.True_Gamer_
So youre saying that you will have to put 4 times stronger CPU and GPU than the ones in the ps4 to reach same results?

The CPU is because of the number of threads. A quad core will be able to handle 8 threads designed for a Jaguar CPU but I am not sure that a dual core i3, even with HyperThreading would be sufficient, you might get away with an AMD 8 core CPU instead but at the moment the Intel stuff is much better. GTA4 runs badly on anything less than a tri core for this reason. A 7950 is around 2x the performance of the PS4 GPU but when creating a PC game you do not write as close to the metal as you can on PS4 so you do need a bit more brute force to handle it so I am playing it safe by over specifying. The 7870 Ghz might be fast enough but right now I do not know how much they will be able to squeeze out of the PS4. I think that realistically the eye candy will be higher with this PC than the PS4 throughout the lifetime of the PS4.
Avatar image for btk2k2
btk2k2

440

Forum Posts

0

Wiki Points

1

Followers

Reviews: 0

User Lists: 0

#4 btk2k2
Member since 2003 • 440 Posts

This thread got destroyed MS already clainmed this sh** with the xbox 360 which actually had a stronger GPU,i can't imagine how sad it will be now that they have one weaker..:lol:

tormentos
The PS3 vs 360 was a case of one having some advantages and the other having some advantages with the PS3 being a royal PITA to extract the maximum performance. This time there is nothing in the Xbox1 specification that might make up for the GPU horsepower deficit. The CPU is the same and runs as the same clock speed. The ESRAM is the best solution they had to the low bandwidth DDR3 but it has a programming overhead. The move engines are to make sure data is in the ESRAM when required to reduce the amount of time the GPU has to access the DDR3 directly. It is built around compromises which were the result of wanting 8GB of ram guaranteed.
Avatar image for btk2k2
btk2k2

440

Forum Posts

0

Wiki Points

1

Followers

Reviews: 0

User Lists: 0

#5 btk2k2
Member since 2003 • 440 Posts

[QUOTE="ronvalencia"]

[QUOTE="Mr720fan"]

ron just beating down cows at every turn, love the mastery, and ron did you see the new digital foundry article??? looks like 192 gb/s possible for x1 hahahaha oh what a night!!!

Mr720fan

The said article's link is http://www.eurogamer.net/articles/digitalfoundry-xbox-one-memory-better-in-production-hardware

 

 

thank you mastery owner.    Love how the cows just are so delusional, your factual evidence is delicious.     Thanks again RON!!!   look at dem cows :lol:

It is a good article. I like the DF articles and it is really interesting how the ESRAM can be utilised. The thing is that is not the most important performance differentiator because the bandwidth in the X1 was, if handled correctly, enough to feed the GPU. The important differentiator is the shader and fillrate performance and that will not be made up in any way other than through a clock speed increase which is very unlikely to happen.
Avatar image for btk2k2
btk2k2

440

Forum Posts

0

Wiki Points

1

Followers

Reviews: 0

User Lists: 0

#6 btk2k2
Member since 2003 • 440 Posts
The debate is about the effective gap between the machines. Having the same performance would negate having two performance points.

PS4 doesn't have 7970 level GPU to show the "massive" difference with the X1

ronvalencia
The difference between a 12CU part and an 18CU part is sufficient for there to be a difference in image quality and performance. There does not need to be a leap all the way upto a 7970 to show a difference. The W5000 card you like to use as similar to the X1 GPU is still slower than the 7850 which is fractionally worse than the PS4 GPU. The PS4 will have a 40-50% performance advantage for devs to play with. In PC terms a 40-50% increase is generational. The difference on release between the 6970 and the 7970 was less than 50%. The GTX 680 was less than 50% faster than the GTX580. A 40-50% performance difference is large.
Avatar image for btk2k2
btk2k2

440

Forum Posts

0

Wiki Points

1

Followers

Reviews: 0

User Lists: 0

#7 btk2k2
Member since 2003 • 440 Posts
One big difference between W5000 and X1 is with ROPS, but a modern 16 ROPs can still handle 1080p. 32 ROPs comes into play when you go higher than 1080p e.g. Eyefinity resolutions like 5760x1080p, but before that happens, you run into other bottlenecks (for lesser GCNs with 32 ROPS e.g. W5000, 7850, 7870, W7000).

-----

W5000's 8 CUs and related TMUs are disabled. If we keep the current GCN's stream processor vs TMU ratio, X1's TMU count should be similar to X1.

------

Like X1, W5000 still has 256bit memory controllers and it's related L2 caches. 7770 and 7790 has 128bit memory controllers and it's related L2 caches.

L2 cache attempts hide the latency difference with external memory. X1's L2 cache would have to work harder than PS4's version, but X1 has SRAM insurance.

Cache%20Hierarchy.png

Note the relationship between L2 cache and memory controller.

-------

ike X1, W5000/7790 still has "2 primitives per cycle" geometry engine improvements. 7770 has "1 primitive per cycle" geometry engine.

This leaves us the memory bandwidth question.

7770 has 72 GB/s

7790 has 96 GB/s

X1 has 68 GB/s + 102 GB/s SRAM (very low latency and overheads) + JIT hardware LZ/JPEG compression/decompression (pick a known compression ratio for LZ or JPEG).

ronvalencia
You have written a lot but actually said nothing. The X1 GPU has enough bandwidth to feed it if properly programmed for. If the devs make mistakes or are lazy then the performance will be affected. The PS4 also has enough bandwidth the feed the GPU but this does not require specific programming to achieve because of the nature of the memory subsystem. The PS4 also has more execution resources meaning it can do more. It is like having a brick laying competition. You have a conveyor belt with bricks on it and you have builders laying the bricks. As long as new bricks arrive fast enough to keep all the builders working then you are good. In this ananolgy the Xbox has 12 builders with a conveyor belt fast enough to feed the builders and the PS4 has 18 builders with a conveyor belt fast enough to feed the builders. If they have to build the same size wall which group will finish faster?
Avatar image for btk2k2
btk2k2

440

Forum Posts

0

Wiki Points

1

Followers

Reviews: 0

User Lists: 0

#8 btk2k2
Member since 2003 • 440 Posts
The article says the ESRAM theoretical bandwidth has increased and the achieved bandwidth so far is upto 133GB/s. That is pretty impressive but it does not say anything about DDR3 because this has nothing to do with the DDR3. I am sure this will help in certain scenarios but the biggest issue for the Xbox1 is still the shader and fillrate deficit which cannot be magically overcome.
Avatar image for btk2k2
btk2k2

440

Forum Posts

0

Wiki Points

1

Followers

Reviews: 0

User Lists: 0

#9 btk2k2
Member since 2003 • 440 Posts
[QUOTE="btk2k2"][QUOTE="ronvalencia"]I made sure I picked a game with heavy geometry (wasteful tessellation) that would gimped older AMD hardware (1 primitive per cycle). Superior ROPs is nothing when your front end gimps the rest of the GPU.ronvalencia
No console version will do that though because why would a developer purposefully waste resources? It is not an example of something that will occur in a real world scenario on the consoles. Especially after the backlash that Crytek received when it was discovered that they were purposefully gimping one vendors set of hardware to make the competition look better because they were paid to.

AMD changed their GPU designs with improve tessellation i.e. starting with 7790 and above.

What does that have to do with anything? The Xbox1 and PS4 are both using GPU's with the same tessellation engine. A dev would have to design the game to overwhelm the tessellation engine for the performance to equalise in both systems and no developer will do that because it would be a waste of resources on both consoles and they could achieve the same image quality in other ways that have less overhead.
Avatar image for btk2k2
btk2k2

440

Forum Posts

0

Wiki Points

1

Followers

Reviews: 0

User Lists: 0

#10 btk2k2
Member since 2003 • 440 Posts
Note the word "effective". The FLOPs gap between X1 and PS4 is smaller than 7850 and 7970.

The FLOPS gap between X1 and PS4 = ~0.61 TFLOPS

The FLOPS gap between 7850 and 7970 (XT1) = ~2.027 TFLOPS

The FLOPS gap between 7850 and 8970-OEM = ~2.539 TFLOPS

ronvalencia
Why are you bringing the 7970 into this? Tahiti is a completely different animal to Pictarin(78xx), Cape Verde(7770/7750)) and Bonnair(7790). Your point also makes no sense because we are making a comparison between Xbox and PS4 and we use % rather than raw numbers because that indicates the effective difference between them. The performance difference if the IQ is normalised and we assume that memory bandwidth is not starving the GPU will be in the region of 40-50%. If a game gets 60FPS on the Xbox1 the PS4 could run it at 84-90FPS. That increase in performance will mean more eye candy can be enabled without sacrificing FPS.