There is no thread dedicated solely to this topic so I thought I'd create one.
With the launch of haswell intel is now in a very strong place regarding its igps.
On mobile hd 4600 appears to be in a position where it can compete with trinity on a fairly even footing.
http://www.notebookcheck.net/Review-Intel-HD-Graphics-4600.93188.0.html
This compares the a10-4600m to the i7-4930MX. The two are relatively equal with the hd 4600 leading at lower resolutions and trinity leading at higher resolutions. Note that lower resolutions tend to matter more because many games are not playable at higher settings (intel seems to optimize their drivers for low resolutions and settings and fps tends to drop off when settings are cranked up). However, the 4930mx runs at 1350mhz, significantly faster than is expected for the i5 and i3 level hd 4600 igp. At a speed of 1150 mhz (expected for i3/i5 haswell given the speed for ivy bridge where i5 runs at 1200 mhz and i3 at 1100 mhz) and slightly less cache I expect the hd4600 to run about 10% slower than the i7-4930mx and about 12% behind the a10-4600m. Notebook acknowledges that there are testing inconsistencies in their review and looking at the Tom's review we can see that the hd4600 generally performs much better in comparison (testing the i7-4770k with its gpu at 1250 mhz) possibly because of throttling on the mobile chips (?).
http://www.tomshardware.com/reviews/core-i7-4770k-haswell-review,3521.html
http://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested/6
Distilled down the a10-5800k is, against the 4770k (both using 1600 mhz CL9 RAM) at playable settings (must be over 30 fps or the greatest fps settings for a gain, not interested in 4 vs 8 fps at 1080p because nothing is playable).
game - percent a10-5800k over 4770k--[notes]
BF3 (720p low) - 5.5%--[4770k stutters horribly in this game]
Bioshock Infinite (720p low) - 20%
Hitman Absolution (720p low) - 3%--[stuttering is fairly bad on both igps]
Skyrim (720p med) - 49%--[a10-5800k exhibits high frame variance-intel is significantly more competitive on low settings in this game, intel is barely playable]
WOW MOP (720p good quality) - MINUS 13% --[both stutter quite a bit but the a10 is quite a bit worse]
In conclusion the a10-5800k is faster in 2 games, a bit slower in one (arguably the most important given the number of WOW players and the fact that few try to play BF3 or other AAA games on an igp) and about equal in two. Resolution does not appear to affect intel, scaling between the two is very similar (only hitman).
From the anandtech review (haswell is 2400 mhz ram, trinity is 2133 mhz ram)
game - percent a10-5800k over 4770k --[notes]
Metro last light (768p med) -15%--[mobile haswell is significantly more powerful than trinity, hd4000 is about 20% faster than a10-4600m and 4770k is about 55% faster, nothing is playable at these settings however]
Bioshock infinite (768p med) - 48%--[mobile should be very close]
Sleeping Dogs (768p med) - 32% -[mobile should be neck and neck]
Tomb Raider (768p med) - 20% - [mobile haswell slightly faster]
BF3 (768p med) - 34% better - [mobile will be close]
Crysis 3 (768p low) - 32% better - [mobile close but nothing playable]
(Crysis warhead and grid 2 dont have all the numbers).
In conclusion it can be seen that as the settings are turned up amd's lead over intel tends to grow (vs toms hardware). Expect given these numbers that mobile haswell will be very close to trinity given the fact that mobile haswell will have slightly lower clocks and less cache. It can also be seen that perhaps (need to isolate variables to be sure) faster RAM helps trinity more than haswell. On average trinity a10-5800k is about 30% faster than haswell. HD 4600 is about 30% faster than HD 4000.
(In this review I am ignoring Iris/HD 5200 given the massive difference in pricing).
Note also that for this AT Iris Pro review the power figures are pretty much worthless. TDP != power consumption, especially under gaming loads. For instance the rmbp i7 quad + 650m (which isn't anywhere close to 45 watt tdp) cannot be drawing 90 watts simply because it has an 85 watt adapter for the ENTIRE NOTEBOOK. The maximum the CPU + GPU can possibly be drawing once you remove the screen, RAM, mobo, etc is around 70-75 watts.
So it looks like HD 4600 in i5/i3 mobile will be about as powerful as trinity (looking at in the AT tests the 4600 in the 4770k is about 10-15% faster than the 7660G) and about 10-15% behind the top level richland chips. On the desktop the a10-6800k is about 10% faster cpu wise (4.4 vs 4.0 ghz) and about 5% faster gpu wise (844 vs 800 mhz). Mobile will be very close but in desktop richland is still going to be about 35-40% faster than your average HD 4600 chip.
As can be seen trinity/richland is around 30% faster than HD 4600. Richland is essentially indistinguishable from trinity (2% gain).
Intel is also quite powerful in compute, very competitive vs trinity/richland. Tesselation works quite well on HD 4600.
Looking toward the future Kaveri is expected to greatly improve amd's igp performance. Expected launch dates are Q4 2013 for desktop and Q1 2014 for mobile. If Kaveri contains 512 GCN shaders then at 800 mhz it will match the desktop 7750 (if it uses GDDR5). On mobile however, it may be difficult to stick that level of power into a 35 watt chip, especially given the fact that AMD needs to improve its mobile CPU performance and 32nm to 28 nm isn't a whole lot of room. It may also be difficult given the fact that 28 nm gcn cards are drawing about 25 watts under gaming loads (7730m with 512 gcn shaders at 575-675 mhz is rated at 32 watts using DDR3--this is the mobile card so integration into an apu is going to cut off less power vs a desktop gpu).
With the launch of haswell intel is now in a very strong place regarding its igps.
On mobile hd 4600 appears to be in a position where it can compete with trinity on a fairly even footing.
http://www.notebookcheck.net/Review-Intel-HD-Graphics-4600.93188.0.html
This compares the a10-4600m to the i7-4930MX. The two are relatively equal with the hd 4600 leading at lower resolutions and trinity leading at higher resolutions. Note that lower resolutions tend to matter more because many games are not playable at higher settings (intel seems to optimize their drivers for low resolutions and settings and fps tends to drop off when settings are cranked up). However, the 4930mx runs at 1350mhz, significantly faster than is expected for the i5 and i3 level hd 4600 igp. At a speed of 1150 mhz (expected for i3/i5 haswell given the speed for ivy bridge where i5 runs at 1200 mhz and i3 at 1100 mhz) and slightly less cache I expect the hd4600 to run about 10% slower than the i7-4930mx and about 12% behind the a10-4600m. Notebook acknowledges that there are testing inconsistencies in their review and looking at the Tom's review we can see that the hd4600 generally performs much better in comparison (testing the i7-4770k with its gpu at 1250 mhz) possibly because of throttling on the mobile chips (?).
http://www.tomshardware.com/reviews/core-i7-4770k-haswell-review,3521.html
http://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested/6
Distilled down the a10-5800k is, against the 4770k (both using 1600 mhz CL9 RAM) at playable settings (must be over 30 fps or the greatest fps settings for a gain, not interested in 4 vs 8 fps at 1080p because nothing is playable).
game - percent a10-5800k over 4770k--[notes]
BF3 (720p low) - 5.5%--[4770k stutters horribly in this game]
Bioshock Infinite (720p low) - 20%
Hitman Absolution (720p low) - 3%--[stuttering is fairly bad on both igps]
Skyrim (720p med) - 49%--[a10-5800k exhibits high frame variance-intel is significantly more competitive on low settings in this game, intel is barely playable]
WOW MOP (720p good quality) - MINUS 13% --[both stutter quite a bit but the a10 is quite a bit worse]
In conclusion the a10-5800k is faster in 2 games, a bit slower in one (arguably the most important given the number of WOW players and the fact that few try to play BF3 or other AAA games on an igp) and about equal in two. Resolution does not appear to affect intel, scaling between the two is very similar (only hitman).
From the anandtech review (haswell is 2400 mhz ram, trinity is 2133 mhz ram)
game - percent a10-5800k over 4770k --[notes]
Metro last light (768p med) -15%--[mobile haswell is significantly more powerful than trinity, hd4000 is about 20% faster than a10-4600m and 4770k is about 55% faster, nothing is playable at these settings however]
Bioshock infinite (768p med) - 48%--[mobile should be very close]
Sleeping Dogs (768p med) - 32% -[mobile should be neck and neck]
Tomb Raider (768p med) - 20% - [mobile haswell slightly faster]
BF3 (768p med) - 34% better - [mobile will be close]
Crysis 3 (768p low) - 32% better - [mobile close but nothing playable]
(Crysis warhead and grid 2 dont have all the numbers).
In conclusion it can be seen that as the settings are turned up amd's lead over intel tends to grow (vs toms hardware). Expect given these numbers that mobile haswell will be very close to trinity given the fact that mobile haswell will have slightly lower clocks and less cache. It can also be seen that perhaps (need to isolate variables to be sure) faster RAM helps trinity more than haswell. On average trinity a10-5800k is about 30% faster than haswell. HD 4600 is about 30% faster than HD 4000.
(In this review I am ignoring Iris/HD 5200 given the massive difference in pricing).
Note also that for this AT Iris Pro review the power figures are pretty much worthless. TDP != power consumption, especially under gaming loads. For instance the rmbp i7 quad + 650m (which isn't anywhere close to 45 watt tdp) cannot be drawing 90 watts simply because it has an 85 watt adapter for the ENTIRE NOTEBOOK. The maximum the CPU + GPU can possibly be drawing once you remove the screen, RAM, mobo, etc is around 70-75 watts.
So it looks like HD 4600 in i5/i3 mobile will be about as powerful as trinity (looking at in the AT tests the 4600 in the 4770k is about 10-15% faster than the 7660G) and about 10-15% behind the top level richland chips. On the desktop the a10-6800k is about 10% faster cpu wise (4.4 vs 4.0 ghz) and about 5% faster gpu wise (844 vs 800 mhz). Mobile will be very close but in desktop richland is still going to be about 35-40% faster than your average HD 4600 chip.
As can be seen trinity/richland is around 30% faster than HD 4600. Richland is essentially indistinguishable from trinity (2% gain).
Intel is also quite powerful in compute, very competitive vs trinity/richland. Tesselation works quite well on HD 4600.
Looking toward the future Kaveri is expected to greatly improve amd's igp performance. Expected launch dates are Q4 2013 for desktop and Q1 2014 for mobile. If Kaveri contains 512 GCN shaders then at 800 mhz it will match the desktop 7750 (if it uses GDDR5). On mobile however, it may be difficult to stick that level of power into a 35 watt chip, especially given the fact that AMD needs to improve its mobile CPU performance and 32nm to 28 nm isn't a whole lot of room. It may also be difficult given the fact that 28 nm gcn cards are drawing about 25 watts under gaming loads (7730m with 512 gcn shaders at 575-675 mhz is rated at 32 watts using DDR3--this is the mobile card so integration into an apu is going to cut off less power vs a desktop gpu).
Facebook
Twitter